HEALTH TECHNOLOGY ASSESSMENT

HEALTH TECHNOLOGY ASSESSMENT VOLUME 17 ISSUE 54 NOVEMBER 2013 ISSN 1366-5278 Lithium or an atypica antipsychotic drug in the management of treatment-resistant depression: a systematic review and economic evauation SJ Edwards, V Hamiton, L Nherera and N Trevor DOI 10.3310/hta17540

Lithium or an atypica antipsychotic drug in the management of treatment-resistant depression: a systematic review and economic evauation SJ Edwards, * V Hamiton, L Nherera and N Trevor BMJ Technoogy Assessment Group (BMJ-TAG), London, UK *Corresponding author Decared competing interests of authors: SJE was an empoyee of AstraZeneca UK Ltd unti August 2010. AstraZeneca hods the marketing authorisation for Seroque (quetiapine), an atypica antipsychotic drug that is incuded in this report. Pubished November 2013 DOI: 10.3310/hta17540 This report shoud be referenced as foows: Edwards SJ, Hamiton V, Nherera L, Trevor N. Lithium or an atypica antipsychotic drug in the management of treatment-resistant depression: a systematic review and economic evauation. Heath Techno Assess 2013;17(54). Heath Technoogy Assessment is indexed and abstracted in Index Medicus/MEDLINE, Excerpta Medica/EMBASE, Science Citation Index Expanded (SciSearch ) and Current Contents / Cinica Medicine.

Heath Technoogy Assessment HTA/HTA TAR ISSN 1366-5278 (Print) ISSN 2046-4924 (Onine) Five-year impact factor: 5.804 Heath Technoogy Assessment is indexed in MEDLINE, CINAHL, EMBASE, The Cochrane Library and the ISI Science Citation Index and is assessed for incusion in the Database of Abstracts of Reviews of Effects. This journa is a member of and subscribes to the principes of the Committee on Pubication Ethics (COPE) (www.pubicationethics.org/). Editoria contact: nihredit@southampton.ac.uk The fu HTA archive is freey avaiabe to view onine at www.journasibrary.nihr.ac.uk/hta. Print-on-demand copies can be purchased from the report pages of the NIHR Journas Library website: www.journasibrary.nihr.ac.uk Criteria for incusion in the Heath Technoogy Assessment journa Reports are pubished in Heath Technoogy Assessment (HTA) if (1) they have resuted from work for the HTA programme, and (2) they are of a sufficienty high scientific quaity as assessed by the reviewers and editors. Reviews in Heath Technoogy Assessment are termed systematic when the account of the search appraisa and synthesis methods (to minimise biases and random errors) woud, in theory, permit the repication of the review by others. HTA programme The HTA programme, part of the Nationa Institute for Heath Research (NIHR), was set up in 1993. It produces high-quaity research information on the effectiveness, costs and broader impact of heath technoogies for those who use, manage and provide care in the NHS. Heath technoogies are broady defined as a interventions used to promote heath, prevent and treat disease, and improve rehabiitation and ong-term care. The journa is indexed in NHS Evidence via its abstracts incuded in MEDLINE and its Technoogy Assessment Reports inform Nationa Institute for Heath and Care Exceence (NICE) guidance. HTA research is aso an important source of evidence for Nationa Screening Committee (NSC) poicy decisions. For more information about the HTA programme pease visit the website: www.hta.ac.uk/ This report The research reported in this issue of the journa was funded by the HTA programme as project number 10/30/01. The contractua start date was in Juy 2011. The draft report began editoria review in June 2012 and was accepted for pubication in October 2012. The authors have been whoy responsibe for a data coection, anaysis and interpretation, and for writing up their work. The HTA editors and pubisher have tried to ensure the accuracy of the authors' report and woud ike to thank the reviewers for their constructive comments on the draft document. However, they do not accept iabiity for damages or osses arising from materia pubished in this report. This report presents independent research funded by the Nationa Institute for Heath Research (NIHR). The views and opinions expressed by authors in this pubication are those of the authors and do not necessariy refect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Heath. If there are verbatim quotations incuded in this pubication the views and opinions expressed by the interviewees are those of the interviewees and do not necessariy refect those of the authors, those of the NHS, the NIHR, NETSCC, the HA programme or the Department of Heath. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. Pubished by the NIHR Journas Library (www.journasibrary.nihr.ac.uk), produced by Prepress Projects Ltd, Perth, Scotand (www.prepress-projects.co.uk).

Editor-in-Chief of Heath Technoogy Assessment and NIHR Journas Library Professor Tom Waey Director, NIHR Evauation, Trias and Studies and Director of the HTA Programme, UK NIHR Journas Library Editors Professor Ken Stein Chair of HTA Editoria Board and Professor of Pubic Heath, University of Exeter Medica Schoo, UK Professor Andree Le May Chair of NIHR Journas Library Editoria Group (EME, HS&DR, PGfAR, PHR journas) Dr Martin Ashton-Key Consutant in Pubic Heath Medicine/Consutant Advisor, NETSCC, UK Professor Matthias Beck Chair in Pubic Sector Management and Subject Leader (Management Group), Queen s University Management Schoo, Queen s University Befast, UK Professor Aieen Carke Professor of Heath Sciences, Warwick Medica Schoo, University of Warwick, UK Dr Tessa Criy Director, Crysta Bue Consuting Ltd, UK Dr Peter Davidson Director of NETSCC, HTA, UK Ms Tara Lamont Scientific Advisor, NETSCC, UK Professor Eaine McCo Director, Newcaste Cinica Trias Unit, Institute of Heath and Society, Newcaste University, UK Professor Wiiam McGuire Professor of Chid Heath, Hu York Medica Schoo, University of York, UK Professor Geoffrey Meads Honorary Professor, Business Schoo, Winchester University and Medica Schoo, University of Warwick, UK Professor Jane Norman Professor of Materna and Feta Heath, University of Edinburgh, UK Professor John Powe Consutant Cinica Adviser, Nationa Institute for Heath and Care Exceence (NICE), UK Professor James Raftery Professor of Heath Technoogy Assessment, Wessex Institute, Facuty of Medicine, University of Southampton, UK Dr Rob Riemsma Reviews Manager, Keijnen Systematic Reviews Ltd, UK Professor Heen Roberts Professoria Research Associate, University Coege London, UK Professor Heen Snooks Professor of Heath Services Research, Institute of Life Science, Coege of Medicine, Swansea University, UK Pease visit the website for a ist of members of the NIHR Journas Library Board: www.journasibrary.nihr.ac.uk/about/editors Editoria contact: nihredit@southampton.ac.uk NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Abstract Lithium or an atypica antipsychotic drug in the management of treatment-resistant depression: a systematic review and economic evauation SJ Edwards,* V Hamiton, L Nherera and N Trevor BMJ Technoogy Assessment Group (BMJ-TAG), London, UK *Corresponding author Background: Patients with treatment-resistant depression (TRD) are those with major depressive disorder that has not responded adequatey to treatment. The causes of depression are not fuy understood, athough there is evidence to suggest that depression is a compex interaction among bioogica, genetic, psychosocia and environmenta factors. Strategies avaiabe for the treatment of patients with TRD incude pharmacoogica, non-pharmacoogica, and psychoogica and psychosocia interventions. Pharmacoogica treatment options incude switching to a different antidepressant, the addition of another antidepressant of a different cass, or use of an augmenting agent, such as anticonvusants, ithium or atypica antipsychotics (AAPs). However, there is imited evidence avaiabe on the effectiveness of these strategies in the treatment of TRD. Objectives: To estimate the cinica effectiveness and cost-effectiveness of augmentation of seective serotonin reuptake inhibitor (SSRI) antidepressant therapy with either ithium or an AAP drug in the management of peope with treatment-resistant unipoar depression, defined as faiure to respond to two or more antidepressant drugs in their current episode of depression. Data sources: Databases searched were Cochrane Centra Register of Controed Trias (CENTRAL), EMBASE, MEDLINE, PsycINFO and NHS Economic Evauation Database (NHS EED). A databases were searched from inception to August 2011. Additiona data were obtained from manufacturers. Review methods: Systematic reviews of studies evauating cinica effectiveness, economic anayses and quaity of ife (QoL) were executed. Quaity assessment according to predefined criteria was undertaken independenty by two reviewers. Pairwise meta-anayses and mixed-treatment comparisons (MTCs) using both fixed- and random-effects modes were undertaken based on intention-to-treat anayses. A probabiistic de novo mathematica mode was deveoped to synthesise the avaiabe data on costs and cinica outcomes from the UK NHS perspective over a 1-year time horizon (8 weeks of acute treatment captured by a decision tree and 10 months of maintenance treatment captured by a Markov mode). Resuts: Tweve randomised controed trias (RCTs) were identified in the review of cinica effectiveness iterature; 10 considered SSRI + AAP compared with SSRI + pacebo/no treatment, one considered SSRI + AAP compared with SSRI + ithium and one considered SSRI + ithium compared with SSRI + pacebo. The RCTs incuded in the primary anayses used fuoxetine as the background SSRI and oanzapine as the AAP. Resuts of the MTC showed a non-significant trend in favour of ithium augmentation for response [ithium a priori odds ratio (OR) 1.29; 95% credibe interva (CrI) 0.11 to 5.32; ithium post hoc OR 4.15; 95% CrI 0.25 to 20.34 (the tria informing the comparison with ithium reported response using two different definitions)], mean change in Montgomery Åsberg Depression Rating Scae score from baseine (mean difference 1.47, 95% CrI 9.10 to 6.41) and a-cause withdrawas (OR 0.74, 95% CrI 0.10 to 2.66). Four economic evauations (none directy addressing the review question) and 17 studies that reported on Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. v

ABSTRACT QoL were identified and summarised in narrative reviews. The resuts of the de novo modeing indicate that augmentation of SSRI with ithium dominates augmentation of an SSRI with AAP (i.e. it resuted in cost savings of 905 per person per year and generated more heath benefits, estimated to be 0.03 quaity-adjusted ife-years). However, sensitivity anayses showed that the mode was highy sensitive to changes in acute treatment efficacy (response and remission) or discontinuation. The mode was not sensitive to changes in other parameters. Limitations: In patients with TRD, there is a ack of direct evidence comparing the cinica effectiveness of augmenting an SSRI with an AAP compared with augmenting with ithium. RCTs were identified which faciitated comparison of adding AAP with adding ithium via a MTC. However, variations in the definitions of response impemented in the RCTs, together with differences in patient baseine characteristics across RCTs, introduce bias into the anaysis. The direction and extent of the bias is uncertain. Concusions: Augmentation of SSRIs with ithium or AAP is ikey to be beneficia in peope with TRD. Cinica evauation based on the imited evidence identified in this research indicates no statisticay significant difference between the two augmentation strategies. Cost-effectiveness anayses suggest that augmentation with ithium is ess expensive and more effective than augmentation with AAP. However, the uncertainty in the cinica estimates of discontinuation and treatment response is refected in the mode resuts. A RCT comparing the two augmentation strategies, reporting reevant outcomes, incuding QoL, is needed. Study registration: PROSPERO CRD42011001464. Funding: The Nationa Institute for Heath Research Heath Technoogy Assessment programme. vi NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Contents Gossary... List of abbreviations.... Scientific summary... Chapter 1 Background Description of heath probem Incidence and prevaence Current service provision Current service cost Description of technoogy under assessment Chapter 2 Definition of the decision probem Decision probem Overa aims and objectives of assessment Chapter 3 Assessment of cinica effectiveness Methods for reviewing effectiveness Resuts Chapter 4 Assessment of cost-effectiveness Economic iterature review Quaity-of-ife iterature review Chapter 5 De novo economic anaysis Introduction Mode overview Mode structure Effectiveness data Costs Sensitivity anaysis Resuts Heath economics discussion Chapter 6 Overa discussion Statement of principa findings Summary of economic findings Strengths and imitations and uncertainty of the assessment Chapter 7 Concusions Impications for service provision Suggested research priorities Acknowedgements References ix xi xiii 1 1 2 3 4 6 11 11 12 13 13 17 45 45 55 65 65 65 67 68 74 81 85 92 97 97 101 101 105 105 105 107 109 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. vii

CONTENTS Appendix 1 Fina protoco Appendix 2 Literature search strategies for the cinica review Appendix 3 Copy of data extraction form Appendix 4 Tabe of excuded cinica effectiveness studies with rationae Appendix 5 Quaity assessment of cinica trias Appendix 6 Summary of trias incuded for each outcome in the network meta-anaysis Appendix 7 Heath economics iterature search strategies Appendix 8 Quaity assessment of incuded studies (cost-effectiveness studies) Appendix 9 Tabe of excuded heath economics studies with rationae 119 141 147 155 159 169 171 181 187 viii NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Gossary Cost-effectiveness acceptabiity curves A graph that pots a range of possibe cost-effectiveness threshods on the horizonta axis against the probabiity (chance) that the intervention wi be cost-effective on the vertica axis. In technoogy appraisas, cost-effectiveness acceptabiity curves are a means of representing the uncertainty surrounding the cost-effectiveness estimates in reation to the decision. Cost-effectiveness mode An expicit mathematica framework, which is used to represent cinica decision probems and incorporate evidence from a variety of sources to estimate costs and heath outcomes. Discounting Costs and benefits incurred today are usuay vaued more highy than costs and benefits occurring in the future. Discounting heath benefits refects society's preference for benefits to be experienced in the present rather than the future. Discounting costs refects society's preference for costs to be experienced in the future rather than the present. Dominance An intervention is dominated if it has higher costs and worse outcomes than an aternative intervention. Incrementa cost-effectiveness ratio The ratio of the difference in the mean costs of a technoogy compared with the next best aternative to the differences in the mean outcomes. Mixed-treatment comparison An anaysis that compares two or more interventions using a combination of direct evidence (from head-to-head trias of the interventions of interest) and indirect evidence (trias that do not compare the interventions of interest directy in head-to-head trias but have a common comparator). QT interva Time interva on an eectrocardiogram that represents the interva between the start of the eectrica stimuation of the ventrices (the Q wave) and the end of the recharging of the eectrica cyce in the heart (the T wave). Quaity-adjusted ife-year An index of surviva that is adjusted to account for the patient's quaity of ife during this time. Quaity-adjusted ife-years have the advantage of incorporating changes in both quantity and quaity of ife and are used to measure benefits in cost utiity anaysis. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. ix

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 List of abbreviations 5-HT 5-hydroxytryptamine GP genera practitioner AAP ATHF atypica antipsychotic drug antidepressant treatment history form HAMD HCP Hamiton Depression Rating Scae heath-care professiona BAP b.i.d. BMI British Association of Psychopharmacoogy twice daiy body mass index HRQoL HSUV HTA heath-reated quaity of ife heath-state utiity vaue Heath Technoogy Assessment BNF British Nationa Formuary HUI Heath Utiities Index CCDAN CCDANCTR CCMD-3 CEAC CI CMHT CRHTT CrI DIC DSM (-III, -IV) Cochrane Coaboration Depression, Anxiety and Neurosis Review Group CCDAN Controed Trias Register Chinese Cassification of Menta Disorders, Version 3 cost-effectiveness acceptabiity curve confidence interva community menta heath team Crisis Resoution and Home Treatment Team credibe interva deviance information criterion Diagnostic and Statistica Manua of Menta Disorders-Third and Fourth Edition ICD-10 ICER ITT LLFDI MADRS MeSH MD MDD MTC NHS EED NICE NMS Internationa Cassification of Diseases, Tenth Revision incrementa costeffectiveness ratio intention to treat Late-Life Function and Disabiity Instrument Montgomery Åsberg Depression Rating Scae medica subject heading mean difference major depressive disorder mixed-treatment comparison NHS Econonic Evauation Database Nationa Institute for Heath and Care Exceence neuroeptic maignant syndrome ECG eectrocardiogram OD once daiy ECT eectroconvusive therapy OR odds ratio EQ-5D ER European Quaity of Life-5 Dimensions extended reease OWSA PICO one-way sensitivity anaysis popuation, intervention, comparator and outcome Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. xi

LIST OF ABBREVIATIONS p.o. per os SG standard gambe PP per protoco SIP Sickness Impact Profie PRISMA PSA Q-LES-Q QALY Preferred Reporting Items for Systematic Reviews and Meta-Anayses probabiistic sensitivity anaysis Quaity of Life Enjoyment and Satisfaction Questionnaire quaity-adjusted ife-year SNRI SSRI STAR*D TMS serotonin norepinephrine reuptake inhibitor seective serotonin reuptake inhibitor Sequenced Treatment Aternatives to Reieve Depression transcrania magnetic stimuation QoL QUOROM QWB QWB-SA RCT RR SD quaity of ife Quaity of Reporting of Meta-anayses Quaity of We-Being Scae sef-administered Quaity of We-Being Scae randomised controed tria reative risk/risk ratio standard deviation TRD TSD TTO WFSBP WHOQOL-BREF treatment-resistant depression Technica Support Document time trade-off Word Federation of Societies of Bioogica Psychiatry Word Heath Organization Quaity of Life Assessment SE standard error SF-36 Short Form questionnaire- 36 items xii NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Scientific summary Background Patients with treatment-resistant depression (TRD) are those with major depressive disorder (MDD) that has not responded adequatey to treatment. However, there is much uncertainty regarding what constitutes the definition of TRD and whether or not, for exampe, a patient with a faiure to respond to two antidepressants from the same cass coud be defined as treatment resistant. The focus of this review is patients with unipoar TRD and, for the purposes of this report, TRD has been defined as a faiure to respond to two or more antidepressants in the current episode of depression. No UK-specific data on the incidence or prevaence of TRD are avaiabe in the iterature. However, it is understood that up to two-thirds of patients diagnosed with MDD wi have a suboptima response to first-ine treatment with antidepressant drugs. The Word Federation of Societies of Bioogica Psychiatry guideines for bioogica treatment of unipoar depressive disorders state that as many as 50% of non-responders to a first antidepressant tria aso fai to respond to a second, different course of treatment. There are severa strategies avaiabe for the treatment of patients with TRD. These strategies incude pharmacoogica, non-pharmacoogica, and psychoogica and psychosocia interventions. Pharmacoogica treatment options incude switching to a different antidepressant, the addition of another antidepressant of a different cass, or use of an augmenting agent, such as anticonvusant drugs, ithium or atypica antipsychotic drugs (AAPs). This report contains a heath technoogy assessment of ithium and AAPs used as augmentation therapies in the management of patients with TRD who are aready taking seective serotonin reuptake inhibitor (SSRI) antidepressant therapy. Objectives The objective of the project was to estimate the cinica effectiveness and cost-effectiveness of augmentation of SSRI antidepressant therapy with either ithium or an AAP in the management of peope with unipoar TRD. The project was spit into four distinct pieces of work: systematic review of cinica effectiveness of interventions systematic review of cost-effectiveness of interventions systematic review of quaity-of-ife (QoL) studies in depression de novo economic mode. Methods Search methods A systematic review of the iterature was carried out to identify potentiay reevant randomised controed trias (RCTs) comparing augmentation of SSRI antidepressant therapy with either ithium or an AAP in the management of peope with unipoar TRD. Databases searched were EMBASE, MEDLINE, PsycINFO and the Cochrane Centra Register of Controed Trias (CENTRAL). In addition, the registries of the Cochrane Coaboration Depression, Anxiety and Neurosis Review Group were searched. The NHS Economic Evauation Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. xiii

SCIENTIFIC SUMMARY Database (NHS EED) was aso searched for the economic evauation studies. A searches were performed from the date of database inception to August 2011. Further data were obtained from manufacturers. Incusion decisions, quaity assessment and data extraction were undertaken independenty by two reviewers according to predefined criteria. Cinica effectiveness Standard pairwise meta-anaysis was conducted using a fixed-effects mode as the primary anaysis. Mixed-treatment comparisons (MTCs) were conducted using a fixed- and random-effects mode, with the best fitting most appropriate mode chosen for the reporting of resuts. The systematic review was registered on PROSPERO (CRD42011001464). Cost-effectiveness A de novo mathematica mode was deveoped to synthesise the avaiabe data on costs and cinica outcomes from the UK NHS perspective. The mode adopted a 1-year time horizon, consisting of 8 weeks of acute treatment (captured by a decision tree) and 10 months of maintenance treatment (captured by a Markov mode). The primary outcome of interest was the cost per quaity-adjusted ife-year (QALY) gained. The mode required data on the outcomes of remission, response and discontinuation. Acute efficacy data used in the mode were derived from the MTC carried out as part of this review. As a resut of a paucity of cinica effectiveness data, a nove samping approach was used to generate the probabiities required for the economic mode. The approach invoved samping the treatment effect [change in MADRS (Montgomery Åsberg Depression Rating Scae) score from baseine] of each augmentation strategy (from a distribution of possibe effects) and cacuating the proportion of patients (in a cohort of 1000 for each treatment arm) that woud achieve remission or response during the acute treatment phase. The mode assumed that outcomes in the maintenance phase were treatment independent. Resuts Cinica effectiveness Tweve RCTs were identified in the review of cinica effectiveness data. Ten RCTs considered SSRI + AAP compared with SSRI + pacebo/no treatment. Of the remaining two RCTs, one was a comparison of SSRI or serotonin norepinephrine reuptake inhibitor (SNRI) + AAP with SSRI or SNRI + ithium and the fina RCT compared SSRI + ithium with SSRI + pacebo. Six of the 10 SSRI + AAP trias were incuded in the primary anaysis; the remaining four RCTs were incuded in a cass-based sensitivity anaysis. Of the trias considering ithium augmentation as a comparator, ony one was incuded in the primary anaysis. A six trias considering augmentation with an AAP incuded in the primary anaysis evauated fuoxetine (SSRI) + oanzapine (AAP). Furthermore, the ithium tria incuded in the primary anaysis used fuoxetine as the background SSRI in both the comparator group and ithium augmentation group. Resuts for seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor aone (pairwise comparison) Five RCTs reported response based on the MADRS and the remaining RCT used the Hamiton Depression Rating Scae. The resuts of the meta-anaysis (fixed effects) demonstrated a statisticay significant benefit for fuoxetine + oanzapine over fuoxetine aone [odds ratio (OR) 1.48; 95% confidence interva (CI) 1.13 to 1.94] with a moderate eve of statistica heterogeneity (I 2 = 53%; p = 0.07). Five RCTs reported the outcome of remission. Fixed-effects meta-anaysis demonstrated a statisticay significant increase in remissions in patients treated with oanzapine + fuoxetine compared with fuoxetine aone (OR 1.77; 95% CI 1.27 to 2.47) with no statistica heterogeneity (I 2 =0%; p = 0.75). Data on reapse rates were not avaiabe for anaysis. xiv NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Four RCTs reported east square mean difference (MD) from baseine in MADRS score at study end point. Fixed-effects meta-anaysis resuted in a statisticay significant MD of 2.04 (95% CI 3.25 to 0.82) in favour of fuoxetine + oanzapine. However, there was a high eve of heterogeneity that was statisticay significant (I 2 = 73%; p = 0.01). Fixed-effects meta-anaysis of five trias found that oanzapine augmentation therapy was associated with a non-statisticay significant increase in discontinuations (OR 1.25; 95% CI 0.91 to 1.71) with no statistica heterogeneity (I 2 =0%; p = 0.51). Seective serotonin reuptake inhibitor pus ithium compared with seective serotonin reuptake inhibitor pus pacebo response (pairwise comparison) The singe tria comparing fuoxetine + ithium with fuoxetine aone used two definitions of response, one prespecified primary anaysis and one post hoc anaysis. Resuts of the primary and post hoc anayses for response data indicated a non-significant trend in favour of ithium augmentation compared with SSRI aone (OR 1.48; 95% CI 0.37 to 5.95 and OR 3.85; 95% CI 0.80 to 18.62, respectivey). Data on remission or reapse rates were not avaiabe. The MD in change in MADRS score from baseine between fuoxetine + ithium compared with fuoxetine aone was 3.79 (95% CI 11.25 to 3.67) a non-significant improvement from baseine score with fuoxetine + ithium compared with fuoxetine aone. Data on a-cause withdrawas demonstrated fewer withdrawas with a ithium augmentation strategy than with fuoxetine aone, athough this difference was statisticay non-significant (OR 0.68; 95% CI 0.15 to 3.16). Mixed-treatment comparison (seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor pus ithium) Seven RCTs were incuded in the MTC: six for SSRI + AAP compared with SSRI aone and one RCT for SSRI + ithium compared with SSRI aone. Two separate anayses for the outcome of response were conducted because the tria informing the comparison with ithium reported response using two criteria. Anayses of response (random-effects mode) using the ithium primary anaysis and the ithium post hoc anaysis data showed a non-significant trend in favour of treatment with ithium [OR 1.29; 95% credibe interva (CrI) 0.11 to 5.32 and OR 4.15; 95% CrI 0.25 to 20.34, respectivey]. Five trias were incuded in the anaysis for mean change in MADRS (four RCTs were AAPs and one was a ithium RCT). The random-effects mode resuted in a weighted MD of 1.47 (95% CrI 9.10 to 6.41) for the mean change in MADRS score from baseine for fuoxetine + ithium compared with fuoxetine + oanzapine, which suggests a statisticay non-significant trend in favour of ithium augmentation. However, the wide 95% CrI indicates a high eve of uncertainty in this estimate of treatment effect and so the resuts shoud be interpreted with caution. Six trias reported data on a-cause withdrawas. The fixed-effects mode resuts suggested a statisticay non-significant trend in favour of augmentation with ithium (OR 0.74; 95% CrI 0.10 to 2.66) compared with augmenting with AAP. Various sensitivity anayses were carried out, incuding anayses assuming cass effects of SSRIs and AAPs, anaysis of RCTs in which patients had experienced two or more faiures to antidepressants in their current episode, and anaysis of RCTs reporting response based on MADRS score. Resuts of most sensitivity anaysis were consistent with the resuts of the primary anaysis. However, the resut of the sensitivity anaysis assuming a cass effect for SSRIs and AAPs for the outcome of mean change in MADRS differed from the primary anaysis, identifying a statisticay non-significant trend in favour of treatment with SSRI + AAP [MD 1.27; 95% CrI 1.88 to 4.68 (random-effects mode)]. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. xv

SCIENTIFIC SUMMARY Cost-effectiveness The systematic iterature review identified four economic evauations in the management of TRD and five studies that reported utiity vaues for different eves of depression severity and treatment response. Of the economic evauations, none directy addressed the review question but a were used to inform the modeing methods. Of the heath-state utiity vaues studies, one was used in the QALY cacuations of the de novo mode. The monthy cost of the commony prescribed SSRIs and AAPs varied substantiay, whereas the costs of monitoring were modest. The annua cost per patient treated with SSRI + ithium was estimated to be 4739 compared with 5644 for those treated with SSRI + AAP. The difference in cost between the two augmentation strategies is around 905 per person per year, in favour of augmentation with ithium (ithium augmentation is cheaper) and transates to savings of 75 per person per month. The resuts of the de novo modeing indicate that augmentation of SSRI therapy with an AAP is dominated by augmentation of an SSRI with ithium. The difference in costs is modest (cost savings 905 per person per year) and the difference in QALYs is estimated to be 0.03 QALYs. It appears there is no uncertainty about the dominance resut, as ithium augmentation provided more benefits than AAP augmentation in a probabiistic runs. One-way sensitivity anaysis showed that changes in costs had a minima impact on the overa resuts, whereas changes in acute efficacy or discontinuation coud potentiay reverse the direction of the cost-effectiveness resuts. For exampe, assuming a ow eve of acute response (i.e. using the upper CrI) for the treatment effect of ithium (vs. SSRI aone; MD 12.58, 95% CrI 33.0 to 7.84) resuted in AAP augmentation having incrementa cost-effectiveness ratios of < 20,000/QALY compared with ithium augmentation, as opposed to the base case in which ithium dominates. A simiar resut was observed when a high eve of acute response for AAP was assumed (i.e. using the ower CrI) for AAP treatment effect (vs. SSRI aone; MD 11.22, 95% CrI 30.13 to 7.69). However, ithium remained dominant in most of the sensitivity anayses performed. Discussion The major weakness of this anaysis is the ack of head-to-head data on the effectiveness of the comparison of SSRI + AAP with SSRI + ithium in patients with TRD. The MTC resuts demonstrate a non-significant benefit of augmenting with ithium compared with augmenting with AAP for most of the outcomes assessed. These resuts shoud be interpreted with caution, as the definitions of response and characteristics of patients incuded in the ithium RCTs differed from those used in the AAP RCTs. The incusion of sighty different popuations and definition of response may bias the resuts in favour of ithium; however, it shoud be noted that extent of the bias remains unknown. The economic anaysis found that augmenting with ithium was cheaper and more effective than augmenting with AAP. However, these resuts shoud be interpreted with caution in ight of the sensitivity of the mode to changes in the efficacy parameters and the uncertainty around the cinica data. In addition, it is expected that the price of AAPs wi fa once generic versions of branded treatments are avaiabe. It is aso important to note that the trias incuded in the MTC did not report on a of the outcomes of interest for the economic mode and consequenty some of the required parameters were generated using samping methods that have not been previousy vaidated. There was aso a paucity of RCT data on foow-up and maintenance treatment. Concusions The resuts of this review support the concusion that augmentation of SSRIs with ithium or AAP is ikey to be beneficia in peope with TRD, defined as a faiure to respond to two or more antidepressants in the current episode of depression. However, based on the imited number of RCTs identified, the cinica xvi NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 evauation suggests there is no statisticay significant difference between the two augmentation strategies. There is a genera paucity of tria data avaiabe in patients with TRD for SSRI + ithium and SSRI + AAP. The cost-effectiveness resuts suggest that augmentation with ithium is cheaper and more effective compared with augmenting with AAP. However, the resuts are not definitive because the mode is sensitive to the cinica effectiveness parameters of discontinuations and treatment response. The cost-effectiveness of SSRI + ithium and SSRI + AAP wi need to be reconsidered if further tria data become avaiabe. Suggested research priorities A RCT in patients with TRD that compares SSRI + ithium with SSRI + AAP for response, remission and discontinuation in both the acute and maintenance phases of treatment is needed. In addition, data on reapse rates in the ong term woud be beneficia. Adverse events and QoL data shoud aso be prioritised as part of the research. Study registration This study is registered as PROSPERO CRD42011001464. Funding Funding for this study was provided by the Heath Technoogy Assessment programme of the Nationa Institute for Heath Research. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. xvii

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 1 Background Description of heath probem The focus of this review is the acute-phase treatment of patients with unipoar treatment-resistant depression (TRD). Patients with TRD are those with major depressive disorder (MDD) that has not responded adequatey to treatment. However, there is much uncertainty regarding what constitutes the definition of TRD and whether or not, for exampe, a patient with a faiure to respond to two antidepressants from the same cass coud be defined as treatment resistant. 1,2 For the purposes of this report, TRD has been defined as a faiure to respond to two or more antidepressants in the current episode of depression, the definition used in the 2003 Nationa Institute for Heath and Care Exceence (NICE) cinica guideine on the treatment and management of depression in aduts (CG23). 3 This definition for TRD was aso reported by the authors of a arge systematic review of 42 randomised trias to refect the consensus within the iterature (26 trias) 4 and in the Word Federation of Societies of Bioogica Psychiatry (WFSBP) guideines 5 for bioogica treatment of unipoar depressive disorders. In addition, the WFSBP guideines 5 state that as many as 50% of non-responders to a first antidepressant tria aso fai to respond to a second, different course of treatment. However, there is a genera ack of carity or consensus around the ength of treatment required prior to treatment being defined as a faiure and aso the impact of historica treatment faiures on the definition of future episodes of TRD, i.e. whether or not TRD shoud be diagnosed based on antidepressant faiures that have occurred in ony the current episode of depression. Pathophysioogy The aetioogy of depression is not fuy understood, athough there is evidence to suggest depression is a compex interaction among bioogica, genetic, psychosocia and environmenta factors. 2 The highest rates of depression typicay occur in peope between 25 and 44 years od, and femaes are twice as ikey as maes to experience depression, 2 athough how these figures reate to the subgroup of patients with TRD is difficut to know owing to the ack of epidemioogica data and the ack of a consistent definition for TRD. Famiy history of depression is aso a risk factor for depression 2 and a previous history of MDD increases the risk of future episodes (i.e. reapses). 6 In addition, it has been reported that patients with depression have increased morbidity and mortaity. 2 For exampe, they are more ikey to die from cardiovascuar disease 7 or suicide. 8 Diagnosis and assessment of response to treatment Peope presenting with depression may compain of depressed mood, oss of interest or peasure, feeings of guit or ow sef-worth, suicida ideation, disturbed seep or appetite, ow energy and poor concentration. Depression can be diagnosed cinicay using different criteria. The most commony used criteria are the DSM-IV (Diagnostic and Statistica Manua of Menta Disorders-Fourth Edition pubished by the American Psychiatric Association) 9 and the ICD-10 (Internationa Cassification of Diseases, Tenth Revision) criteria deveoped by the Word Heath Organization (WHO). 10 The DSM-IV system requires at east five out of nine symptoms for a diagnosis of major depression, incuding at east one of the foowing two symptoms: ow mood; or oss of interest and peasure. 9 A diagnosis of moderatey severe depressive episode using the ICD-10 cassification system requires the presence of at east three out of ten depressive symptoms, incuding at east two of the foowing three symptoms: ow mood; oss of interest and peasure; or oss of energy. 10 In both cases, symptoms shoud be present for at east 2 weeks and each symptom shoud be present at sufficient severity for most of every day (Box 1 shows the fu diagnostic criteria). The recommended treatment goa in depression is to reach remission, which is defined as the reative absence of cinica symptomatoogy and is usuay determined by reaching a certain score on a treatment response rating scae. 11,12 Response to treatment in TRD is commony measured by a reduction of at east 50% on either the Hamiton Depression Rating Scae (HAMD) 13 or Montgomery Åsberg Depression Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 1

BACKGROUND BOX 1 Diagnostic and Statistica Manua of Menta Disorders-Fourth Edition 9 and ICD-10 10 diagnostic criteria The DSM-IV diagnostic criteria for major depressive disorder Depressed mood most of the day, neary every day, as indicated either by subjective report (e.g. fees sad or empty) or observation made by others (e.g. appears tearfu). Markedy diminished interest or peasure in a, or amost a, activities most of the day, neary every day (as indicated either by subjective account or observation made by others). Significant weight oss when not dieting or weight gain (e.g. a change of > 5% of body weight in a month), or decrease or increase in appetite neary every day. Insomnia or hypersomnia neary every day. Psychomotor agitation or retardation neary every day (observabe by others, not merey subjective feeings of restessness or being sowed down). Fatigue or oss of energy neary every day. Feeings of worthessness or excessive or inappropriate guit (which may be deusiona) neary every day (not merey sef-reproach or guit about being sick). Diminished abiity to think or concentrate, or indecisiveness, neary every day (either by subjective account or as observed by others). Recurrent thoughts of death (not just fear of dying), recurrent suicida ideation without a specific pan, or a suicide attempt or specific pan for committing suicide. The ICD-10 diagnostic criteria for moderate depressive episode persistent sadness or ow mood; and/or oss of interests or peasure fatigue or ow energy disturbed seep poor concentration or indecisiveness ow sef-confidence poor appetite suicida thoughts or acts beak and pessimistic views of the future guit or sef-bame. Rating Scae (MADRS). 14 Both scaes are designed to be administered via a cinica interview and consist of a ist of symptoms of depression which the cinician must assess the patient for during the interview. The cinician then rates the patient's symptoms on the scaes provided for each symptom and adds up the individua scores to provide the overa score. In both cases, the ower the score, the ess severey depressed the patient. The definition of remission on the HAMD is typicay defined as a score of 7 on the 17-item version of the HAMD. 15 However, there is currenty no consensus on a definition of remission for the MADRS, 15 athough cinica expert advisors reported that scores of 10 on the MADRS are commony used. Incidence and prevaence The current NICE cinica guideine on the treatment and management of depression in aduts (CG90) 16 states that there are 130 peope with depression per 1000 of the NHS popuation, athough ony 80 peope per 1000 of the popuation actuay consut their genera practitioner (GP). A survey carried out by the Socia Survey Division of the Office for Nationa Statistics on behaf of the Department of Heath, the Scottish Executive and the Nationa Assemby for Waes in 2000 17 reported a prevaence of the diagnosis of depressive 2 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 disorder of 28 per 1000 of the survey popuation. A comparison in the report 17 was made to an earier survey conducted in 1993, in which the prevaence of depressive disorder was reported as 23 per 1000 peope, thus suggesting that either the prevaence of depression in the UK is rising or that depression is being diagnosed more frequenty. A study ooking at depression in Engand in the adut popuation in 2000 18 reported the tota number of peope suffering from depression in Engand to be 2,661,468, based on cacuations using 1998 data from the Office for Nationa Statistics. The study aso reported that 72% of peope with depression were femae, and 20% were in the 35- to 44-year age band. 18 In CG90 16 it is reported that the prevaence of depression aso varies consideraby according to sex and a wide range of socia and economic factors (e.g. it is higher in femaes and in unempoyed peope). However, it shoud be noted that the statistics reported above refect the tota cinica spectrum of depression and, as such, encompass patients with mid, moderate and severe depression, and thus those with TRD represent ony a subgroup of these. No data specific to the UK incidence or prevaence of TRD were identified, which is possiby because there is no widey agreed definition for TRD. 1 Current service provision There are severa pharmacoogica treatment strategy options for patients with TRD not achieving adequate response with antidepressants. In the British Nationa Formuary (BNF) 19 it is stated that: Faiure to respond to initia treatment with an SSRI [seective serotonin reuptake inhibitor] may require an increase in the dose, or switching to a different antidepressant. Faiure to respond to a second antidepressant may require the addition of another antidepressant of a different cass, or use of an augmenting agent [such as ithium, aripiprazoe (unicensed), oanzapine (unicensed), quetiapine or risperidone (unicensed)], but such adjunctive treatment shoud be initiated ony by doctors with specia experience of these combinations. Other potentia treatment options suggested in CG90 16 incude augmenting with a different agent, such as anticonvusants, pindoo (Visken, Amdipharm), triiodothyronine, benzodiazepines, buspirone or atomoxetine (Strattera, Ei Liy), or the use of eectroconvusive therapy (ECT) or psychoogica and psychosocia interventions such as cognitive behavioura therapy. Reevant nationa guideines The key cinica guideine for depression in aduts in the UK is the NICE cinica guideine on the treatment and management of depression in aduts (CG90; extracts from this guideine have been reproduced here with permission); 16 this guideine is the updated version of CG23. 3 It shoud be noted that in CG90 16 a decision was taken to no onger use the term treatment-resistant depression, as there were concerns that the term impies there is a natura cut-off between peope who respond to one or two antidepressants compared with those who do not, and that this is not supported by the evidence, and the term may be taken by both doctors and patients as a pejorative abe. 16 As a resut, in CG90 16 it was decided to approach the probem of inadequate response by considering sequenced treatment options rather than by a category of patient. This is refected throughout CG90 16 by use of the abe inadequate response to initia interventions. The recommendations in CG90 16 for the sequencing of drug treatments in patients with an initia inadequate response are presented in Box 2, together with the additiona recommendations for monitoring when treatment with ithium or atypica antipsychotic drugs (AAPs) is chosen. In addition to the NICE cinica guideine (CG90) for the treatment and management of depression in aduts, 16 there are guideines pubished by the British Association of Psychopharmacoogy (BAP). 20 The BAP 2008 guideines 20 provide simiar advice to CG90 16 on the use of augmentation therapy in TRD, athough they do not specificay mention a definition of how many treatment faiures are required for a diagnosis of TRD (Box 3). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 3

BACKGROUND BOX 2 The NICE guideine, CG90. 16 Nationa Institute for Heath and Care Exceence (NICE). Depression: the treatment and management of depression in aduts (update). CG90. London: NICE; 2009. URL: http://guidance.nice. org.uk/cg90. Reproduced with permission 1.8.1.1 When reviewing drug treatment for a person with depression whose symptoms have not adequatey responded to initia pharmacoogica interventions: check adherence to, and side effects from, initia treatment increase the frequency of appointments using outcome monitoring with a vaidated outcome measure be aware that using a singe antidepressant rather than combination medication or augmentation is usuay associated with a ower side effect burden consider introducing previous treatments that have been inadequatey deivered or adhered to, incuding increasing the dose consider switching to an aternative antidepressant. Combining and augmenting medications Augmentation is when an antidepressant is used with a non-antidepressant drug and combination is when two antidepressants are used together. 1.8.1.6 If a person with depression is informed about, and prepared to toerate, the increased side effect burden, consider combining or augmenting an antidepressant with: ithium or an antipsychotic such as aripiprazoe, a oanzapine, a quetiapine a or risperidone, a or another antidepressant such as mirtazapine or mianserin. 1.8.1.7 When prescribing ithium: monitor rena and thyroid function before treatment and every 6 months during treatment (more often if there is evidence of rena impairment) consider ECG [eectrocardiogram] monitoring in peope with depression who are at high risk of cardiovascuar disease monitor serum ithium eves 1 week after initiation and each dose change unti stabe, and every 3 months thereafter. a In this guideine, drug names are marked with an asterisk if they do not have UK marketing authorisation for the indication in question at the time of pubication (October 2009). A key probem highighted in both of the guideines 16,20 is that there is imited randomised controed tria (RCT) evidence comparing the different potentia augmentation treatments and thus there is currenty much uncertainty as to which augmentation therapy is the most cinicay effective and/or cost-effective in the management of TRD. Current service cost No data were identified that reported specificay on the economic impact of TRD in the UK. However, the report of a survey carried out by the Socia Survey Division of the Office for Nationa Statistics in 2000 17 estimated the tota cost of adut depression to be over 9B, incuding around 370M of direct treatment costs. It aso estimated that there were 109.7 miion working days ost and 2615 deaths as a resut of 4 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 BOX 3 The BAP 2008 guideines 20 Augmentation/combination treatment (A) Consider adding a second agent especiay if: there is partia/insufficient response on the current antidepressant (D) and there is good toerabiity of current antidepressant (D); switching antidepressant has been unsuccessfu (D) estabish the safety of the proposed combination (S) choose the combinations with the best evidence-base first (S) consider adding ithium (A), oanzapine (A), quetiapine (B), risperidone (B), aripiprazoe (B), triiodothyronine (B) or mirtazapine (B) being aware that the evidence mainy supports ithium and triiodothyronine added to TCAs [tricycic antidepressants] and the other drugs added to SSRIs. Notes: 1. Deveoped from Shekee et a. 1999. 2. Categories of evidence for causa reationships and treatment: Evidence from meta-anaysis of randomised controed trias, a at east one arge, good quaity, randomised controed tria a or repicated, smaer, randomised controed trias* Evidence from sma, non-repicated, randomised controed trias, a at east one controed study without randomisation or evidence from at east one other type of quasi-experimenta study Evidence from non-experimenta descriptive studies, such as uncontroed, comparative, correation and case-contro studies Evidence from expert committee reports or opinions and/or cinica experience of respected authorities 3. Proposed categories of evidence for non-causa reationships: Evidence from arge representative popuation sampes Evidence from sma, we designed, but not necessariy representative sampes Evidence from non-representative surveys, case reports Evidence from expert committee reports or opinions and/or cinica experience of respected authorities 4. Strength of recommendation: A. Directy based on category I evidence B. Directy based on category II evidence or extrapoated b recommendation from category I evidence C. Directy based on category III evidence or extrapoated b recommendation from category I or II evidence D. Directy based on category IV evidence or extrapoated b recommendation from category I, II or III evidence S. Standard of good practice. a Randomised controed trias must have an appropriate contro treatment arm; for primary efficacy this shoud incude a pacebo condition. b Extrapoation may be necessary when evidence is ony indirecty reated, covers ony a part or the area of practice under consideration, or has methodoogica probems or is contradictory. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 5

BACKGROUND depression in 2000. 17 These figures represent the whoe spectrum of depression and so the actua costs of TRD are ikey to be much ower, athough it coud reasonaby be expected that the costs per patient woud be significanty higher for patients with TRD compared with the costs of treatment of mid depression. This assumption is supported by reports that patients with TRD use a disproportionatey arger share of heath-care resources and cost empoyers more in ost productivity than patients with MDD who respond to treatment. 2,21 A more recent economic review was conducted by the King's Fund in 2006 to estimate expenditure on menta heath, incuding depression, in Engand up to 2026. 22 This study 22 estimated the tota costs for depression, incuding prescribed drugs, inpatient care, other NHS services, supported accommodation, socia services and ost empoyment. The tota cost of services for depression in Engand in 2007 was estimated to be 1.7B, with costs projected to reach 3B by 2026. 22 Moreover, addition of costs attributabe to ost empoyment increased the estimated cost of depression in 2007 to 7.5B (projected to reach 12.2B by 2026). 22 These figures are consistent with a more recent report by the UK Menta Heath Foundation in November 2010; 23 this reported that depression costs the UK economy over 9 bn a year in ost earnings an increase of 4 bn since 1999, and a rise of over 500 m in the ast year aone. The figures for the UK Menta Heath Foundation report 23 were cacuated by the Research Service of the House of Commons Library. The estimates identified for the costs of depression are a consistent with a statement in CG90 16 that the indirect costs of depression far outweigh the heath service costs. Description of technoogy under assessment The technoogies under assessment in this report are ithium and AAPs used as augmentation therapies in the management of patients with TRD taking concomitant SSRI antidepressant therapy. A of the treatments under investigation in this review are avaiabe as ora tabet or iquid formuations. 19 Some of the AAPs (aripiprazoe, oanzapine and risperidone) are aso avaiabe for administration parenteray (e.g. intravenousy or intramuscuary). The current usage in the NHS of ithium and AAPs as augmentation therapies in TRD varies across different regions. This is ikey to be due to the absence of nationa guideines or treatment pathways recommending a preference for either augmentation strategy for TRD. The NICE cinica guideine on depression in aduts (CG90) 16 recommends the foowing AAPs to augment the effectiveness of SSRIs in TRD: aripiprazoe (Abiify, Bristo-Myers Squibb); oanzapine (Zyprexa, Ei Liy); quetiapine (Seroque, AstraZeneca); and risperidone (Risperda, Janssen). The augmentation therapies that are the focus of this report have been evauated in patients who have faied to respond to two or more antidepressants in their current episode of depression. Cinica advisors for this report have suggested that augmentation with ithium or an AAP in TRD may be commenced after either one or two antidepressant faiures in the current episode of depression, depending on the patient's medica history and current cinica status. However, the cinica advisors reported that, in their experience, as many as 50% of patients who fai on an initia SSRI wi respond to a second SSRI. The experts thus consider that the popuation of patients with faiure to respond to two or more previous antidepressants in their current episode of depression represents an appropriate popuation in which to evauate the cinica effectiveness and cost-effectiveness of augmentation with either ithium or AAP. The NICE cinica guideine on depression in aduts (CG90) 16 reports on the use of augmentation agents after an initia inadequate response to treatment of depression, athough this is not further defined. In these patients, CG90 16 recommends that augmentation with ithium or an AAP coud be considered as a potentia treatment strategy. In addition, it is considered by the cinica experts for this report that, at this time, ithium is used ess frequenty in the NHS than AAPs in the treatment of patients with TRD. The duration of augmentation therapy is variabe, and is party dependent on the ength of time unti remission is reached. For this review, it is anticipated that treatment with any augmentation agent shoud be for a minimum of 4 weeks prior to the fina efficacy assessment, and treatment with AAPs in particuar is 6 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 recommended for a minimum of 4 weeks before discontinuation due to ack of efficacy. 19 Usuay, it is expected that augmentation therapy woud be discontinued after a period of time in remission, athough there is currenty no set duration for this maintenance of treatment. Cinica advisors for this report suggest that treatment shoud be maintained for around 6 months post diagnosis of remission. CG90 16 aso suggests that treatment shoud be continued for at east 6 months after remission, and in patients at risk of reapse treatment shoud be continued for at east 2 years. In cinica practice, augmentation of antidepressants may occur in primary, secondary or tertiary care, and is usuay in an outpatient setting. In CG90 16 it is recommended that augmentation therapy shoud be started ony in primary care in consutation with a consutant psychiatrist. Cinica experts for this report estimate that approximatey 70% of patients with TRD wi receive care from their GP and a community menta heath team (CMHT). Of the remaining patients, it is estimated that 20% wi be seen by Crisis Resoution and Home Treatment Teams (CRHTTs), which provide intensive home-based support, and the remainder of patients woud receive inpatient care. Foow-up for patients with TRD is usuay dependent on the patient's cinica need and aso the requirement for monitoring associated with the individua augmentation therapy, which is discussed in more detai beow. Seective serotonin reuptake inhibitors The SSRIs icensed for use in the UK are citaopram, escitaopram, fuoxetine, fuvoxamine, paroxetine and sertraine. They are a avaiabe in Engand and Waes as both generic and branded drugs: citaopram (Ciprami, Lundbeck); escitaopram (Cipraex, Lundbeck); fuoxetine (Prozac, Liy); fuvoxamine (Faverin, Abbott Heathcare); paroxetine (Seroxat, GaxoSmithKine); and sertraine (Lustra,Pfizer). SSRIs work by seectivey inhibiting the reuptake of serotonin [5-hydroxytryptamine (5-HT)]; hence, they are termed SSRIs. SSRIs are commony used first-ine for treating depression as they are better toerated and are safer in overdose than other casses of antidepressants. 19 In particuar, the SSRIs are ess sedating and have fewer antimuscarinic and cardiotoxic effects than tricycic antidepressants. 19 Side effects of SSRIs incude gastrointestina effects (e.g. nausea, vomiting, abdomina pain, diarrhoea, constipation), anorexia, rash, dry mouth, anxiety, headache, insomnia, tremor, dizziness, asthenia, drowsiness, convusions, sexua dysfunction, urinary retention, sweating and hyponatraemia. 19 SSRIs are recommended for use with caution in patients with epiepsy, cardiac disease, diabetes meitus, susceptibiity to ange-cosure gaucoma, a history of mania or beeding disorders, or taking drugs that increase the risk of beeding. 19 SSRIs are associated with a risk of seizures and shoud therefore be used with caution in those receiving concurrent ECT. 19 Lithium Lithium is used in the UK to augment antidepressants in patients with TRD, athough this is an unicensed indication. 19 Lithium is more commony used for its icensed indication as a mood-stabiising agent, athough the precise mechanism of action of ithium remains unknown. Lithium is avaiabe in both generic and branded formuations in Engand and Waes. These are as foows: ithium carbonate (Camcoit, Norgine; Lithonate, Teva UK; Liskonum, GaxoSmithKine; Priade tabets, Sanofi-aventis); and ithium citrate (Li-Liquid, Rosemont; Priade iquid, Sanofi-aventis). Lithium sats have a narrow therapeutic toxic ratio and therefore serum ithium concentrations shoud be monitored reguary during treatment. 19 Leves shoud be measured 12 hours after the dose with the aim of achieving a serum-ithium concentration of 0.4 1 mmo/. 19 The BNF recommends that routine serum-ithium monitoring shoud be performed weeky after initiation and after each dose change unti concentrations are stabe, then every 3 months thereafter. 19 It shoud aso be noted that different ithium preparations have different bioavaiabiity and so caution is required when changing the ithium preparation. 19 Rena function shoud aso be monitored at baseine and every 6 months thereafter as ithium is excreted renay, and so rena impairment coud cause ithium eves to buid up, eading to toxicity. 19 Serum ithium eves are aso affected by a patient's sodium or fuid intake, with the risk of ithium toxicity increasing if Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 7

BACKGROUND there is sodium depetion or dehydration. Long-term use of ithium has been associated with thyroid disorders and mid cognitive and memory impairment. 19 Therefore, ong-term treatment requires carefu assessment of risk and benefit, and monitoring of thyroid function every 6 months. 19 Side effects of ithium therapy incude gastrointestina disturbances, fine tremor, rena impairment, poydipsia, eucocytosis, weight gain, cognitive duing, hyperparathyroidism, hyperthyroidism, hypergycaemia, hypermagnesaemia and hypercacaemia. 19 The signs of ithium toxicity incude burred vision, anorexia, vomiting, diarrhoea, musce weakness, poyuria and increasing drowsiness eventuay eading to coma. 19 It is recommended that ithium is avoided if possibe in patients with rena impairment and used with caution in patients with cardiac disease; QT-interva proongation; conditions with sodium imbaance (e.g. Addison's disease); diarrhoea; vomiting; intercurrent infection; concurrent ECT treatment; psoriasis; and myasthenia gravis. 19 In addition, caution is recommended in the use of ithium in surgica patients, patients on diuretics and the edery. 19 Atypica antipsychotic drugs As discussed above for ithium, AAPs can simiary be used as adjunctive therapies to antidepressants in the treatment of patients with TRD. AAP drugs are aso known as the second-generation antipsychotic drugs and act on a range of receptors in comparison with the first-generation antipsychotic drugs that predominanty act on ony one type of receptor. The ony AAP icensed for use in the UK as an adjunctive treatment in MDD is quetiapine. 19 However, there are severa other AAPs that are used as unicensed treatments in patients with unipoar TRD, incuding aripiprazoe, oanzapine and risperidone. In addition, there are other AAPs cassed as second-generation antipsychotic drugs that coud aso potentiay be used, abeit unicensed. Most of the AAPs are sti patented and thus are avaiabe ony as branded drugs. The AAP drugs avaiabe for use in Engand and Waes are as foows: amisupride (Soian, Sanofi-aventis); aripiprazoe (Abiify, Bristo Myers Squibb); cozapine (Cozari, Novartis; Denzapine, Merz; Zaponex, Teva UK); oanzapine (Zyprexa, Liy); paiperidone (Invega, AstraZeneca); quetiapine [Seroque, AstraZeneca; Seroque XL (modified reease), AstraZeneca]; risperidone (Risperda, Janssen-Ciag). In addition to these AAPs, ziprasidone (Geodon /Zedox,Pfizer) is used esewhere in Europe, but is not icensed or used routiney for any indication in the UK. As this report is focused on treatments avaiabe for use in the NHS, ziprasidone wi not be discussed further in this section. The choice of AAP medication is usuay infuenced by the patient's medication history, and consideration of individua patient factors, for exampe, the risk of particuar side effects such as weight gain or impaired gucose toerance. As previousy discussed, AAPs each act on different receptors. These are summarised in Tabe 1. TABLE 1 Mode of action of the AAPs 19 AAP Amisupride Aripiprazoe Cozapine Oanzapine Paiperidone Quetiapine Risperidone Mechanism of action Seective dopamine receptor antagonist with high affinity for mesoimbic D 2 and D 3 receptors Dopamine D 2 partia agonist with weak 5-HT 1a partia agonism and 5-HT 2A receptor antagonism Dopamine D 1, dopamine D 2, 5-HT 2A, apha-1 adrenoceptor, and muscarinic-receptor antagonist Dopamine D 1,D 2,D 4, 5-HT 2,H 1 - and muscarinic-receptor antagonist Metaboite of risperidone; dopamine D 2, 5-HT 2A, apha-1 adrenoceptor, and H 1 -receptor antagonist Dopamine D 1, dopamine D 2, 5-HT 2, apha-1 adrenoceptor, and H 1 -receptor antagonist Dopamine D 2, 5-HT 2A, apha-1 adrenoceptor, and H 1 -receptor antagonist 8 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Fu bood count, urea and eectroytes, and iver function test monitoring are required at the start of therapy with antipsychotic drugs, and then annuay thereafter. 19 In addition, cozapine requires differentia white bood ce monitoring weeky for 18 weeks then fortnighty for up to 1 year, and then monthy as part of the cozapine patient monitoring service. 19 Bood ipids and weight shoud aso be measured at baseine, at 3 months, and then yeary, and fasting bood gucose shoud be measured at baseine, at 4 6 months, and then yeary. 19 Patients taking cozapine or oanzapine shoud have fasting bood gucose tested at baseine, after 1 month's treatment, and then every 4 6 months. 19 It is aso advisabe to monitor proactin concentration reguary. Bood pressure monitoring is aso advised before starting therapy and frequenty during dose titration of antipsychotic drugs, and ECG monitoring may aso be required if the patient has cardiovascuar risk factors. 19 There are numerous side effects associated with AAP drugs and the side effects contribute significanty to the reasons for non-adherence to therapy. 19 Most antipsychotic drugs increase proactin concentration because dopamine inhibits proactin reease, but aripiprazoe reduces proactin because it is a dopamine-receptor partia agonist. 19 Risperidone and amisupride are most ikey to cause symptomatic hyperproactinaemia. 19 The cinica symptoms of hyperproactinaemia incude sexua dysfunction, reduced bone minera density, menstrua disturbances, breast enargement and gaactorrhoea. 19 Other side effects associated with AAPs incude cardiovascuar side effects, such as tachycardia, arrhythmias and hypotension. 19 Hypergycaemia and sometimes diabetes can occur, particuary with cozapine, oanzapine, quetiapine and risperidone. 19 A antipsychotic drugs may cause weight gain, athough the risk and extent varies, with cozapine and oanzapine being the most commony associated with weight gain. 19 Cozapine and quetiapine can cause postura hypotension (especiay during initia dose titration), which may be associated with syncope or refex tachycardia in some patients. 19 Hypersaivation is aso associated with cozapine therapy. 19 In addition, other possibe side effects incude drowsiness, agitation, restessness, increased appetite, insomnia, dizziness, headache, confusion, gastrointestina disturbances, venous thromboemboism, and antimuscarinic symptoms (e.g. dry mouth, constipation, difficuty with micturition, burred vision and aso, very rarey, precipitation of ange-cosure gaucoma). 19 Neuroeptic maignant syndrome (NMS) is a rare, but potentiay fata, side effect of a antipsychotic drugs and requires discontinuation of the antipsychotic drug. 19 NMS is characterised by hyperthermia, a fuctuating eve of consciousness, musce rigidity, paor, irreguar puse, tachycardia, sweating and urinary incontinence. It is recommended that AAPs are used with caution in patients with cardiovascuar disease, a history of epiepsy or those on concomitant drugs that increase the QT interva (on an ECG). 19 In addition, caution is required in the edery owing to an increased risk of mortaity associated with antipsychotic drugs and an increased risk of other serious side effects. 19 Anticipated costs associated with intervention The direct costs associated with the interventions under review (SSRIs, AAPs and ithium) are imited to the price of the individua tabet or iquid formuations, as there is no requirement for them to be administered in a speciaised setting. In addition, there are some costs associated with the monitoring requirements of each therapy, athough these costs vary between ithium and AAPs, as we as among the individua AAPs. The costs of the interventions aong with the wider costs associated with each intervention are discussed in detai in the cost-effectiveness section of this report (see Chapter 5, Drug costs). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 9

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 2 Definition of the decision probem This section states the key factors that wi be addressed by this report and defines the scope of the assessment (decision probem) in terms of these key factors, in ine with the definitions agreed in the pubished project protoco (see Appendix 1). 24 The protoco for this systematic review was registered on PROSPERO, which is an internationa prospective register of systematic reviews (CRD42011001464); 25 the protoco is aso avaiabe in fu on the NIHR Heath Technoogy Assessment (HTA) programme website (www.hta.ac.uk/project/2599). 24 Decision probem This report aims to address the question What is the cinica effectiveness and cost-effectiveness of ithium or an AAP in the management of treatment-resistant unipoar depression in aduts? The panned popuation, intervention, comparator and outcomes (PICO) for this report was as foows: Popuation: aduts with treatment-resistant unipoar depression defined as faiure to respond to at east two previous antidepressants in the current episode of depression. Restrictions were not imposed on the maximum number of previous antidepressant drugs aowed in order to avoid reducing the amount of data avaiabe for anaysis, as it was noted a priori that there may be imited reevant SSRI RCT data avaiabe. However, this decision assumed that there was a consistent reative treatment effect independent of ine of therapy (i.e. addition of an AAP or ithium had the same reative benefit whether given with third-ine SSRI or fourth-ine SSRI, etc.) and so a sensitivity anaysis was prespecified to assess the impact of this assumption. Intervention: an SSRI (defined as citaopram, escitaopram, fuoxetine, fuvoxamine, paroxetine or sertraine), pus an AAP drug (aso known as second-generation antipsychotic, and defined as amisupride, aripiprazoe, cozapine, oanzapine, paiperidone, quetiapine, risperidone or ziprasidone). Comparator: an SSRI (defined as citaopram, escitaopram, fuoxetine, fuvoxamine, paroxetine or sertraine), pus ithium (ithium carbonate or ithium citrate or ithium). Outcomes: response (measured by a reduction of at east 50% in HAMD 13 or MADRS 14 score) remission (using individua tria definitions) mean change from baseine MADRS 14 score quaity of ife (QoL) as reported using a vaidated QoL rating scae 25 [e.g. Short Form questionnaire-36 items (SF-36)] adverse events (tota number of events, and the individua adverse events deemed most burdensome to patients) withdrawas (a cause) as a surrogate outcome for adherence to medication reapse rate mortaity cost-effectiveness. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 11

DEFINITION OF THE DECISION PROBLEM Subgroup anayses The a priori subgroup anayses deemed to be most important were as foows: different durations of depression (i.e. time since first onset of current episode of depression) cass of previous antidepressants (e.g. SSRI or tricycic antidepressant) sex (i.e. mae and femae) age (i.e. those < 75 years and those 75 years) peope with different severities of depression (i.e. based on tria entry HAMD score 13 ). The reason for seecting these subgroups is that they were highighted by cinica experts to be the most cinicay important subgroups. This is because: Peope who have had TRD for onger periods of time are ikey to be more difficut to treat and, thus, coud be ess ikey to respond to augmentation therapy. Previous cass of antidepressant therapy may have an impact on the response to future treatments (i.e. if two SSRIs have been faied in the current episode rather than two different casses of antidepressants). It is unknown whether or not sex has an effect on response to treatment in TRD, but more femaes tend to be treated for depression than maes and thus RCTs may have a higher femae mae ratio. Peope of < 75 years of age are known to have different pathophysioogies for their depression and aso to respond differenty to antidepressants than peope aged 75 years. Peope with more severe depression at baseine (i.e. higher HAMD 13 score) require a greater improvement to enter remission and so coud potentiay be ess ikey to enter remission. It is thus considered that these subgroups of patients coud respond differenty to augmentation therapy and so each subgroup wi be anaysed for the primary outcome in this review if sufficient data are identified to enabe such comparison. Overa aims and objectives of assessment The aim of this report is to compare the augmentation of SSRI antidepressant therapy with either ithium or an AAP in the management of peope with treatment-resistant unipoar depression. The key areas that this report pans to address are: identifying and reviewing the existing evidence reating to the cinica efficacy of augmentation of SSRIs with ithium or an AAP reporting the cost-effectiveness of augmentation of SSRIs with ithium compared with that of augmentation of SSRIs with an AAP identifying what the potentia areas for future research might be in the pharmaceutica management of TRD. 12 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 3 Assessment of cinica effectiveness Methods for reviewing effectiveness A review of the evidence for cinica effectiveness was undertaken systematicay foowing the genera principes recommended in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Anayses) statement [formery the QUOROM (Quaity of Reporting of Meta-anayses) statement] 26 and based on the PICO defined in Chapter 2. Search strategy The search strategy comprised the foowing eements: 1. searching of eectronic bibiographic databases 2. contact with cinica experts in the fied 3. review of the reference ists of retrieved papers 4. searching of the Cochrane Coaboration Depression, Anxiety and Neurosis Review Group (CCDAN) Controed Trias Register (CTR) databases (CCDANCTR-Studies and CCDANCTR-References). The foowing eectronic databases were searched: (a) EMBASE (searched from 1974 to August 2011) (b) MEDLINE (searched from inception to August 2011) (c) PsycINFO (searched from inception to August 2011) (d) Cochrane Centra Register of Controed Trias (CENTRAL) (searched from inception to August 2011). Fu detais of the search strategies used are provided in Appendix 2. In addition, to assist the drawing up of fina recommendations, the website CinicaTrias.gov was searched to identify reevant ongoing cinica trias. Trias considered reevant were those that when competed may have an impact on the resuts of this review. Cinica experts in the reevant therapy areas were contacted for detais of trias (pubished and unpubished) of which they may be aware. The references from any reevant review papers or RCTs uncovered in the search were aso examined for additiona references potentiay reevant to the review. The CCDANCTR-Studies and CCDANCTR-References were searched using terms consistent with the search terms used in the other eectronic bibiographic databases as a vaidation exercise of the searches. The searches on the CCDANCTR-Studies and CCDANCTR-References registers were conducted on 7 December 2011. Abstract appraisa Tites and abstracts of studies identified by the search process were assessed independenty by two reviewers (VH and SB) for incusion. For cases in which the reviewers were unabe to reach a consensus on whether or not the fu text shoud be obtained for further appraisa, the fu text was obtained. When potentiay reevant data were avaiabe in ony an abstract format then attempts were made to contact the corresponding author and drug manufacturer to obtain the fu pubication or additiona information if possibe. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 13

ASSESSMENT OF CLINICAL EFFECTIVENESS The a priori incusion criteria appied to the review: for the review of cinica effectiveness, ony RCTs were incuded aduts 18 years peope with unipoar depression TRD defined as faiure to respond to at east two previous antidepressants in the current episode of depression SSRI given as baseine treatment and patient randomised to either ithium or an AAP comparator treatment of SSRI pus either ithium, AAP, pacebo or no treatment minimum duration of 4 weeks' treatment with study medication for the current episode of depression studies reporting on one or more of the foowing outcomes: response QoL adverse events adherence to medication or withdrawas (a cause) reapse rate mortaity cost-effectiveness. The a priori excusion criteria appied to the review: non-randomised studies narrative reviews, editorias, opinions studies performed in animas studies not focusing on the treatment of the acute phase of depression (i.e. those focusing soey on maintenance therapy) bipoar depression or bipoar disorder diagnosis prior to study entry underying medica condition or another substantia comorbid psychiatric condition (e.g. psychosis) trias reporting ony post-crossover resuts trias using non-ssri antidepressants as the baseine treatment for augmentation with ithium or an AAP. Study incusion assessment Two reviewers (VH and SB) independenty assessed the fu-text papers of the trias identified during the abstract assessment stage for incusion and any differences in opinion were arbitrated by a third reviewer (SJE). Data extraction strategy A sampe of five papers was fuy independenty data extracted by two reviewers (VH and SB) using a standardised data extraction form (for a copy of the data coection form, pease see Appendix 3) and then vaidated by one reviewer (SB). Agreement between the two reviewers was high and so, owing to time constraints, the remaining papers were independenty extracted by one reviewer (VH) and vaidated by a second reviewer (SB). Discrepancies in the data extracted by the two reviewers were resoved through discussion, with invovement of a third reviewer (SJE) if necessary. Data from intention-to-treat (ITT) anayses were extracted and it was panned that per-protoco (PP) data woud aso be extracted for use in a sensitivity anaysis, athough PP data were not reported in any of the incuded papers. For the purpose of this review, ITT was defined as patients being anaysed in the treatment group to which they were aocated at randomisation regardess of whether they received the wrong intervention, withdrew or were ost to foow-up. Shoud a tria not report ITT data then missing data were treated as treatment faiures to aow the anaysis to conform to an ITT anaysis. 14 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Study authors and drug manufacturers were contacted to suppy any additiona information not incuded in pubished sources (incuding reevant subgroup data and additiona methodoogica data required for the quaity assessment). Quaity assessment strategy Outcomes from the studies that met the incusion criteria were assessed using the updated risk of bias too deveoped by the Cochrane Coaboration (March 2011). 27 These criteria assess the foowing areas: 1. random sequence generation 2. aocation conceament 3. binding of participants and personne 4. binding of outcomes assessment 5. incompete outcome data 6. seective reporting 7. other bias. Based on these criteria, an assessment for each outcome reported in the tria was aocated based on the identified risk of bias. The three bias assessment categories used were ow risk, high risk and uncear risk. Ony trias that were deemed to be at ow or uncear risk of bias were incuded in the main anaysis, with pans to incude the trias rated as high risk in a sensitivity anaysis; no tria was rated as being at high risk of bias. Two reviewers (VH and SB) independenty rated the tria outcomes for incusion and any differences in opinion were arbitrated by a third reviewer (SJE). Outcomes reported by each RCT were considered appropriate for incusion uness the tria demonstrated a high risk of bias across severa of the seven risk-of-bias domains assessed for that outcome. No tria was excuded from any of the outcomes anaysed based on the risk of bias assessments. Methods of anaysis/synthesis Data have been tabuated and, where appropriate, meta-anaysis undertaken to estimate a summary measure of effect on reevant outcomes based on ITT anayses. Standard pairwise meta-anaysis was conducted when more than one tria was identified for incusion for any pair of treatments under investigation. This was carried out using a fixed-effects mode with the Mante Haensze method. 28 Sensitivity anayses were conducted using a random-effects mode with the DerSimonian and Laird method. 29 Ony one direct head-to-head tria was identified comparing augmentation with AAP with ithium (avaiabe as two abstracts and one poster). 30 32 The main anayses of this tria consisted of peope with resistance to either one or two antidepressants in their current episode of depression. This tria aso incuded a mixture of different SSRIs and venafaxine, a serotonin norepinephrine reuptake inhibitor (SNRI), and was excuded from the primary anayses. The decision to excude it from the primary anayses for this review was a resut of being unabe to obtain subgroup data in the subgroup of patients meeting the incusion criteria for this review (i.e. patients with resistance to two or more antidepressants and taking SSRIs, with the data reported separatey for each SSRI). As a resut, it was necessary to carry out an indirect comparison to estimate the efficacy of SSRI + AAP compared with SSRI + ithium. A mixed-treatment comparison (MTC; aso caed a mutipe-treatment meta-anaysis and network meta-anaysis) was chosen as the method to estimate the effects of SSRI + AAP compared with SSRI + ithium. A MTC can be seen as an extension of traditiona pairwise meta-anaysis. 29,33 35 The MTC was conducted using a fixed- and random-effects mode, with the most appropriate mode chosen for the reporting of the resuts. This was determined by the mode with the owest deviance information criterion (DIC). 36 DIC measures the fit of the mode whie penaising for the number of effective Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 15

ASSESSMENT OF CLINICAL EFFECTIVENESS parameters. 34,37 For the chosen mode, the consistency of the evidence was assessed using the posterior mean residua deviance, which shoud approximate the number of unconstrained data points in a good-fitting mode. For dichotomous outcomes, the odds ratio (OR) is reported as the summary statistic, and, for continuous outcomes, the mean difference (MD). The primary anaysis is: response (measured by a reduction of at east 50% in HAMD 13 or MADRS 14 score). The secondary anayses are: QoL as reported using a vaidated QoL rating scae 25 (e.g. SF-36) adverse events (tota number of events, individua adverse events for which comparabe data were avaiabe for both augmentation with AAP and with ithium, and withdrawa rates due to an adverse event) withdrawas (a cause) as a surrogate outcome for adherence to medication reapse rate mortaity (a cause). In addition, remission rates and mean change from baseine MADRS score 14 were aso chosen a priori to further assess response to treatment as these were additiona cinica parameters that were required for the economic mode. Eight-week outcome data were coected where reported. If 8-week data were not avaiabe, outcome data from the nearest avaiabe time point were coected. Subgroup anayses were panned in the foowing popuations on ony the primary outcome (response), subject to the avaiabiity of data: different durations of depression (i.e. time since first onset of current episode of depression, short term < 6 months, ong term > 6 months) cass of previous antidepressants (e.g. SSRI or tricycic antidepressant) sex (i.e. mae and femae) age (i.e. those of 75 years and those of < 75 years od) peope with different severities of depression, that is, based on tria entry HAMD 13 rating using the foowing categories: 13 8 13 = mid depression 14 18 = moderate depression 19 22 = severe depression 23 = very severe depression. In the absence of suitabe data to perform a meta-anaysis, the avaiabe data have been tabuated where possibe and discussed in a narrative review. Heterogeneity Heterogeneity in pairwise meta-anaysis has been expored through consideration of the study popuations, methods and interventions, by visuaisation of resuts and, in statistica terms, by the chi-squared test for homogeneity and the I 2 -statistic. Statisticay significant heterogeneity has been defined as p < 0.05. Leves of inconsistency have been assessed using I 2 and defined as foows. I 2 of: 0 25% = ow eve of inconsistency; 26 50% = moderate eve of inconsistency; and > 50% = high eve of inconsistency. 38 16 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 When statisticay significant heterogeneity was detected in any of the primary or secondary anayses, hypothesis-generating subgroup anaysis was conducted, athough the resuts from such anayses are highighted in the text and shoud be treated with caution. Meta-regression was panned if significant statistica heterogeneity was identified among trias anaysed and there were 10 or more trias in the review. However, there were insufficient trias in the review to consider any meta-regression for the pairwise meta-anayses. For the MTC, where a random-effects mode was deemed the best fit, the degree of heterogeneity has been investigated by evauating the posterior mean tau-squared statistic. 39 Sensitivity anaysis The foowing sensitivity anayses were specified a priori on the primary anaysis: assuming a cass effect with SSRIs and AAPs different number of prior antidepressants for the current episode of depression changing the quaity assessment to incude the tria outcomes excuded on grounds of methodoogica quaity, i.e. those categorised as being of high risk of bias changing the anaysis from using ITT data to PP data. In addition, the foowing post hoc sensitivity anaysis was conducted: imiting the primary anaysis to trias reporting response measured by a 50% reduction on the MADRS scae. 14 None of the trias incuded in this review was rated as high risk of bias and so the sensitivity anaysis incuding such trias was not required. In addition, no tria reported PP data and so this sensitivity anaysis coud not be performed. Pubication bias For each of the primary pairwise meta-anayses, a funne pot was used to assess pubication bias. A regression of normaised effect compared with precision was cacuated as a test for sma study effects (using p < 0.10 to indicate a significant resut). 40,41 Resuts Quantity of research avaiabe The search of eectronic databases identified 3717 potentiay reevant artices, which, after initia screening, resuted in the identification of 61 potentiay reevant fu-text artices that were ordered for further screening. An additiona three RCTs 42 44 were identified from the reference ist of a systematic review by Wang et a., 45 and a further RCT 46 was identified from a systematic review by Neson et a. 11 Both systematic reviews were from the 61 fu-text artices assessed. In addition, an unpubished poster 32 was provided in response to a request for further information on one RCT. 30 Foowing the fu assessment of a 66 fu-text papers, a tota of 11 studies 30,43,46 53 reported in 15 pubications were identified by both reviewers (VH and SB) as meeting the criteria for incusion: one pubication 53 incuded data from two studies, one RCT 30 was pubished as two abstracts 30,31 with additiona data avaiabe from an unpubished poster, 32 and one RCT 51 was pubished in four pubications 51,54 56 (one fu-text paper and three abstracts). From here on, each RCT wi be referred to by ony the primary source of the data incuded in this review, that is, Franco et a. 30 and Sheton et a. 51 It shoud aso be noted that two trias 49,52 incuded some patients who had faied to respond to ony one antidepressant in their current episode of depression and had a historica faiure to a second antidepressant in a previous episode of depression. Another RCT 43 may aso have incuded such patients, athough the numbers are not reported in Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 17

ASSESSMENT OF CLINICAL EFFECTIVENESS the paper. A three RCTs were incuded in the primary anayses, athough a post hoc decision was taken to perform a sensitivity anaysis to assess the impact of excuding these trias from the anaysis. In addition, it shoud be noted that one additiona study 57 that was potentiay suitabe for incusion was excuded foowing appraisa of the fu-text papers and after discussion with cinica experts because it was deemed to invove unusuay high doses of AAP and the AAP used was one that is not icensed or routiney used in the NHS. The RCT was three armed and compared two different doses of the AAP ziprasidone with pacebo augmentation of SSRI. The ziprasidone doses used were 80 mg twice daiy (b.i.d.) and 160 mg b.i.d., and the tria was tited a piot study. In addition, it is noted that there is currenty a cinica tria in progress in patients with MDD and faiure to respond to an SSRI in their current episode of depression that is assessing the efficacy of augmentation of SSRIs with ziprasidone at doses of 20 80 mg/day compared with augmentation with pacebo. This tria is isted on CinicaTrias.gov as a Phase II tria with an anticipated competion date of March 2013 (CinicaTrias.gov identifier: NCT00633399). 58 Of the 11 RCTs agreed by both reviewers (VH and SB) as suitabe for incusion, 10 RCTs were for a comparison of SSRI + AAP with SSRI + pacebo/no treatment. 43,46 53 The remaining study 30 compared SSRI or SNRI + AAP with SSRI or SNRI + ithium. Baseine antidepressants in this study were venafaxine and mixed SSRIs and thus it was agreed among the reviewers (VH, SB and SE) that it shoud not be incuded in the primary anaysis as the tria author was unabe to suppy suitabe subgroup data. No studies were identified that compared SSRI + ithium with SSRI + pacebo/no treatment. Owing to the absence of suitabe trias for the primary anaysis incuding ithium as a comparator, a pragmatic decision was taken to review a the previousy screened fu-text papers evauating ithium (n = 20), with a view to identifying trias that most cosey matched the incusion criteria. Foowing this review of previousy excuded papers, a singe study 59 that met a of the incusion criteria, with the exception of the popuation criterion, was identified. The tria reported in Katona et a. 59 considered the comparative effectiveness of SSRI + ithium with SSRI + pacebo in patients who had faied one or more antidepressant regimens. Furthermore, in ight of the new NICE guideine for depression in aduts (CG90; extracts from this guideine have been reproduced here with permission), 16 it was considered that this tria woud suffice as a proxy for a ithium tria in the required popuation; CG90 16 states that a natura cut-off between peope who respond to one or two antidepressants compared with those who do not... is not supported by the evidence. The decision to incude this tria in the review was vaidated by a third reviewer (SJE). In addition, CG90 16 and a systematic review of pacebo controed trias of ithium augmentation therapy in TRD 60 was used to vaidate that a of the other potentiay reevant SSRI + ithium compared with SSRI + pacebo trias had been identified and excuded appropriatey. However, the patient popuation of the surrogate tria for ithium augmentation coud be ess treatment resistant than the patients in the trias informing treatment augmentation with AAP. The potentia impact of this difference in the tria popuations is discussed further in the discussion section (see Assessment of effectiveness, beow). The search of CinicaTrias.gov identified no cinica trias for the comparison of SSRI + AAP with SSRI + ithium in the popuation of interest (i.e. peope with TRD and a faiure to respond to two or more antidepressants in their current episode of depression) that were competed within the past 12 months or registered as sti recruiting patients or ongoing. In addition, cinica experts for this review were not aware of any additiona pubished or unpubished reevant trias. For a fu breakdown of studies incuded and excuded at each stage of the search and appraisa process, see Figure 1 (PRISMA diagram). For detais of the fu-text studies excuded and the individua reasons for excusion, see Appendix 4. The tota number of trias agreed for incusion in this review was 12 RCTs in 16 pubications 30,43,46 53,59 (two RCTs are reported in one pubication 53 ). CG90 16 reported that the DSM-IV too was used to diagnose depression in most of the evidence reviewed in the guideine and, thus, DSM-IV was the preferred diagnostic 18 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Records identified through database searching (n = 3717) CENTRAL = 756 EMBASE = 1303 MEDLINE = 938 PsycINFO = 720 Additiona records identified from other sources (reference ists of systematic reviews) (n = 4) Records after dupicates removed (n = 1863) Records screened (n = 1863) Records excuded (n = 1797) Fu-text artices assessed for potentia surrogate trias (n = 34) Fu-text artices assessed for eigibiity (n = 66) Studies incuded meeting a priori pre-specified incusion criteria (n = 11, in 15 pubications) Fu-text artices excuded (n = 51) Unabe to obtain fu paper = 8 Did not meet popuation and/or intervention/outcome incusion criteria = 15 Systematic review = 27 Drug not icensed for use in UK/drug dose considered non-comparabe = 1 Studies incuded as proxy trias (n = 1) Tota number of studies incuded (n = 12, in 16 pubications) FIGURE 1 Preferred Reporting Items for Systematic Reviews and Meta-Anayses (PRISMA) fow diagram for studies incuded and excuded from the cinica effectiveness review. measure in the guideine. A but one of the RCTs incuded in this review required patients to have a diagnosis of MDD based on the DSM-III (one RCT) or DSM-IV (nine RCTs) criteria. The one RCT 43 identified that did not impement DSM criteria to diagnose depression used the Chinese Cassification of Menta Disorders, Version 3 (CCMD-3) measure. A sensitivity anaysis excuding this study is reported in the resuts section of the report [see Quaity assessment (sensitivity anaysis 3)]. For a summary of the characteristics of each of the studies incuded in this review, see Tabe 2. The search of the CCDANCTR-Studies and CCDANCTR-References resuted in the identification of 1487 artices. Initia screening identified three additiona potentiay reevant papers. A three of these artices represented additiona conference abstracts for a study aready incuded foowing the primary eectronic database searches (CENTRAL, EMBASE, MEDLINE and PsycINFO), and the decision was taken not to incude them in the resuts as none of them provided any additiona information to that avaiabe in the fu-text pubication. Quaity assessment A 12 of the incuded RCTs were assessed for quaity using the Cochrane risk-of-bias too. 27 In the overa assessments for each study, as we as the majority of the assessments for the individua outcomes of interest, Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 19

ASSESSMENT OF CLINICAL EFFECTIVENESS TABLE 2 Characteristics of incuded studies Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Berman et a. RCT DSM-IV diagnosis of MDD Aripiprazoe pus SSRI (2007) 47 USA (24 sites) and or venafaxine history of faiure to achieve a response to between one and three antidepressants after at east 6 weeks' treatment at adequate dose and prospective faiure to respond to an adequate dose of an investigator-assigned open-abe antidepressant (SSRI or venafaxine) pus singe bind (patient binded) adjunctive pacebo during an 8-week screening period Aripiprazoe was given at an initia dose of 5 mg/day, which was titrated weeky within a range of 2 20 mg/day (maximum 15 mg/day if on fuoxetine or paroxetine) on the basis of response and toerabiity Dose of antidepressant received prior to randomisation was given throughout the acute phase of treatment Mean dose during fina study week (mg/day) (SD): Aripiprazoe 11.8 Antidepressant doses not reported SSRI or venafaxine pus pacebo Dose of antidepressant received prior to randomisation was given throughout the acute phase of treatment Mean moda dose (mg/day) (SD): Doses not reported No. of patients in each group Aripiprazoe pus SSRI or venafaxine = 182 patients SSRI or venafaxine pus pacebo = 176 patients Duration of treatment Additiona comments 6 weeks 27.4% of patients were on venafaxine ER Distribution of the SSRI antidepressants in the aripiprazoe arm was: Escitaopram 33.9% Fuoxetine 17.5% Paroxetine controed reease 7.9% Sertraine 11.9% Distribution of the SSRI antidepressants in the pacebo arm was: Escitaopram 30.2% Fuoxetine 14.5% Paroxetine controed reease 11.6% Sertraine 17.4% 20 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Berman et a. RCT DSM-IV diagnosis of MDD Aripiprazoe pus SSRI or (2009) 48 USA (36 sites) and venafaxine history of faiure to achieve a response to between one and three antidepressants after at east 6 weeks' treatment at adequate dose and prospective faiure to respond to an adequate dose of an open-abe antidepressant (SSRI or venafaxine) pus singe bind (patient binded) adjunctive pacebo during an 8-week screening period Aripiprazoe was given at an initia dose of 5 mg/day, which was titrated weeky within a range of 2 20 mg/day (maximum 15 mg/day if on fuoxetine or paroxetine) on the basis of response and toerabiity Dose of antidepressant received prior to randomisation was given throughout the acute phase of treatment Mean dose (mg/day) (SD): Aripiprazoe 10.7 Antidepressant doses not reported SSRI or venafaxine pus pacebo Dose of antidepressant received prior to randomisation was given throughout the acute phase of treatment Mean dose (mg/day) (SD): Pacebo 13.9 Antidepressant doses not reported No. of patients in each group Duration of treatment Additiona comments Aripiprazoe pus SSRI or venafaxine = 177 patients SSRI or venafaxine pus pacebo = 6 weeks In tota, 26.2% of patients in the aripiprazoe pus SSRI or venafaxine group were on venafaxine ER, and 28.8% of patients in the SSRI or venafaxine pus pacebo group were on venafaxine ER 172 patients Foowing randomisation a sex imbaance was noted between the two tria arms, with more femaes being randomised to the aripiprazoe arm (78% vs. 68%) Distribution of the SSRI antidepressants in the tria was: Escitaopram 29.6% Fuoxetine 14.2% Paroxetine controed reease 8.9% Sertraine 19.8% continued Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 21

ASSESSMENT OF CLINICAL EFFECTIVENESS TABLE 2 Characteristics of incuded studies (continued) Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Corya et a. RCT DSM-IV diagnosis of MDD Oanzapine pus fuoxetine: (2006) 49 data pooed from four arms, 16 countries (90 sites); no further detais reported and history of faiure to achieve a satisfactory response to an SSRI after at east 6 weeks' treatment at a therapeutic dose foowed by a prospective partia response (< 30% improvement in MADRS tota score 14 ) after 7 weeks' open-abe treatment with venafaxine (ead-in phase) assessing various doses (daiy): Oanzapine 6 mg pus fuoxetine 25 mg Oanzapine 12mg pus fuoxetine 25 mg Oanzapine 6 mg pus fuoxetine 50 mg Oanzapine 12mg pus fuoxetine 50 mg Fuoxetine aone 25 mg or 50 mg aone (daiy) Mean moda doses (mg/day; average of a groups): oanzapine = 7.9; fuoxetine = 37.5 Feng et a. RCT CCMD-3 diagnosis of depression Oanzapine pus fuoxetine (2008) 43 China (one site) and patients had to have previousy taken two or more antidepressants with different mechanisms of action in sufficient quantities and for a sufficient duration (each drug treatment time 6 weeks) with no significant improvement in depressive symptoms Oanzapine 2.5 5 mg/day Fuoxetine 20 40 mg/day Fuoxetine aone 20 40 mg/day Mean daiy doses not reported No. of patients in each group Oanzapine pus fuoxetine = 243 patients Fuoxetine aone = 60 patients Oanzapine pus fuoxetine = 30 patients Fuoxetine aone = 30 patients Duration of treatment Additiona comments 12 weeks Retrospective faiure to antidepressant not imited to current episode Eight-armed RCT; specified a priori that data from groups receiving four highest doses of oanzapine and fuoxetine woud be pooed for anaysis. Other arms in the RCT were a pseudopacebo (oanzapine 1 mg/day pus fuoxetine 5 mg/day), oanzapine aone, and venafaxine RCT reports that a subgroup of patients (69.2%) had retrospective faiure in their current episode; no data on number of peope in each group reported 8 weeks Retrospective faiures to antidepressant not imited to current episode 22 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Franco et a. RCT (open abe) DSM-IV diagnosis of MDD Quetiapine ER pus SSRI or (2010) 30 venafaxine 12 countries (107 sites) and history of faiure to respond to two antidepressants, no further detais reported Quetiapine 300 mg/day Antidepressant doses not reported Lithium pus SSRI or venafaxine Lithium dose 0.6 1.0 µmo/ Antidepressant doses not reported Mean daiy doses not reported Katona et a. RCT DSM-III-revised diagnosis of MDD Fuoxetine pus ithium (1995) 59 UK; number of sites invoved not reported and patients were eigibe for augmentation with ithium if they had faied to respond to 6 weeks' treatment with an antidepressant (either fuoxetine or ofepramine) during Phase I of the tria Fuoxetine 20 mg Lithium given at an initia dose of 400 mg and increased to 800 mg for days 3 7. Lithium administered to achieve a 12-hour post-dose pasma eve of 0.6 1.0 mmo/ Faiure to respond defined as: Fuoxetine pus pacebo reduction in HAMD 13 of < 50% HAMD 13 at end of Phase I of 13 Fuoxetine 20 mg Lithium carbonate pacebo was identica in appearance, taste, and weight to the ithium carbonate tabet Mean daiy doses not reported No. of patients in each group Quetiapine pus SSRI or venafaxine = 114 patients Quetiapine pus SSRI or venafaxine = 110 patients Fuoxetine 20 mg pus ithium = 17 patients Fuoxetine 20 mg pus pacebo = 16 patients Duration of treatment Additiona comments 6 weeks Subgroup of a tria (tota tria 688 patients) that comprised peope with a history of faiure to respond to one or more antidepressants Three-armed RCT; third arm assessed the effects of quetiapine aone Of the whoe tria popuation, 36.2% of the quetiapine group was on venafaxine, and 32.6% of the ithium group was on venafaxine 6 weeks Patients entering augmentation phase were those who had faied to respond to treatment in a doube-bind controed comparison of efficacy and toerabiity of fuoxetine and ofepramine RCT assessed augmentation of fuoxetine and of ofepramine with ithium compared with pacebo (four-armed RCT) continued Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 23

ASSESSMENT OF CLINICAL EFFECTIVENESS TABLE 2 Characteristics of incuded studies (continued) Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Marcus et a. RCT DSM-IV diagnosis of MDD Aripiprazoe pus SSRI (2008) 50 USA (36 sites) and or venafaxine history of faiure to achieve a response to between one and three antidepressants after at east 6 weeks' treatment at adequate dose and prospective faiure to respond to an adequate dose of an open-abe investigatorassigned antidepressant (SSRI or venafaxine) pus singe-bind (patient-binded) pacebo during an 8-week screening period Aripiprazoe was given at an initia dose of 5 mg/day, which was titrated weeky within a range of 2 20 mg/day (maximum 15 mg/day if on fuoxetine or paroxetine) on the basis of response and toerabiity Dose of antidepressant received prior to randomisation was given throughout the acute phase of treatment Mean dose in fina week (mg/day) (SD): Aripiprazoe 11.0 Antidepressant doses not reported SSRI or venafaxine pus pacebo Dose of antidepressant received prior to randomisation was given throughout the acute phase of treatment Mean dose in fina week (mg/day) (SD): Pacebo 15.3 Antidepressant doses not reported No. of patients in each group Aripiprazoe pus SSRI or venafaxine = 191 patients SSRI or venafaxine pus pacebo = 190 patients Duration of treatment Additiona comments 6 weeks 28.1% of patients were on venafaxine Distribution of the SSRI antidepressants in the tria was: Escitaopram 30.4% Fuoxetine 14.7% Paroxetine controed reease 7.1% Sertraine 19.7% 24 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Mattingy et a. RCT DSM-IV diagnosis of MDD Quetiapine pus SSRI or SNRI (2006) 46 USA (no further detais reported) and history of at east one faiure to achieve a satisfactory response to an antidepressant after at east 4 weeks' treatment at a cinicay appropriate dose within the current episode of MDD, and a further treatment faiure defined as HAMD-17 rating of 20 after at east 6 weeks' treatment with an SSRI or SNRI, which must have been different from the earier antidepressant faied Quetiapine titrated from 50 mg to 200 400 mg/day pus ongoing treatment with an SSRI or SNRI Mean fina dose (mg/day) (SD): quetiapine = 268 (71.1); SSRI/SNRI = NR SSRI or SNRI pus pacebo Ongoing treatment with an SSRI or SNRI pus pacebo Mean fina dose (mg/day) (SD): Pacebo = 341 (53.9) SSRI/SNRI = NR Sheton et a. RCT DSM-IV diagnosis of MDD Oanzapine pus fuoxetine (2001) 51 USA (two sites) and history of faiure to achieve a response to an antidepressant other than an SSRI after at east 4 weeks' treatment and prospective faiure to respond to an adequate dose of fuoxetine (SSRI) during a 6-week screening period Oanzapine was given at an initia dose of 5 mg/day, which was titrated weeky within a range of 5 20 mg/day on the basis of response and toerabiity Dose of fuoxetine received 1 week before randomisation was given throughout the acute phase of treatment Mean moda dose (mg/day) (SD): oanzapine = 13.5 (4.1); fuoxetine = 52.0 (10.3) No. of patients in each group Quetiapine pus SSRI or SNRI = 26 patients SSRI or SNRI pus pacebo = 14 patients Oanzapine pus fuoxetine = 10 patients Fuoxetine pus pacebo = 10 patients Duration of treatment Additiona comments 8 weeks Two patients (one in each treatment arm) were on both an SSRI and an SNRI. A further nine patients (six in quetiapine group and three in the pacebo group) were on SNRIs (not further defined) 8 weeks Three-armed RCT; third arm assessed the effects of oanzapine aone continued Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 25

ASSESSMENT OF CLINICAL EFFECTIVENESS TABLE 2 Characteristics of incuded studies (continued) Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions Fuoxetine pus pacebo Dose of fuoxetine received 1 week before randomisation was given throughout the acute phase of treatment Mean moda dose (mg/day) (SD): 52.0 (14.0) Sheton et a. RCT DSM-IV diagnosis of MDD Oanzapine pus fuoxetine (2005) 52 USA and Canada (71 sites) and history of at east one faiure to achieve a satisfactory response to an SSRI after at east 4 weeks' treatment at a therapeutic dose occurring within either the current episode or a historica episode of MDD, and prospective treatment faiure to 7 weeks' treatment with nortriptyine (Aegron, King) (open-abe, dose-escaation phase); prospective faiure defined as < 30% improvement (decrease) in MADRS tota score 14 from baseine Patients coud receive either oanzapine 6 mg/day pus fuoxetine 25 mg/day or oanzapine 12 mg/day pus fuoxetine 50 mg/day Oanzapine was started at dose of 6 mg/day and coud be increased to 12 mg/day at cinician's discretion Fuoxetine was started at dose of 25 mg/day and coud be increased to 50 mg/day at cinician's discretion Mean moda dose (mg/day) (SD): oanzapine = 8.5 (3.1); fuoxetine = 35.6 (12.7) Fuoxetine pus pacebo Fuoxetine was started at dose of 25 mg/day and coud be increased to 50 mg/day at cinician's discretion Mean moda dose (mg/day) (SD): 35.8 (12.8) No. of patients in each group Oanzapine pus fuoxetine = 146 patients Fuoxetine pus pacebo = 142 patients Duration of treatment Additiona comments 8 weeks Retrospective faiure to antidepressant not imited to current episode RCT reports that a subgroup of patients (62.8%) had retrospective faiure in their current episode; no data on number of peope in each group reported Four-armed RCT; the remaining arms assessed the effects of oanzapine aone and nortriptyine 26 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Author (date) Study type and setting Criteria for randomisation to acute-phase treatment Interventions No. of patients in each group Duration of treatment Additiona comments Thase et a. (2007), 53 studies a and b RCTs DSM-IV diagnosis of MDD Oanzapine pus fuoxetine USA and Canada; no. of sites invoved not reported Two separate identica 8-week, doube-bind, parae-group RCTs reported in the same pubication and history of a faiure to achieve a satisfactory response to an antidepressant (except fuoxetine) after at east 6 weeks' treatment at a therapeutic dose within the current episode of MDD (based on the investigators' cinica judgement) and prospective faiure to 8 weeks' open-abe treatment with fuoxetine (ead-in phase); patients were excuded from the ead-in phase if they showed evidence of psychotic features or response to fuoxetine ( 25% decrease in the IVR HAMD-17 score or an IVR HAMD-17 score of < 18 or a > 15% decrease between weeks 7 and 8 of the ead-in phase) Oanzapine 6, 12 or 18 mg/day: patients were started at 6 mg/day and were required to be titrated to the higher dose uness there were issues with safety or toerabiity Fuoxetine 50 mg/day Mean moda dose (mg/day) (SD): oanzapine = 8.6 (4.7); fuoxetine = 48.8 (7.8) Fuoxetine aone 50 mg/day Mean moda dose (mg/day) (SD): 49.5 (4.9) Fuoxetine aone = 102 patients Study 1: Oanzapine pus 8 weeks Three-armed RCT; third arm assessed the effects of oanzapine aone fuoxetine = 102 patients Fuoxetine aone = 104 patients Study 2: Oanzapine pus fuoxetine = 98 patients ER, extended reease; HAMD-17, the origina version of the HAMD, which was based on 17 items; IVR, interactive voice response; NR, not reported; SD, standard deviation. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 27

ASSESSMENT OF CLINICAL EFFECTIVENESS a of the trias were rated as uncear risk of bias. This was generay the resut of a ack of information being reported in the methods, and that, despite contacting study authors, the additiona information coud not be obtained. This reason for rating studies as having uncear risk of bias is not unusua as it has been reported esewhere that uncear risk is ikey to be assigned owing to poor reporting of how a tria was conducted rather than a poory conducted tria. 61 For fu detais of the risk-of-bias assessments for each study see Appendix 5. Assessment of effectiveness The RCTs meeting the incusion criteria for the primary anayses in the cinica effectiveness review comprise trias comparing SSRI + AAP with SSRI + pacebo/no treatment, and SSRI + ithium with SSRI + pacebo. These trias were used to create a network for the MTC to address the review question regarding comparison of the cinica effectiveness of SSRI + AAP with that of SSRI + ithium. The individua cinica effectiveness resuts are presented separatey beow for each of the foowing comparisons: 1. SSRI + AAP vs. SSRI + pacebo/no treatment 2. SSRI + ithium vs. SSRI + pacebo 3. SSRI + AAP vs. SSRI + ithium. Seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor pus pacebo/no treatment A tota of six trias were identified that met the criteria for incusion in the primary anayses. 43,49,51 53 A six RCTs compared fuoxetine (SSRI) + oanzapine (AAP) with fuoxetine (SSRI) aone (or fuoxetine + pacebo). Corya et a. 49 and Sheton et a. 51,52 were reported as using an AAP pacebo tabet, and the two studies reported in Thase et a. 53 were doube bind, which suggests that an AAP pacebo tabet was used. Feng et a. 43 aso reported imited information on the conduct of the tria: there was no mention of binding and so it is unikey that a pacebo tabet was provided to the fuoxetine-aone treatment group. For simpicity, from here onwards, SSRI + pacebo/no treatment wi be referred to as SSRI aone. It shoud aso be noted that the Thase et a. studies 53 were two identica concurrent studies that were reported in a singe pubication. For the purpose of this review, many of the resuts in the ITT popuation are imited to a pooed anaysis of these two studies 53 because the appropriate data for the individua studies coud not be obtained. However, the use of pooed data in the anayses is highighted in the corresponding text. A further four trias 46 48,50 were incuded in the cass-based sensitivity anaysis because they aowed a range of antidepressants, incuding SNRIs such as venafaxine, as the baseine for augmentation and they did not present individua subgroup resuts for each SSRI. Three of these trias 47,48,50 compared an antidepressant + aripiprazoe with antidepressant + pacebo and the remaining tria 46 compared an antidepressant + quetiapine with antidepressant + pacebo. The antidepressants incuded various SSRIs and SNRIs, athough in a three of the aripiprazoe trias the SNRI was imited to venafaxine. A four trias were incuded in a sensitivity anaysis to assess the efficacy of augmentation if AAPs and SSRIs are assumed to have a cassbased effect, rather than assuming that different drugs within each cass have different efficacy. As over 70% of patients in each tria received an SSRI as baseine therapy it was agreed by the reviewers (SB, SJE and VH) to incude the four trias 46 48,50 in the sensitivity anaysis. Subgroup data for those on SSRI aone as baseine therapy were aso sought from corresponding authors. These SSRI subgroup data were provided for two of the trias 47,50 for the outcome of mean change in MADRS score 14 from baseine at study end point from a pooed anaysis. 62 Other corresponding authors either did not repy or were unabe to provide data on this subgroup. Resuts for seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor aone Response This was predefined in the protoco as a reduction of 50% in MADRS 14 or HAMD 13 score from baseine at the tria end point. It was reported in a six trias 43,49,51 53 that met the incusion criteria for this comparison, athough the data for Thase et a. 53 are reported as a pooed anaysis. Five of the trias 49,51 53 28 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 reported response based on the MADRS scae 14 and the remaining tria 43 used the HAMD scae. The resuts of the meta-anaysis (fixed effects) demonstrated a statisticay significant benefitoffuoxetine + oanzapine over fuoxetine aone [OR 1.48; 95% confidence interva (CI) 1.13 to 1.94] with a moderate eve of statistica heterogeneity (I 2 = 53%; p = 0.07) (Figure 2). Visua inspection of the funne pot did not appear to infer pubication bias (Figure 3). A regression of normaised effect compared with precision (otherwise known as Egger's regression test or more simpy the Egger's test ) was cacuated as a test for sma study effects (using p < 0.10 as an indicator of a significant resut). 40 The Egger's test was not statisticay significant (p = 0.17). A random-effects mode was used in a sensitivity anaysis to expore whether or not the resuts were sensitive to the choice of using a fixed-effects mode in the primary anaysis. The random effects meta-anaysis remained statisticay significant in favour of treatment with fuoxetine + oanzapine (OR 1.60; 95% CI 1.01 to 2.53). These resuts suggest that augmentation of fuoxetine with oanzapine significanty improves the ikeihood of treatment response in patients with TRD compared with those receiving treatment with fuoxetine (SSRI) aone. Study Corya et a. (2006) 49 Feng et a. (2008) 43 Sheton et a. (2005) 52 Sheton et a. (2001) 51 Thase et a. (2007) 53 SSRI + atypica Response Tota 100 15 6 40 80 243 30 10 146 200 SSRI aone Response Tota 19 7 1 41 60 60 30 10 142 206 OR (95% CI) 1.51 (0.83 to 2.75) 3.29 (1.08 to 9.95) 13.50 (1.20 to 152.21) 0.93 (0.56 to 1.55) 1.62 (1.07 to 2.45) OR (95% CI) Fixed effects (95% CI) Heterogeneity: c² = 8.53, df = 4 (p = 0.07); I ² = 53% Test for overa effect: z = 2.85 (p = 0.004) Random effects (95% CI) Heterogeneity: t² = 0.13; c² = 8.53, df = 4 (p = 0.07); I ² = 53% Test for overa effect: z = 1.99 (p = 0.05) 1.48 (1.13 to 1.94) 1.60 (1.01 to 2.53) 0.01 0.1 1 10 100 Favours SSRI aone Favours SSRI + atypica FIGURE 2 Resuts of meta-anaysis for 50% response comparing SSRI + AAP with SSRI aone. df, degrees of freedom. 0 0.5 SE [og(or)] 1.0 1.5 2.0 0.01 0.1 1 10 100 OR FIGURE 3 Funne pot for response. SE, standard error. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 29

ASSESSMENT OF CLINICAL EFFECTIVENESS Mean change in MADRS score This anaysis was conducted to provide cinica data for use in the economic mode (see Chapter 5, Acute treatment phase). Four trias were incuded in the meta-anaysis. 49,52,53 Using the east square MD from baseine in MADRS score at study end point resuted in a statisticay significant MD of 2.04 (95% CI 3.25 to 0.83) in favour of fuoxetine + oanzapine. This equates to a mean reduction of two points on the MADRS scae with oanzapine augmentation of fuoxetine compared with fuoxetine aone. However, there was a high eve of heterogeneity that was statisticay significant (I 2 = 73%; p = 0.01). One possibe expanation for the heterogeneity coud be reated to the study popuation of Thase et a. (study b). 53 It reported a much arger MD ( 5.9) between fuoxetine + oanzapine and fuoxetine aone compared with the other trias incuded in the meta-anaysis (vaues between 0.2 and 2.36). Simiary, Thase et a. (study b) 53 was the ony tria to report a statisticay significant difference in mean change in MADRS score (p < 0.001). Random effects meta-anaysis aso resuted in a statisticay significant MD in favour of fuoxetine + oanzapine compared with fuoxetine aone (MD 2.40; 95% CI 4.76 to 0.04) (Figure 4). An exporatory meta-anaysis was conducted to assess the impact of removing Thase et a. (study b). 53 This resuted in the remova of the significant statistica heterogeneity (I 2 =0%; p = 0.38) and the statisticay significant effect on MD 1.15 (95% CI 2.49 to 0.19), athough the trend was sti in favour of fuoxetine + oanzapine compared with fuoxetine aone. Visua inspection of the funne pot did not appear to indicate pubication bias and the Egger's test was not statisticay significant (p = 0.23). Remission In tota five RCTs 43,49,52,53 reported remission, athough the data for Thase et a. 53 are reported as a pooed anaysis. It was noted that the criteria for remission varied sighty between the RCTs. In Corya et a. 49 and Sheton et a., 52 the definition of remission used was two consecutive MADRS tota scores of 8, whereas Thase et a. 53 used a definition of a MADRS tota score of 10 at the study end point. Feng et a. 43 used the HAMD and their definition of remission was a HAMD score of < 7. Cinica experts for this review consider that the most commony used definition of remission in patients with TRD is a MADRS tota score of 10; there is some uncertainty regarding the equivaent definition for remission using the HAMD rating scae, athough the cinica experts for this report consider that a score of 7 is commony used. The resuts for a five trias 43,49,52,53 were anaysed as reported in the RCTs with no attempt made to convert them to a standard definition of remission. The resuts of the fixed-effects meta-anaysis demonstrated a statisticay significant increase in remissions in patients treated with oanzapine + fuoxetine compared with fuoxetine aone (OR 1.77; 95% CI 1.27 to 2.47), with no statistica heterogeneity (I 2 =0%; p = 0.75) (Figure 5). There was no evidence of pubication bias in the funne pot and the Egger's test was not statisticay significant (p = 0.61). Study Corya et a. (2006) 49 Sheton et a. (2005) 52 Thase et a. (2007a) 53 Thase et a. (2007b) 53 SSRI + atypica Mean change SD 14.06 9.2 8.71 8.46 11 10 14.5 10.4 SSRI aone Mean change SD 11.7 8.83 8.51 8.34 9.4 10 8.6 9.65 MD (95% CI) 2.36 ( 4.88 to 0.16) 0.20 ( 2.14 to 1.74) 1.60 ( 4.35 to 1.15) 5.90 ( 8.70 to 3.10) MD (95% CI) Fixed effects (95% CI) Heterogeneity: c² = 10.93, df = 3 (p = 0.01); I ² = 73% Test for overa effect: z = 3.30 (p = 0.0010) 2.04 ( 3.25 to 0.83) Random effects (95% CI) 2.40 ( 4.76 to 0.04) Heterogeneity: t² = 4.18; c² = 10.93, df = 3 (p = 0.01); I ² = 73% Test for overa effect: z = 1.99 (p = 0.05) 10 5 0 5 10 Favours SSRI + atypica Favours SSRI aone FIGURE 4 Resuts of meta-anaysis for mean change in MADRS score (SSRI + AAP vs. SSRI aone). df, degrees of freedom; SD, standard deviation. 30 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Study Corya et a. (2006) 49 Feng et a. (2008) 43 Sheton et a. (2005) 52 Thase et a. (2007) 53 SSRI + atypica Remission Tota 69 7 25 54 243 30 146 200 SSRI aone Remission Tota 10 3 19 34 60 30 142 206 OR (95% CI) 1.98 (0.95 to 4.13) 2.74 (0.63 to 11.82) 1.34 (0.70 to 2.55) 1.87 (1.15 to 3.03) OR (95% CI) Fixed effects (95% CI) Heterogeneity: χ² = 1.21, df = 3 (p= 0.75); I² = 0% Test for overa effect: z = 3.39 (p = 0.0007) Random effects (95% CI) Heterogeneity: t² = 0.00; c² = 1.21, df = 3 (p = 0.75); I² = 0% Test for overa effect: z = 3.35 (p = 0.0008) 1.77 (1.27 to 2.47) 1.77 (1.27 to 2.47) 0.01 0.1 1 10 100 Favours SSRI aone Favours SSRI + atypica FIGURE 5 Resuts of meta-anaysis for remission comparing SSRI + AAP vs. SSRI aone. df, degrees of freedom. Quaity of ife QoL data were ony reported in one paper as a pooed anaysis for the two Thase et a. 53 RCTs. The data were imited to changes on the Sheehan disabiity scae and the SF-36 scae, and are summarised in Tabe 3. In the QoL domains assessed, patients receiving fuoxetine + oanzapine generay showed greater improvements than with fuoxetine-aone patients, athough the difference in the SF-36 menta heath summary score was not statisticay significant (p > 0.05). Withdrawas (a cause) Data on a-cause withdrawas were reported in five trias, 49,51 53 athough the data for the two Thase et a. studies 53 were reported as a pooed anaysis. The resuts of the fixed-effects meta-anaysis suggest a statisticay non-significant reduction in discontinuations with fuoxetine aone compared with fuoxetine + oanzapine (OR 1.25; 95% CI 0.91 to 1.71), with no statistica heterogeneity (I 2 =0%; p = 0.51) (Figure 6). Visua inspection of the funne pot did not appear to indicate pubication bias and the Egger's test was not statisticay significant (p = 0.13). Withdrawas due to adverse events A tota of five studies 43,49,52,53 were suitabe for incusion in this meta-anaysis. Thase et a. 53 reported resuts ony for a pooed anaysis of the two studies. The resuts of the meta-anaysis show a statisticay significanty ower risk of withdrawa due to an adverse event with fuoxetine aone than with fuoxetine + oanzapine (OR 3.85; 95% CI 2.03 to 7.29), with no statistica heterogeneity (I 2 =0%;p = 0.40) (Figure 7). The funne pot and the Egger's test were deemed inappropriate for use for this outcome as the meta-anaysis contained fewer than four trias. Adverse events A diverse range of adverse events were reported across the trias incuded in this review. As it seemed ikey that a substantia anaytica effort woud be rewarded with itte gain, a pragmatic TABLE 3 Quaity-of-ife resuts for Thase et a. 53 pooed anaysis Quaity-of-ife measure Fuoxetine + oanzapine: n (SD) Fuoxetine aone: n (SD) p-vaue Mean improvement from baseine on Sheehan Disabiity Scae eisure item Mean improvement from baseine on Sheehan Disabiity Scae famiy item Mean change from baseine on SF-36 summary menta score Mean change from baseine on SF-36 summary physica score 1.6 (2.8) 1.1 (2.6) 0.027 1.7 (2.7) 1.2 (2.6) 0.047 8.9 (12.6) 7.3 (12.3) 0.175 2.1 (9.0) 0.4 (8.7) 0.028 SD, standard deviation. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 31

ASSESSMENT OF CLINICAL EFFECTIVENESS Study Corya et a. (2006) 49 Sheton et a. (2005) 52 Sheton et a. (2001) 51 Thase et a. (2007) 53 SSRI + atypica WDRL Tota 60 243 30 146 1 10 52 200 SSRI aone WDRL Tota 12 60 28 142 3 10 40 206 OR (95% CI) 1.31 (0.65 to 2.63) 1.05 (0.59 to 1.87) 0.26 (0.02 to 3.06) 1.46 (0.91 to 2.33) OR (95% CI) Fixed effects (95% CI) Heterogeneity: c² = 2.33, df = 3 (p = 0.51); I² = 0% Test for overa effect: z = 1.37 (p = 0.17) 1.25 (0.91 to 1.71) Random effects (95% CI) 1.25 (0.91 to 1.73) Heterogeneity: t² = 0.00; c² = 2.33, df = 3 (p = 0.51); I² = 0% Test for overa effect: z = 1.39 (p = 0.17) 0.01 0.1 1 10 100 Favours SSRI + atypica Favours SSRI aone FIGURE 6 Resuts of meta-anaysis for a-cause withdrawas comparing SSRI + AAP vs. SSRI aone. df, degrees of freedom; WDRL, number of patients withdrawing from the study. Study Corya et a. (2006) 49 Feng et a. (2008) 43 Sheton et a. (2005) 52 Sheton et a. (2001) 51 Thase et a. (2007) 53 SSRI + atypica Adverse events Tota 29 0 10 0 27 243 30 146 10 200 SSRI aone Adverse events Tota 3 0 4 0 5 60 30 142 10 206 OR (95% CI) 2.57 (0.76 to 8.76) Not estimabe 2.54 (0.78 to 8.28) Not estimabe 6.27 (2.36 to 16.64) OR (95% CI) Fixed effects (95% CI) Heterogeneity: c² = 1.85, df = 2 (p = 0.40); I ² = 0% Test for overa effect: z = 4.13 (p < 0.0001) Random effects (95% CI) Heterogeneity: t² = 0.00; c² = 1.85, df = 2 (p = 0.40); I ² = 0% Test for overa effect: z = 4.05 (p < 0.0001) 3.85 (2.03 to 7.29) 3.77 (1.98 to 7.15) 0.01 0.1 1 10 100 Favours SSRI + atypica Favours SSRI aone FIGURE 7 Resuts of meta-anaysis for withdrawas due to adverse events comparing SSRI + AAP vs. SSRI aone. df, degrees of freedom. decision was taken to anayse ony the tota number of peope reporting adverse events and the individua adverse events for which there were comparabe data for both adjuvant treatment regimens. The resuts with ithium and from the indirect comparison are presented ater in this report. Ony two RCTs 43,52 informed the comparison of fuoxetine + oanzapine with oanzapine aone. The resut of a meta-anaysis of the two trias was not statisticay significant but did suggest a trend favouring treatment with fuoxetine aone compared with fuoxetine + oanzapine (OR 1.60; 95% CI 0.91 to 2.83), with no statistica heterogeneity (I 2 =0%; p = 0.39). The individua adverse events with sufficient data for anaysis were tremor, somnoence, increased appetite, dry mouth and headache. The individua adverse events anayses generay resuted in statisticay significant resuts when the fixed-effects mode was used (in favour of fuoxetine aone, with the exception of headache, which favoured fuoxetine + oanzapine) and statisticay non-significant resuts with wide 95% CIs when a random-effects mode was used. The meta-anayses had high eves of heterogeneity that was statisticay significant. The resuts for the individua adverse events shoud thus be interpreted with caution. For fu detais of the adverse events resuts, see Figure 8. The funne pot and the Egger's test were deemed inappropriate for use for any of the adverse event outcomes as no meta-anaysis contained four or more trias. 32 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Study Any adverse event Feng et a. (2008) 43 Sheton et a. (2005) 52 SSRI + atypica Events Tota 14 128 30 146 SSRI aone Events Tota 8 119 Fixed effects (95% CI) Heterogeneity: c² = 0.75, df = 1 (p = 0.39); I ² = 0% Test for overa effect: z = 1.64 (p = 0.10) 30 142 OR (95% CI) 2.41 (0.82 to 7.10) 1.37 (0.71 to 2.67) 1.60 (0.91 to 2.82) Random effects (95% CI) 1.60 (0.91 to 2.83) Heterogeneity: t² = 0.00; c² = 0.75, df = 1 (p = 0.39); I ² = 0% Test for overa effect: z = 1.63 (p = 0.10) OR (95% CI) Dry mouth Corya et a. (2006) 49 Thase et a. (2007) 53 13 57 243 200 7 18 60 206 Fixed effects (95% CI) Heterogeneity: c² = 15.83, df = 1 (p < 0.0001); I ² = 94% Test for overa effect: z = 3.66 (p = 0.0002) 0.43 (0.16 to 1.12) 4.16 (2.35 to 7.38) 2.46 (1.52 to 3.98) Random effects (95% CI) 1.38 (0.15 to 12.89) Heterogeneity: t² = 2.43; c² = 15.83, df = 1 (p < 0.0001); I ² = 94% Test for overa effect: z = 0.28 (p = 0.78) Headache Corya et a. (2006) 49 Thase et a. (2007) 53 10 25 243 200 17 40 60 206 Fixed effects (95% CI) Heterogeneity: c² = 11.02, df = 1 (p = 0.0009); I ² = 91% Test for overa effect: z = 4.14 (p < 0.0001) 0.11 (0.05 to 0.25) 0.59 (0.34 to 1.02) 0.38 (0.24 to 0.60) Random effects (95% CI) 0.26 (0.05 to 1.39) Heterogeneity: t² = 1.32; c² = 11.02, df = 1 (p = 0.0009); I ² = 91% Test for overa effect: z = 1.57 (p = 0.12) Increased appetite Corya et a. (2006) 49 Thase et a. (2007) 53 16 64 243 200 7 12 60 206 Fixed effects (95% CI) Heterogeneity: c² = 20.96, df = 1 (p < 0.00001); I ² = 95% Test for overa effect: z = 4.89 (p < 0.00001) 0.53 (0.21 to 1.36) 7.61 (3.95 to 14.64) 3.60 (2.16 to 6.02) Random effects (95% CI) 2.06 (0.15 to 28.18) Heterogeneity: t² = 3.39; c² = 20.96, df = 1 (p < 0.00001); I ² = 95% Test for overa effect: z = 0.54 (p = 0.59) Somnoence Corya et a. (2006) 49 Thase et a. (2007) 53 22 35 243 200 5 11 Fixed effects (95% CI) Heterogeneity: c² = 3.82, df = 1 (p = 0.05); I ² = 74% Test for overa effect: z = 3.17 (p = 0.002) 60 206 1.10 (0.40 to 3.02) 3.76 (1.85 to 7.64) 2.56 (1.43 to 4.59) Random effects (95% CI) 2.15 (0.64 to 7.16) Heterogeneity: t² = 0.56; c² = 3.82, df = 1 (p= 0.05); I ² = 74% Test for overa effect: z = 1.24 (p = 0.21) Tremor Sheton et a. (2001) 51 Thase et a. (2007) 53 17 21 146 200 3 18 142 206 6.11 (1.75 to 21.32) 1.23 (0.63 to 2.38) Fixed effects (95% CI) 1.93 (1.11 to 3.37) Heterogeneity: c² = 5.07, df = 1 (p = 0.02); I ² = 80% Test for overa effect: z = 2.32 (p = 0.02) Random effects (95% CI) 2.50 (0.51 to 12.19) Heterogeneity: t² = 1.06; c² = 5.07, df = 1 (p = 0.02); I ² = 80% Test for overa effect: z = 1.14 (p = 0.26) 0.01 0.1 1 10 100 Favours SSRI + atypica Favours SSRI aone FIGURE 8 Resuts of meta-anaysis for adverse events (SSRI + AAP vs. SSRI aone). df, degrees of freedom. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 33

ASSESSMENT OF CLINICAL EFFECTIVENESS In addition to the adverse events reported here, there were aso data from one tria 49 on the adverse event of dizziness, for which there are simiar data for ithium, which are presented ater in this report. The Corya et a. tria 49 reported an OR of 0.31 (95% CI 0.13 to 0.73), suggesting a statisticay significant reduction in dizziness with fuoxetine + oanzapine compared with fuoxetine aone. The tota number of patients receiving fuoxetine aone was sma in comparison with the number receiving fuoxetine + oanzapine (60 vs. 243, respectivey). Reapse rate This outcome was not reported in any of the RCTs suitabe for incusion in this anaysis. Mortaity This outcome was not reported in any of the RCTs suitabe for incusion in this anaysis. Sensitivity anaysis assuming cass response The sensitivity anaysis, assuming that SSRIs have a cass effect and a AAPs have a cass effect, was conducted on the primary outcome of response (Figure 9). Response was chosen as it was expected a priori to be the outcome most commony reported by trias. The meta-anaysis for this outcome incuded a 10 trias identified for the comparison of SSRI + AAP with SSRI aone. 43,46 53 The resuts demonstrate a statisticay significant improved response with SSRI + AAP compared with SSRI aone when either the fixed or random-effects modes are used (fixed effects, OR 1.78; 95% CI 1.48 to 2.15; random effects, OR 1.83; 95% CI 1.38 to 2.42). This is consistent with the primary anaysis for response, where no cass effect is assumed (fixed effects, OR 1.48; 95% CI 1.13 to 1.94). This meta-anaysis incudes some patients receiving SNRI as a baseine treatment to be augmented with an AAP. The overa impact of this on the resuts is uncear as any bias introduced woud equay affect both tria arms within a RCT. The funne pot for this anaysis appeared to be symmetrica and the Egger's test was not statisticay significant (p = 0.14). Seective serotonin reuptake inhibitor pus ithium compared with seective serotonin reuptake inhibitor pus pacebo Ony one RCT 59 was identified for incusion in the anayses of SSRI + ithium compared with SSRI aone. As highighted earier, it shoud be noted that this tria is a surrogate tria used because of an absence of ithium trias in the required popuation. The patient popuation of Katona et a. 59 had faied to respond to one or more antidepressants in the current episode of depression, rather than two or more antidepressants as specified by the popuation incusion criterion. The use of a proxy tria faciitated an indirect comparison between augmentation of SSRI with ithium and augmentation of SSRI with an AAP. However, patients within this tria may be ess treatment resistant compared with the patients in the trias informing treatment SSRI + atypica Study Response Tota Berman et a. (2009) 48 Berman et a.(2007) 47 Corya et a. (2006) 49 Feng et a. (2008) 43 Marcus et a. (2008) 50 Mattingy et a. (2006) 46 Sheton et a. (2005) 52 Sheton et a. (2001) 51 Thase et a. (2007) 53 81 61 100 15 60 14 40 6 80 174 181 243 30 185 26 146 10 200 SSRI aone Response Tota 45 169 41 172 19 60 7 30 32 184 3 14 41 142 1 10 60 206 OR (95% CI) 2.40 (1.53 to 3.77) 1.62 (1.02 to 2.59) 1.51 (0.83 to 2.75) 3.29 (1.08 to 9.95) 2.28 (1.40 to 3.72) 4.28 (0.96 to 19.01) 0.93 (0.56 to 1.55) 13.50 (1.20 to 152.21) 1.62 (1.07 to 2.45) OR (95% CI) Fixed effects (95% CI) Heterogeneity: c² = 14.43, df = 8 (p = 0.07); I ² = 45% Test for overa effect: z = 6.01 (p < 0.00001) Random effects (95% CI) Heterogeneity: t² = 0.07; c² = 14.43, df = 8 (p = 0.07); I ² = 45% Test for overa effect: z = 4.26 (p < 0.0001) 1.78 (1.48 to 2.15) 1.83 (1.38 to 2.42) 0.01 0.1 1 10 100 Favours SSRI aone Favours SSRI + atypica FIGURE 9 Resuts of sensitivity meta-anaysis for response (SSRI + AAP vs. SSRI aone). df, degrees of freedom. 34 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 augmentation with AAP. The potentia impact of this difference in the tria popuations is discussed further in the discussion section (see Assessment of effectiveness, beow). Resuts for seective serotonin reuptake inhibitor pus ithium compared with seective serotonin reuptake inhibitor aone Response Two different definitions of treatment response were presented in the report for Katona et a., 59 neither of which directy met the review criteria of response (i.e. a 50% reduction in MADRS or HAMD score from baseine score) (Box 4). The post hoc definition for response was chosen by the tria authors to enabe the data from the tria to be compared with that of other simiar trias in a meta-anaysis that was conducted by the tria author and aso reported in Katona et a. 59 The resut of the tria using the a priori definition for response demonstrated a non-significant trend in favour of fuoxetine + ithium compared with fuoxetine aone (OR 1.48; 95% CI 0.37 to 5.95). Using the post hoc criteria for response there is a arger treatment effect that favours fuoxetine + ithium compared with fuoxetine aone but the difference between treatment effects remains non-significant (OR 3.85; 95% CI 0.80 to 18.62). Mean change in MADRS score The baseine MADRS score in Katona et a. 59 was 26.06 [standard deviation (SD) 4.93] in the fuoxetine + ithium group, and 26.25 (SD 5.29) in the fuoxetine-aone group. The MADRS scores at week 6 of augmentation therapy were 15.88 (SD 10.27) in the fuoxetine + ithium group, and 19.45 (SD 7.39) in the fuoxetine + pacebo group. The between-group mean change from baseine MADRS score at week 6 was 3.79 (fuoxetine + ithium vs. fuoxetine aone; 95% CI 11.25 to 3.67) and was statisticay non-significant. Remission Data for this outcome were not reported in the pubished RCT and were not avaiabe from the tria author or sponsor. Quaity of ife Data for this outcome were not reported in the pubished cinica tria and were not avaiabe from the tria author or sponsor. BOX 4 Definitions of response used in Katona et a. 59 Definition 1 (a priori definition) Reduction in HAMD score of at east 50% from Phase I baseine (i.e. prior to a 6-week ead-in phase of fuoxetine treatment). Reduction in HAMD score of at east 25% from Phase II baseine (i.e. prior to commencement of ithium or pacebo augmentation therapy). Fina HAMD score of < 13. Definition 2 (post hoc definition) Reduction in HAMD score of at east 50% from Phase I baseine (i.e. prior to a 6-week ead-in phase of fuoxetine treatment). Reduction in HAMD score of at east 25% from Phase II baseine (i.e. prior to commencement of ithium or pacebo augmentation therapy). Fina HAMD score of < 10. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 35

ASSESSMENT OF CLINICAL EFFECTIVENESS Withdrawas (a cause) Discontinuations from the Katona et a. 59 tria were sighty ower with fuoxetine + ithium (4/17) compared with fuoxetine aone (5/16). This difference was not statisticay significant (OR 0.68; 95% CI 0.15 to 3.16). The reasons for withdrawas in the ithium group were patient's decision (n = 1), physician's decision (n = 2) and protoco vioation (n = 1). The reasons for withdrawa with fuoxetine aone were ack of efficacy (n = 1), patient's decision (n = 2), protoco vioation (n = 1), and adverse event (n = 1). Withdrawas due to adverse events As aready described above, there was ony one withdrawa due to an adverse event reported in Katona et a. 59 This was in the fuoxetine-aone group. No further detais were provided in the additiona information received from the study sponsor for this outcome. Adverse events There was a non-statisticay significant difference in favour of treatment with fuoxetine-aone for the tota number of adverse events compared with the fuoxetine + ithium group (OR 1.95; 95% CI 0.43 to 8.83). The individua adverse event data for the outcomes of interest (as discussed above in the SSRI + AAP vs. SSRI aone comparison) are presented in Tabe 4. None of the resuts shows a statisticay significant difference between treatment groups. However, it shoud be noted that these are not necessariy the adverse events that woud be commony expected with treatment with ithium or fuoxetine; these wi be discussed further in the discussion section (see Assessment of effectiveness, beow). In addition, as a resut of the sma number of peope in the Katona et a. 59 tria, the number of peope reporting any singe adverse event is reativey sma and the study was not powered to detect between-group differences in adverse events. The resuts presented shoud thus be interpreted with caution. Reapse rate Data for this outcome were not reported in the pubished cinica tria and were not avaiabe from the tria author or sponsor. Mortaity Data for this outcome were not reported in the pubished cinica tria and were not avaiabe from the tria author or sponsor. Sensitivity anaysis assuming cass response Anaysis not required, as no additiona trias were identified as suitabe for incusion in this comparison. Seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor pus ithium Seven trias 43,49,51 53,59 were identified that met the criteria for incusion in the network meta-anaysis (MTC). Of these trias, six RCTs 43,49,51 53 compared fuoxetine + oanzapine with fuoxetine aone, and one tria compared fuoxetine + ithium with fuoxetine aone. 59 See Figure 10 for network diagram. TABLE 4 Adverse event resuts reported in Katona et a. 59 Fuoxetine + ithium Fuoxetine aone Adverse event n N n N OR (95% CI) Dizziness 0 17 1 16 0.30 (0.01 to 7.79) Dry mouth 2 17 1 16 0.44 (0.04 to 5.36) Headache 3 17 2 16 0.58 (0.08 to 4.01) Increased appetite 1 17 0 16 0.30 (0.01 to 7.79) Somnoence 1 17 2 16 2.00 (0.16 to 24.48) Tremor 1 17 0 16 0.30 (0.01 to 7.79) Tota no. of peope experiencing an adverse event 13 17 10 16 1.95 (0.43 to 8.83) 36 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Fuoxetine + atypica antipsychotic Fuoxetine + pacebo/no treatment Fuoxetine + ithium Direct comparison Indirect comparison Corya et a. (2006) 49 Feng et a. (2008) 43 Sheton et a. (2001) 51 Sheton et a. (2005) 52 Thase et a. (2007a) 53 Thase et a. (2007b) 53 Katona et a. (1995) 59 FIGURE 10 Network of RCTs incuded in the primary network meta-anaysis. In addition, the three RCTs 47,48,50 reporting antidepressant + aripiprazoe compared with antidepressant aone, the RCT 46 reporting antidepressant + quetiapine compared with antidepressant aone, and the RCT 30 comparing antidepressant + quetiapine with antidepressant + ithium were incuded in a cass-based sensitivity anaysis. The resuts from the MTC for the anayses of SSRI + AAP compared with SSRI aone, and SSRI + ithium compared with SSRI aone are reported in Tabe 5, together with the data for the main comparison of interest: SSRI + AAP compared with SSRI + ithium. A summary tabe detaiing which trias were incuded in each anaysis for each of the outcomes is presented in Appendix 6. The consistency of the evidence was assessed for each outcome using the posterior mean residua deviance, which shoud approximate the number of unconstrained data points in a good-fitting mode. For a outcomes other than mean change in MADRS and the sensitivity anaysis for discontinuations, the numbers of unconstrained data points were simiar to the vaue of the posterior mean residua deviance, suggesting that the chosen modes were good-fitting. The modes used for mean change in MADRS and the sensitivity anaysis for discontinuations wi be discussed further with the resuts beow. Where a random-effects mode was deemed the best fit, the degree of heterogeneity was investigated by evauating the posterior mean tau-squared (hereafter referred to as tau ). 39 The vaues of tau varied between 0.16 (sensitivity anaysis 1) and 0.47 (mean change in MADRS) for a but two outcomes (dry mouth and headache), suggesting the presence of some heterogeneity. 63 The vaues of tau for dry mouth and headache were higher (0.62 and 0.55, respectivey), suggesting moderate heterogeneity in these anayses, athough there were data from ony four trias in each anaysis. 63 Response Seven trias 43,49,51 53,59 were suitabe for incusion in this anaysis. A decision was taken to conduct two separate anayses for the outcome of response owing to the tria informing the comparison with ithium 59 reporting response using two different criteria. Neither of these criteria was directy comparabe with the definition of response used in the other trias incuded in this anaysis. A standard definition of response used in RCTs for depression is a 50% reduction in MADRS or HAMD score compared with baseine score. 64 The definitions used in the Katona et a. tria 59 are described in Box 4. For the purposes of comparabiity with the other trias in this anaysis, it is the response from Phase II baseine that is critica for the Katona et a. tria. 59 It is difficut to fuy assess the impact that the differences among the tria definitions of response may have on the overa resuts for this outcome, but this wi be discussed and expored further in the discussion section (see Assessment of effectiveness, beow). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 37

ASSESSMENT OF CLINICAL EFFECTIVENESS TABLE 5 Summary of resuts from the network meta-anaysis Outcome SSRI + AAP vs. SSRI aone: OR (95% CrI) SSRI + ithium vs. SSRI aone: OR (95% CrI) SSRI + ithium vs. SSRI + AAP: OR (95% CrI) Response (using ithium a 1.96 a (1.01 to 3.88) 2.26 (0.24 to 8.96) 1.29 (0.11 to 5.32) priori data) b Response (using ithium 1.99 a (1.03 to 4.01) 7.44 (0.59 to 35.30) 4.15 (0.25 to 20.34) post hoc data) b,c Mean change in MADRS 2.19 a ( 3.62 to 0.79) 3.11 ( 10.67 to 5.13) 1.47 ( 9.10 to 6.41) score from baseine b,d Withdrawas (a cause) e 1.27 (0.90 to 1.75) 0.92 (0.13 to 3.32) 0.74 (0.10 to 2.66) Tota adverse events e 1.71 (0.93 to 2.94) 2.87 (0.45 to 10.66) 1.84 (0.24 to 7.05) Somnoence b 2.75 (0.75 to 6.87) 1.14 (0.01 to 7.29) 0.58 (0.004 to 4.01) Dry mouth b 1.95 (0.45 to 5.51) 53.24 (0.13 to 124.8) 47.47 (0.06 to 92.44) Headache b 0.32 a (0.08 to 0.86) 4.48 (0.14 to 27.37) 20.49 (0.41 to 134.6) Sensitivity anaysis 1 b 1.99 a (1.41 to 2.93) 2.11 (0.86 to 4.95) 1.07 (0.44 to 2.32) Sensitivity anaysis 2 b 3.13 (0.85 to 10.29) 7.96 (0.56 to 37.70) 3.73 (0.13 to 19.01) Sensitivity anaysis 3 b 1.64 (0.89 to 3.32) 7.60 (0.68 to 34.57) 5.18 (0.36 to 23.24) Sensitivity anaysis 4 b,d 2.82 a ( 3.76 to 1.92) 1.89 ( 5.36 to 1.38) 1.27 ( 1.88 to 4.68) Sensitivity anaysis 5 e 1.24 (0.97 to 1.56) 0.91 (0.12 to 3.17) 0.75 (0.10 to 2.62) CrI, credibe interva. a Statisticay significant at p < 0.05. b Random-effects mode. c Primary anaysis for this review. d MD. e Fixed-effects mode. Notes Sensitivity anaysis 1: Cass-based sensitivity anaysis for response. Sensitivity anaysis 2: Sensitivity anaysis for response imiting to trias reporting faiure to two or more antidepressants in the current episode of depression. Sensitivity anaysis 3: Sensitivity anaysis for response imiting to trias reporting response based on the MADRS scae. Sensitivity anaysis 4: Cass-based sensitivity anaysis for mean change in MADRS. Sensitivity anaysis 5: Cass-based sensitivity anaysis for discontinuations. The resuts of the MTC for the outcome of response using the a priori data from Katona et a. 59 for the comparison of fuoxetine + ithium with fuoxetine + oanzapine were a non-significant trend in favour of treatment with ithium [OR 1.29; 95% credibe interva (CrI) 0.11 to 5.32], athough this was not statisticay significant. When the ithium data that met the post hoc definition of response from Katona et a. 59 were used (the primary anaysis for this review) there was a arger treatment effect favouring ithium augmentation over AAP augmentation but the difference between groups remained non-significant (OR 4.15; 95% CrI 0.25 to 20.34). In both instances, the best-fitting mode, as determined by the mode with the owest DIC, was a random rather than a fixed-effects mode (75 vs. 77 for ithium a priori anaysis and 74 vs. 77 for ithium post hoc anaysis). Mean change in MADRS scores Five RCTs provided data for this anaysis [Corya et a. (2006), 49 Katona et a. (1995), 59 Sheton et a. (2005) 52 and Thase et a. (2007a+b) 53 ] and the preferred mode was a random rather than a fixed-effects mode (DIC 26 vs. 28, respectivey). The MTC resuted in a MD of 1.47 (95% CrI 9.10 to 6.41) for the mean change in MADRS score from baseine for fuoxetine + ithium compared with fuoxetine + oanzapine, 38 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 which suggests a statisticay non-significant trend in favour of ithium augmentation. However, the wide 95% CrI indicates a arge amount of uncertainty in this estimate of treatment effect and so the resuts shoud be interpreted with caution. Remission There were no suitabe data to conduct an anaysis for this outcome. Quaity of ife There were no suitabe data to conduct an anaysis for this outcome. Withdrawas (a cause) Six RCTs 49,51 53,59 were suitabe for incusion in this anaysis and the preferred mode was a fixed-effects mode rather than a random-effects mode (DIC 58 vs. 60, respectivey). The resuts of the MTC for the comparison of fuoxetine + ithium with fuoxetine + oanzapine suggest a statisticay non-significant trend in favour of treatment with SSRI + ithium, i.e. withdrawas are ess ikey (OR 0.74; 95% CrI 0.10 to 2.66). Withdrawas due to adverse events There were insufficient data to conduct an anaysis for this outcome. Adverse events Three RCTs 43,52,59 provided data for the anaysis of the tota number of peope reporting adverse events. The fixed- rather than the random-effects mode was best-fitting (DIC 34.0 vs. 34.3, respectivey) and resuted in a statisticay non-significant benefit for fuoxetine + ithium compared with fuoxetine + oanzapine (OR 1.84; 95% CrI 0.24 to 7.05). However, the number of events informing this anaysis is sma and so the resuts shoud be interpreted with caution. The individua adverse events with sufficient data for comparison in the MTC were imited to somnoence, dry mouth and headache. A three outcomes were anaysed using a random-effects mode and they a incuded data from four trias. 49,53,59 The ORs for the comparison of fuoxetine + ithium with fuoxetine + oanzapine were as foows: somnoence OR 0.58 (95% CrI 0.004 to 4.01) dry mouth OR 47.47 (95% CrI 0.06 to 92.44) headache OR 20.49 (95% CrI 0.41 to 134.6). The 95% CrIs for a three of these outcomes are extremey wide and thus the estimates for the ORs are extremey uncertain and shoud be interpreted with caution. Reapse rate There were no suitabe data to conduct an anaysis for this outcome. Mortaity There were no suitabe data to conduct an anaysis for this outcome. A priori subgroup anayses The foowing prespecified subgroup anayses coud not be conducted owing to an absence of suitabe data from the cinica effectiveness trias incuded in this review: Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 39

ASSESSMENT OF CLINICAL EFFECTIVENESS (a) different durations of depression (i.e. time since first onset of current episode of depression) (b) casses of previous antidepressants (e.g. SSRI or tricycic antidepressant) (c) sex (d) age (i.e. those < 75 years and those 75 years) (e) peope with different severities of depression. Sensitivity anayses Detais of the five sensitivity anayses are as foows. Sensitivity anaysis 1 This sensitivity anaysis assumed that there is a cass effect for a SSRIs and aso a cass effect for a AAPs. It enabed the incusion of a 12 RCTs identified as suitabe for incusion in this review. 30,43,46 53,59 The random rather than the fixed-effects mode was best fitting (DIC 141 vs. 143, respectivey) and suggests a statisticay non-significant trend favouring treatment with SSRI + ithium compared with SSRI + AAP for the outcome of response (OR 1.07; 95% CrI 0.44 to 2.32), based on 50% reduction in MADRS or HAMD score and the post hoc definition of response used in the tria by Katona et a. 59 This is consistent with the resuts from the primary anaysis. Sensitivity anaysis 2 This sensitivity anaysis was decided post hoc to assess the impact of incuding some trias with patients with a faiure to respond to two or more antidepressants of which ony one treatment faiure was in the current episode of depression (the remaining faiures were in historica episodes of depression). The trias incuded in the sensitivity anaysis were imited to those in which a patients had a faiure to respond to two or more antidepressants in their current episode of depression. Four trias 51,53,59 met this strict criterion. A random rather than fixed-effects mode was best fitting for this outcome (DIC 35 vs. 36, respectivey). The resuts favour augmentation of SSRI with ithium over augmentation with oanzapine (OR 3.73; 95% CrI 0.13 to 19.01), athough this resut was non-significant. This is consistent with the resuts from the primary anaysis. Sensitivity anaysis 3 This sensitivity anaysis was conducted to assess the impact of imiting the primary anaysis to trias reporting response based on the MADRS scae. It resuted in the excusion of ony one previousy incuded RCT, 43 eaving a tota of six trias 49,51 53,59 incuded in this sensitivity anaysis. The random rather than fixed-effects mode was best fitting (DIC 64.1 vs. 64.5, respectivey) and resuted in a statisticay non-significant trend in favour of treatment with SSRI + ithium compared with SSRI + oanzapine (OR 5.18; 95% CI 0.36 to 23.24). This is consistent with the resuts of the primary anaysis. Sensitivity anaysis 4 This anaysis assumed a cass effect for a SSRIs and a cass effect for a AAPs and was used to assess the impact of the decision to use a non-cass-based assumption for the primary anaysis on the outcome measure of mean change in MADRS score. This sensitivity anaysis was conducted in addition to the sensitivity anayses on the primary outcome measure to enabe a cass-based sensitivity anaysis to be performed in the economic mode. Eight RCTs 30,47 50,52,53,59 were suitabe for incusion in this sensitivity anaysis. These trias comprised data from the six trias 30,49,52,53,59 in the primary anaysis on the comparison of SSRI (or venafaxine) + quetiapine with SSRI (or venafaxine) + ithium, as we as data from the three SSRI (or venafaxine) + aripiprazoe trias compared with SSRI (or venafaxine)-aone trias. 47,48,50 The data from two of the SSRI + aripiprazoe compared with SSRI-aone trias, 47,50 were avaiabe as a pooed anaysis 62 that was broken down by individua SSRI, and so these data were incuded as this enabed the excusion of the data from patients with baseine venafaxine treatment, thus reducing the amount of potentia cinica heterogeneity. The random rather than fixed-effects mode was deemed to be the best fitting (DIC 45 vs. 47, respectivey) and resuted in a MD of 1.27 (95% CrI 1.88 to 4.68) for SSRI + ithium compared with SSRI + AAP. This suggests a statisticay non-significant trend in favour of treatment with SSRI + AAP, which is in contrast to the resuts of the primary anaysis, in which the trend was in favour of 40 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 treatment with SSRI + ithium. However, both anayses demonstrate no significant difference in treatment effect between treatment with SSRI + ithium and with SSRI + AAP. Sensitivity anaysis 5 The sensitivity anaysis assumed a cass effect for a SSRIs and a cass effect for a AAPs for the outcome measure of withdrawas (a cause). This sensitivity anaysis was conducted to provide the additiona cinica data required to enabe a cass-based sensitivity anaysis to be performed in the economic mode. Ten trias were suitabe for incusion in the anaysis. 46 53,59 These trias incuded the six trias 49,51 53,59 from the primary anaysis, the three trias 47,48,50 with data on SSRI (or venafaxine) + aripiprazoe compared with SSRI (or venafaxine) aone and the Mattingy et a. 46 tria with data on SSRI (or SNRI) + quetiapine compared with SSRI (or SNRI) aone. The tria 30 for the comparison of SSRI (or venafaxine) + quetiapine with SSRI (or venafaxine) + ithium did not report subgroup data for peope with TRD defined as faiure to respond to two or more antidepressants in the current episode of depression and so coud not be incuded in this anaysis. The fixed rather than random-effects mode was the best fitting (DIC 94 vs. 96, respectivey). The resuting OR for the comparison of SSRI + ithium with SSRI + AAP was 0.75 (95% CrI 0.10 to 2.62). This suggests a statisticay non-significant trend in favour of treatment with SSRI + ithium (i.e. withdrawas are ess ikey). This resut is consistent with the resuts of the primary anaysis. Discussion and summary findings The avaiabe cinica effectiveness data informing the comparison of SSRI + AAP with SSRI + ithium in the primary anaysis were based on fuoxetine + oanzapine 43,49,51 53 compared with fuoxetine + ithium. 59 The resuts from the MTC of the star-shaped network demonstrate no significant differences between treatment regimens for any of the outcomes assessed. A non-significant trend in benefit was observed for the ithium-based augmentation strategy compared with the oanzapine-based augmentation strategy for response, mean change in MADRS from baseine, and fewer discontinuations. The resuts of the MTC aso demonstrated a non-significant trend in favour of the oanzapine-based augmentation strategy compared with the ithium-based augmentation strategy for fewer adverse events. However, care shoud be taken when interpreting non-significant resuts. When the resuts of the MTC are compared with the individua resuts for the pairwise meta-anayses there is genera agreement with the resuts obtained when SSRI is used as the baseine. This suggests a reasonabe fit of the modeing approach used within the MTC. The radiating star shape of the network means that ony the trias providing the resuts from the pairwise meta-anayses are providing the resuts within the MTC. The resuts for the ithium-based augmentation strategy compared with SSRI aone in the MTC tend to have wider 95% CrIs than the 95% CIs provided from the singe tria informing this comparison. 59 This is probaby owing to the random-effects mode tending to be the preferred mode for the MTC outcomes assessed. The resuts for both the pairwise meta-anaysis and MTC estimates of SSRI + AAP compared with SSRI aone showed a statisticay significant benefit in favour of augmentation with AAP for the outcomes of response and mean change in MADRS score. The equivaent resuts for SSRI + ithium compared with SSRI aone showed a statisticay non-significant trend in favour of augmentation with ithium. The resuts for ithium augmentation coud be considered inconcusive because they do not reach statistica significance, athough it shoud be noted that they are based on data from ony a sma subgroup of patients in one RCT. 59 In addition, other pubications have reported resuts that suggest ithium is an effective augmentation strategy. 60,65,66 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 41

ASSESSMENT OF CLINICAL EFFECTIVENESS A recent meta-anaysis by Crossey et a. 60 incuded 10 RCTs and demonstrated a statisticay significant benefit in terms of response rate with ithium augmentation compared with pacebo (OR 3.11; 95% CI 1.80 to 5.37). This meta-anaysis is not entirey comparabe with the patient popuation under review here because it incuded patients with bipoar disorder, patients on various antidepressants (incuding tricycic antidepressants), and patients with a minimum of one previous antidepressant faiure. However, the Crossey et a. 60 meta-anaysis does provide evidence to suggest that ithium is an effective augmentation agent in TRD and thus is supportive of the resuts from the MTC. In addition, the resuts from the fu tria popuation for the Katona et a. 59 RCT incuded in the MTC demonstrate a statisticay significant benefit in terms of response with ithium augmentation compared with pacebo (OR 3.21; 95% CI 1.09 to 9.48). Reapse rates and mortaity were prespecified as outcomes of interest for this review. However, none of the identified trias reported comparabe mortaity or reapse rate data. In addition, extremey imited subgroup data, if any, were reported for the trias incuded in the cinica effectiveness review and no tria reported suitabe subgroup data for the prespecified subgroup anayses. The trias incuded in this review were vaidated against the trias incuded in CG90 16 and those incuded in two separate systematic reviews; one for the comparison of AAP augmentation with pacebo in MDD 11 and the other for ithium augmentation compared with pacebo in TRD. 60 A three pubications incuded additiona trias compared with this review, but none of the additiona trias was found to be suitabe for incusion. Moreover, each tria was excuded for mutipe reasons based on the incusion criteria for this review. Mosty, the additiona RCTs did not meet the foowing criteria: the augmentation of SSRI as baseine therapy; 67 82 a 4-week minimum duration of treatment; 74 80,81 or two or more faiures of antidepressant therapy in the current episode of depression 67 73 (it is uncear how many trias in Crossey et a. 60 are affected by this criterion). The resuts of both the Crossey et a. 60 meta-anaysis evauating ithium compared with pacebo augmentation and the Neson et a. 11 review of AAP compared with pacebo augmentation demonstrated that the respective augmentation agents were statisticay significanty more effective than pacebo at achieving treatment response in peope with MDD or TRD. Sensitivity anayses The sensitivity anaysis based on a cass effect (sensitivity anaysis 1) aowed the incusion of two additiona AAPs (aripiprazoe and quetiapine) in the MTC. For the outcomes assessed (response), the non-significant trend favouring ithium augmentation observed in the primary anaysis is diminished (mean OR changes from 4.15 to 1.07) and remains non-significant. This coud be supportive evidence for no cinicay meaningfu difference between the two augmentation strategies, a refection of the ack of information avaiabe for the comparison, or an indication that the assumption of a cass effect is fawed and that there is a difference between the individua treatments within a cass. With regards to the ast concern, the tria providing information on quetiapine in the MTC 30 demonstrated a non-significant trend in favour of a quetiapinebased augmentation strategy compared with ithium-based augmentation (OR 1.25; 95% CI 0.74 to 2.12). The resut of this tria is thus in contrast to the resut of the MTC and therefore suggests a difference in treatment effect between the different AAPs. In the primary anaysis, RCTs were incuded if the majority of patients in each individua tria had experienced two or more previous faiures on an antidepressant in their current episode. A sensitivity anaysis (sensitivity anaysis 2) was conducted to determine if restricting this criterion to RCTs in which the whoe popuation had experienced two or more previous faiures in their current episode had a substantia impact on the resuts of the MTC. The resuts were consistent with the primary anaysis for response, that is, a non-significant trend in favour of ithium-based augmentation (using ithium post hoc data). A further sensitivity anaysis (sensitivity anaysis 3) was conducted as an exporatory anaysis to assess the impact of imiting trias in the primary anaysis to those reporting response on the MADRS scae. It resuted in the excusion of a singe RCT 43 that was conducted in China and in which a different cassification 42 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 system had been used for the diagnosis of depression. The other trias in the anaysis were mainy based in the USA and used the DSM cassification for depression. However, the resuts of this sensitivity anaysis were in keeping with the primary anaysis for response. An additiona sensitivity anaysis (sensitivity anaysis 4) was conducted for the economic mode on the outcome of mean change in MADRS score from baseine assuming a cass effect. Athough the resuts were simiar to the primary anaysis in terms of no significant difference, the trend changed from being in favour of ithium-based augmentation to being in favour of AAP-based augmentation. This change in trend is associated with the incusion of the Franco et a. tria. 30 As discussed earier, this tria demonstrates a non-significant benefit for response for quetiapine-based augmentation over ithium-based augmentation and the same trend is observed for reduction in MADRS scores. However, a MD in MADRS of 1.47 in favour of ithium-based augmentation from the primary anaysis or a MD of 1.16 in MADRS in favour of AAP-based augmentation is unikey to be cinicay meaningfu. 83 The fina sensitivity anaysis (sensitivity anaysis 5) was aso conducted for the benefit of the economic mode. This evauated the impact of a cass-based assessment of withdrawas (a cause). The resuts were simiar to the primary anaysis and demonstrated no significant difference in the risk of patients withdrawing from treatment with either augmentation strategy. Limitations The major imitation in this review of the cinica effectiveness of SSRI + AAP compared with SSRI + ithium in patients with TRD after faiing on two or more previous antidepressants was the absence of a direct comparison in a RCT. The number of RCTs avaiabe with data in the specified popuation, and at the right stage in their disease management, for comparison of the two augmentation strategies within a MTC was sma. In the primary anaysis this consisted of six RCTs for fuoxetine + oanzapine 43,49,51 53 and one RCT for fuoxetine + ithium. 59 In addition, this comparison was made possibe ony by incuding a surrogate tria in a sighty ess-severe popuation to aow the anaysis to contain the ithium tria. 59 This tria comprised around 50% of peope who did not meet the definition of TRD for this review, as they had faied ony one prior antidepressant in their current episode of depression. 59 Patients in this tria coud be ess severe than patients incuded for the fuoxetine + oanzapine comparison and so had a greater potentia to respond to treatment. However, as this potentia to have a greater response woud be simiar in both treatment groups within the tria the impact of this potentia source of bias on the anaysis is difficut to determine. Simiary, the RCTs identified for the primary anaysis provided sufficient data for ony a stepwise indirect comparison of the data, i.e. A versus B versus C, without the possibiity of drawing further strength and cohesion in the primary anaysis that a tria of A versus C woud have provided. This is compounded by the SSRI + ithium compared with SSRI aone ink being formed by ony a singe, sma tria (n = 33). 59 The impact of this can be observed in the reativey high eves of uncertainty obtained in any of the resuts of the MTC. In addition, the singe tria informing the comparison of SSRI + ithium with SSRI aone has what may be considered counterintuitive resuts for withdrawas (a cause), as fewer withdrawas occur in this study with fuoxetine + ithium than fuoxetine aone (4/17 vs. 5/16, respectivey). Lithium is known to have a different side effect profie to the SSRIs, with the additiona inconvenience of requiring frequent bood tests during treatment. 19 The resuts for this outcome shoud be treated with caution. None of the trias identified for incusion in the review provided information on a the outcomes assessed, which is another source of uncertainty around the estimated treatment effects. However, the fundamenta issue with regards to uncertainty is most profoundy driven by the singe sma tria identified for the ithium augmentation strategy. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 43

ASSESSMENT OF CLINICAL EFFECTIVENESS Definitions of response to treatment were aso different among the trias identified for incusion. The definition of response used in this review was 50% reduction in MADRS/HAMD score from baseine. This singe tria informing the ithium augmentation strategy described two different definitions used (one a priori and one post hoc), 59 neither of which was consistent with the one used in this review. This potentiay makes the resuts from this outcome ess reiabe for the comparison of the two augmentation strategies. Finay, a further imitation is that the network of RCTs identified by the systematic review was imited to those treatments within the scope of the review. Important inking studies (e.g. tricycic antidepressants or SNRIs used as a baseine for augmentation) coud have provided additiona information to enhance the precision and reiabiity of the estimates generated by the MTC. Overa The resuts of this review support the concusion that augmentation of SSRIs with ithium or an AAP is ikey to be beneficia in peope with TRD, defined as faiure to respond to two or more antidepressants in their current episode of depression. However, based on the imited number of RCTs identified in this research, the cinica evauation suggests there is no statisticay significant difference between the two augmentation strategies. There is a genera paucity of RCT data avaiabe in patients with TRD for SSRI + ithium and SSRI + AAP. 44 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 4 Assessment of cost-effectiveness In addition to addressing the decision probem that is the focus of this review, this report aims to compare the cost-effectiveness of the augmentation of SSRI therapy with ithium with augmentation with an AAP in a patient popuation with TRD (the decision probem is discussed further in Chapter 2, Decision probem). Assessment of the comparative cost-effectiveness of these two augmentation strategies is required to faciitate the effective and efficient aocation of heath-care resources within the NHS. It was anticipated that the evidence reating to the cost-effectiveness of augmenting SSRI treatment with ithium or with an AAP in a TRD patient popuation woud be imited. Therefore, a de novo economic evauation was carried out to compare the expected costs and benefits of the two augmentation strategies of interest. To ensure a of the reevant evidence from the economic iterature was incorporated into the mode, two systematic iterature reviews were conducted aongside mode deveopment. These reviews aimed to identify: economic evauations that coud inform the methodoogica approach to the deveopment of a de novo mode or improve understanding of the economic consequences of the disease area (economic iterature review) utiity vaues associated with depression and treatments for depression (QoL iterature review). The foowing sections in this chapter and the first six sections in Chapter 5 (see Introduction, Mode overview, Mode structure, Effectiveness data, Costs, and Sensitivity anaysis) describe the methodoogy and resuts of each review and detai the structure, assumptions and resuts of the de novo economic evauation. Economic iterature review A systematic review of the iterature was carried out to identify potentiay reevant economic evauations in the management of TRD. As discussed above, it was anticipated that the cost-effectiveness evidence reating to augmentation strategies in TRD woud be imited. Consequenty, the decision was taken to broaden the search to economic evauations of any intervention in patients with TRD. Therefore, the aim of the economic iterature review was to identify evidence that coud: inform the methodoogica approach to the deveopment of a de novo decision-anaytic mode or improve understanding of the economic consequences of the disease area. Literature search terms and strategies used to identify cost-effectiveness studies reevant to the decision probem were based on vaidated search strategies deveoped by Haynes et a. 84 and Dickersin et a. 85 Mutipe databases encompassing medica and economic iterature were searched to maximise the potentia of capturing reevant studies. Databases searched were: EMBASE (for the period 1988 to August 2011) MEDLINE, incuding MEDLINE In-Process & Other Non-Indexed Citations (for the period 1950 to August 2011) Cochrane Centra Register of Controed Trias (CENTRAL) (from inception to August 2011) HTA database (from inception to August 2011) NHS Economic Evauation Database (NHS EED) accessed via Wiey Onine Library (for the period 1999 to August 2011) PsycINFO (from inception to August 2011). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 45

ASSESSMENT OF COST-EFFECTIVENESS Detais of the search strategy are provided in Appendix 7. The search terms used covered condition, popuation and intervention and no country or anguage restrictions were appied. A references were exported to the Reference Manager (Thomson ResearchSoft, San Francisco, CA, USA) bibiographic database and dedupicated. In addition, the reference ists of identified systematic reviews were hand-searched for references. Incusion and excusion criteria, specific to the cost-effectiveness iterature review, were deveoped and are dispayed in Tabe 6. The search identified 438 studies, of which 104 were dupicates. One reviewer (LN) carried out the first appraisa of the tite and abstract of the 334 studies (eve one screening) for potentia incusion. At this stage, 301 papers were excuded. The 33 abstracts identified as being potentiay reevant were independenty appraised (eve two screening) by a second reviewer (NT) and 10 abstracts were excuded at this stage. The fu-text pubication of 23 studies identified as potentiay reevant by both reviewers (LN and NT) were ordered and independenty assessed for incusion, using the criteria outined in Tabe 6. Foowing the fu-text review, 19 studies were excuded and four economic evauations 86 89 were identified for fina incusion (Tabe 7). The study seection process is summarised in Figure 11. Overview of incuded economic evauations The four economic evauations identified in the systematic iterature review were carried out in various countries: Scotand, 86 Singapore, 87 Thaiand 88 and the USA. 89 The study by Benedict et a. 86 was from the perspective of the Scottish NHS. Therefore, the costs and resources considered in the evauation by Benedict et a. 86 were the most reevant to the decision probem that is the focus of this review. However, it is we estabished that TRD is associated with a substantia societa burden. 16,22 Benedict et a. 86 considered the impact of the societa perspective in sensitivity anaysis. The US study by Simpson et a. 89 took a simiar approach to that of Benedict et a. 86 (payer perspective as the base case and societa perspective in sensitivity anaysis), whereas the studies carried out in Thaiand 88 and Singapore 87 were conducted from a payer and societa perspective, respectivey. A summary of the four incuded economic evauations is given in Tabe 7; the quaity of each evauation (individuay discussed beow; see Narrative review of incuded studies) was assessed using the Phiips checkist 90 (the checkist for each study can be found in Appendix 8). A of the retrieved studies were recent, being pubished in 2009 or 2010, and are therefore ikey to refect current cinica practice in reation to TRD. Each study incuded a popuation of patients who were treatment resistant (either the whoe study popuation or a subgroup); TRD is defined as per the popuation criteria for this review. The heath states used in each of the identified studies informed the number and type of heath states incuded in the de novo mode (see Chapter 5, Mode overview and Mode structure). However, the TABLE 6 Incusion and excusion criteria used in the review of the economic evauation iterature Criteria Incusion Excusion Type of study Fu economic evauation Systematic review Popuation Geographica ocation Peope with TRD, defined in ine with the criteria for this review (patients who have faied to sufficienty respond to two or more adequate ines of SSRI therapy in their current episode of depression) Pubications from any country Peope with psychotic conditions, incuding bipoar I or II disorder None Interventions Any None Outcomes of interest QALYs, other heath outcome measures and expected costs None QALY, quaity-adjusted ife-year. 46 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Records identified through database searching (n = 438) Dupicates (n = 104) Records screened after removing dupicates at eve one screening (n = 334) Records excuded at eve one screening (n = 301) Records screened at eve two (n = 33) Records excuded at eve two screening (n = 10) Fu-text reports assessed for eigibiity (n = 23) Fu-text reports excuded (n = 19) Incuded papers in the review (n = 4) Augmenting with atypica antipsychotic in TRD (n = 1) Other TRD cost-effectiveness (n = 3) FIGURE 11 Number of cost-effectiveness studies identified and excuded at each stage in the appraisa process. methodoogica approach to the simuation of patients varied across the identified studies. Two of the studies used a decision tree mode, 87,88 one study used a Markov mode 86 and the ast 89 used a hybrid mode, combining a decision tree to simuate the acute phase of treatment and a Markov component to simuate the maintenance phase of treatment. The hybrid approach used by Simpson et a. 89 accounted for both the short (duration of acute therapy) and medium (1-year time horizon) term impications of treatment. Therefore, the study by Simpson et a. 89 was considered to be the most reevant to the current decision probem. Furthermore, none of the identified studies considered the onger-term (> 1 year after acute treatment) impications of treatment and the time horizons evauated ranged from 6 weeks 88 to 1 year. 86,89 This refects the paucity of ong-term foow-up data avaiabe for TRD patients. 16 Ony one study 88 evauated augmentation of antidepressant therapy [augmented with an AAP (aripiprazoe)]; the three remaining studies 86,87,89 evauated antidepressant monotherapy regimens compared with either each other or transcrania magnetic stimuation (TMS; Neuronetics, Inc., Mavern, PA, USA). Incrementa cost per quaity-adjusted ife-year (QALY) gained was assessed in a studies except that by Xie et a., 87 which imited the outcomes assessed to the percentage of primary and secondary care patients achieving remission and the associated costs. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 47

ASSESSMENT OF COST-EFFECTIVENESS TABLE 7 Summary of incuded economic evauations Author, year, country Perspective Mode type Patient popuation Intervention vs. comparator Outcomes ICER (per QALY gained) Leeahanaj et a. (2010), 88 Thaiand Payer Decision tree Patients with non-psychotic MDD who have incompetey responded to at east two antidepressants, incuding one prospective 8-week antidepressant treatment course Augmentation with aripiprazoe vs. augmentation with pacebo Incrementa cost per remission and incrementa cost per QALY 3201 Thai baht (2009) Xie et a. (2009), 87 Singapore Societa Decision tree Patients with MADRS score of 22 treated either in primary care or secondary care Escitaopram vs. venafaxine or fuvoxamine Percentage of primary and secondary care patients achieving remission and associated costs N/A Benedict et a. (2010), 86 Scotand Payer and societa Markov Patients with moderate to severe MDD (HAMD-17 score of 19) and patients with severe MDD (HAMD- 17 score of 25) were considered separatey Duoxetine (Cymbata, Ei Liy) vs. venafaxine ER or mirtazapine or SSRI Incrementa cost per QALY Primary care popuation (HAMD-17 score of 19): Duoxetine vs. venafaxine ER: duoxetine dominant Duoxetine vs. mirtazapine: 2353 Duoxetine vs. SSRIs: 6304 Secondary care popuation (HAMD-17 of 25): Duoxetine vs. venafaxine ER: duoxetine dominant Duoxetine vs. mirtazapine: duoxetine dominant Simpson et a. (2009), 89 USA Payer and societa Hybrid (decision tree pus Markov) Patients with pharmacotherapyresistant major depression in the current iness episode as measured by the ATHF (moderate to severe depression and moderate to severe resistance) TMS vs. sham TMS Incrementa cost per QALY Direct costs ony: TMS vs. pharmacotherapy as usua TMS vs. sham treatment US$34,999 Incuding productivity costs: TMS vs. sham treatment US$6667 TMS dominated pharmacotherapy ATHF, antidepressant treatment history form; ER, extended reease; ICER, incrementa cost-effectiveness ratio; N/A, not appicabe. 48 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Narrative review of incuded studies Leeahanaj et a. 2010 Leeahanaj et a. 88 undertook a cost-effectiveness anaysis comparing aripiprazoe with pacebo as adjunctive therapy to antidepressants in the treatment of non-psychotic MDD. The study incuded data from two trias in patients with MDD and with a history of inadequate response to between one and three adequate (appropriate dose given for > 6 weeks) antidepressant trias. Patients who then subsequenty faied (achieved partia response) to respond to treatment in a prospective 8-week tria of a different antidepressant (to those previousy administered) were randomised to adjunctive therapy with aripiprazoe or with pacebo. A cost-effectiveness anaysis was undertaken from a Thai payer perspective, beginning at the point of patient randomisation. Mode structure and assumptions A decision tree mode was constructed to cacuate the comparative costs and effects of adjunctive therapy with aripiprazoe or pacebo in Thaiand. The 6-week time horizon of the cost-effectiveness anaysis matched the duration of the randomised phase of the trias from which data were taken for anaysis. Patients entered the mode at the point of randomisation and received either aripiprazoe or pacebo. After randomisation, patients coud remit, discontinue treatment, or remain on therapy without achieving remission. Patients who did not remit were assumed to be hospitaised for ECT, after which a patients were assumed to be in a state of remission, incuding those who discontinued their therapy. The primary outcome of the mode was cost per remission, with remission defined as a tota MADRS score 14 of 10. Cost per QALY was assessed as a secondary mode outcome. Efficacy data In the mode, the efficacy of aripiprazoe is parameterised soey with data on remission and discontinuation. Remission and discontinuation rates are from a pooed anaysis of two identica US-based RCTs of adjunctive aripiprazoe in patients with MDD with inadequate response to antidepressant therapy. The authors reported that a utiity vaue of 0.8 was appied to the state of remission (the ony end heath state), from which the overa cost per QALY is then cacuated. This utiity vaue is from a study by Revicki et a. 91 Resource use and cost data The anaysis was conducted from a heath-care payer perspective and therefore accounts for ony direct heath-care costs, incuding drug costs (aripiprazoe and antidepressants), hospitaisation costs, and the cost of ECT. Patients achieving remission after augmentation therapy incurred the cost of 6 weeks of antidepressant therapy pus the cost of their augmentation drug (aripiprazoe = 195.5 baht per day, pacebo = 0 baht per day). Patients who did not remit or discontinue from augmentation therapy incurred 6 weeks of augmentation and antidepressant therapy, pus a further 20 days of antidepressant therapy. Patients discontinuing augmentation therapy incurred ony 20 days of antidepressant therapy. In addition, patients who did not remit after augmentation therapy aso incurred the costs of 20 inpatient bed-days and ECT; the estimate of hospitaisation time was obtained from a survey (no further detais of the survey are reported) and the mode assumed that ECT was performed eight times within the hospitaisation period. A costs were taken from the records of a oca hospita. Summary of resuts Remission rates foowing augmentation and overa costs of augmentation were higher with aripiprazoe than with pacebo (25.7% and 30,970 baht vs. 15.4% and 28,409 baht, respectivey). For aripiprazoe compared with pacebo, the incrementa cost per remission is reported as 2561 baht and the incrementa cost per QALY is reported as 3201 baht, respectivey. Based on these resuts, aripiprazoe is deemed not to be cost-effective according to the wiingness-to-pay threshod in Thaiand (not reported). Sensitivity anaysis indicated that the remission rate for aripiprazoe woud need to increase from 25.7% to 34.8% to achieve cost-effectiveness in Thaiand. Aternativey, the cost of aripiprazoe woud need to fa by 48.9%. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 49

ASSESSMENT OF COST-EFFECTIVENESS Critique The study by Leeahanaj et a. 88 evauated the popuation of interest for this review and used a simpe decision tree. However, the study has severa faws, which may imit its appicabiity to this review. Athough the time horizon used matched the ength of tria data avaiabe, 6 weeks is, as the authors concede, a short time frame and wi not inform the ikey ong- or medium-term cost-effectiveness of aripiprazoe. Moreover, the effectiveness data were derived from two US-based studies, which may not be generaisabe to a Thai popuation. Adverse events were omitted from the anaysis, which may resut in overestimation of the benefit of aripiprazoe. Most importanty, the key imitation of this study as a cost-effectiveness anaysis is the assumption that a patients remit foowing treatment with ECT. The consequence of this assumption is that there is no difference in the number of remissions or QALYs gained between treatments. Therefore, this anaysis may be considered akin to a cost-minimisation anaysis rather than a cost-effectiveness anaysis, an approach that is inappropriate between two treatments with significanty different efficacy estimates. Xie et a. 2009 Xie et a. 87 carried out a cost-effectiveness anaysis of escitaopram compared with venafaxine and aso with fuvoxamine in the treatment of patients with MDD. The authors constructed a decision-anaytic mode with two consecutive pathways; the first simuated patients initiating treatment and the second simuated patients who had two faied antidepressant treatments in primary care (who were therefore eigibe for augmentation, combination therapy or hospitaisation). A patients had a MADRS score 14 of 22 and the anaysis was conducted from a Singaporean societa perspective. The focus for this review is the second mode pathway, which considered patients with a MADRS score 14 of 22, treated in secondary care foowing the faiure of two treatments. Mode structure and assumptions The cost-effectiveness anaysis presented in the study considers patients with MDD for 6 months foowing the initiation of therapy, in accordance with the Singaporean depression cinica guideine 2004. 92 The mode used is a decision tree with two consecutive pathways. The first considers patients as they are initiating therapy in either primary or secondary care, according to cinica practice in Singapore. The second considers patients in secondary care who have faied two initia treatments. The primary outcome of the mode was incrementa cost per additiona remission achieved, with remission defined as a MADRS score 14 of 12. Foowing initiation, patients coud achieve remission, require titration because of ack of efficacy or switch therapy as a resut of adverse events. Patients achieving remission had the option of stopping treatment prematurey or continuing with therapy, after which patients either remained in remission or reapsed (with reapses defined as a MADRS score 14 of 25). The mode structure aowed a maximum of one titration and one treatment switch, after which patients who had sti not remitted were moved into the secondary care pathway and were eigibe for augmentation, combination therapy or hospitaisation. The rates of augmentation, combination therapy and hospitaisation were derived from a survey of GPs and psychiatrists and a outcomes were independent of treatment received. Foowing augmentation or combination therapy, patients coud achieve fu/partia response or no response. Responding patients had the option of prematurey ceasing therapy or remaining on therapy, after which they coud either remit or reapse. The fina mode heath states are remission, reapse or hospitaisation. Efficacy data Where possibe, data from head-to-head trias were used to parameterise the efficacy of escitaopram, venafaxine and fuvoxamine. However, the absence of a head-to-head tria of escitaopram compared with fuvoxamine ed to the use of efficacy data for citaopram as a proxy for the efficacy of fuvoxamine; evidence of non-significant difference in the efficacy and toerabiity of citaopram and fuvoxamine was used to justify the appropriateness of this approach. The rate of remission achieved with each drug is the main effectiveness parameter used in the mode. In the comparison of escitaopram with venafaxine, the remission rates of escitaopram and venafaxine are 50 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 taken from a meta-anaysis of patients treated in primary and secondary care. 93 However, in the comparison of escitaopram with fuvoxamine, the remission rates associated with escitaopram and fuvoxamine are taken from the head-to-head tria of escitaopram compared with citaopram (with the remission rates of citaopram used as a proxy for the remission rates of fuvoxamine). Foowing titration, the efficacy of each drug is assumed to decrease sighty, based on evidence from a cited cost-effectiveness anaysis. 94 Remission rates achieved foowing a treatment switch are assumed to be the same for a therapies, irrespective of the starting therapy. In the secondary care mode pathway, the rate of response (either fu or partia) repaces the rate of remission as the main driver of effectiveness. The response rate experienced by patients treated with augmentation or combination therapy is taken from a study of augmentation therapy by Posternak et a. 95 and response rates are assumed to be equa for augmentation or combination treatment strategy (as the outcomes of the secondary care mode pathway are independent of treatment aocation). The rate of reapse is another key cinica parameter in the mode that is assumed to be independent of treatment aocation. The rate of reapse demonstrated in studies of citaopram is used to represent the rate of reapse expected in first-ine therapy and foowing augmentation or combination therapy. A sighty ower reapse rate is associated with switch therapy and a higher rate of reapse is associated with patients who prematurey stop antidepressant therapy foowing remission. Resource use and cost data The perspective of this anaysis was societa and therefore incuded both direct and indirect costs, incuding ost productivity. Estimates of expected resource use were not readiy avaiabe from the iterature and so the authors conducted a survey among GPs and psychiatrists with experience in treating MDD. As part of this survey, the amount of contact with GPs or speciaists, hospitaisations and working days ost were estimated for each phase of treatment. The unit cost of drug acquisition, professiona consutation (either GP or psychiatrist) and hospitaisation were obtained from oca hospitas. The cost of absenteeism was cacuated using the human capita method. Summary of resuts Escitaopram dominated both fuvoxamine and venafaxine, with higher rates of remission and ower costs [escitaopram (68.1%, US$2845) vs. venafaxine (66.0%, US$3176); escitaopram (64.7%, US$3133) vs. fuvoxamine (60.0%, US$3297)]. Probabiistic sensitivity anaysis (PSA) demonstrated that, out of 5000 runs, 95% resuted in the dominance of escitaopram over venafaxine and 5% resuted in escitaopram yieding fewer heath benefits and ower costs than venafaxine. Escitaopram dominated fuvoxamine in 98% of runs and resuted in incrementa cost-effectiveness ratios (ICERs) of < US$10,000 in 2% of runs. No resuts were presented for the secondary care pathway, as transition probabiities were independent of treatment aocated. Critique The decision probem, objective and perspective of this anaysis were ceary stated and the outcome of cost per additiona remission was appropriate to address the decision probem. A pragmatic decision tree mode, originay deveoped in Europe, was we adapted to appy to the Singaporean heath-care system, based on consutation with medica professionas invoved in the treatment of MDD. The seected time horizon of 6 months was compatibe with the cinica guideine of depression in Singapore. However, this wi not have captured the onger-term costs and consequences of the treatment of MDD. The mode accounted for patients who were resistant to initia therapies in a pragmatic and consistent manner, based on consutation with the appropriate heath-care professionas (HCPs). Generay, the assumptions made regarding the pathway of resistant patients seem to be reasonabe. One assumption that contradicts avaiabe evidence is the assumption that reapse rates foowing augmentation or combination therapy woud be the same as foowing first-ine treatment. 96 However, this assumption can be regarded as conservative as any bias woud be directed against escitaopram. The source of effectiveness data was ceary described for patients at a Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 51

ASSESSMENT OF COST-EFFECTIVENESS stages of the mode pathways. However, as highighted by the authors, the use of tria data from Western Europe may not be generaisabe to an Asian popuation. In addition, it seems counterintuitive to assume different efficacy for the same drug, depending on which comparator is used; the efficacy estimates woud perhaps have been more robust had the avaiabe evidence been used to inform a MTC of a three considered treatment options. Resource-use data were ceary described and estimated from a survey of experienced cinicians in both primary and secondary care. Benedict et a. 2010 Benedict et a. 86 conducted an economic evauation comparing duoxetine with extended-reease (ER) venafaxine and aso with mirtazapine in the treatment of MDD in Scotand from a Scottish NHS perspective. Patients with moderate to severe MDD (HAMD-17 score of 19) and patients with severe MDD (HAMD-17 score of 25) were considered separatey in primary and secondary care mode scenarios. SSRIs were considered as a treatment option in patients with moderate to severe MDD, whereas patients with severe MDD referred to secondary care were assumed to have previousy received mutipe ines of SSRI therapy and therefore the comparators were imited to ER venafaxine and mirtazapine. Mode structure and assumptions A cost utiity mode was constructed to compare the QALYs and costs accrued with ER venafaxine and mirtazapine with those obtained with duoxetine. The anaysis was carried out over a 48-week time horizon. The time horizon was chosen to capture treatment duration as recommended by NICE 16 (treatment continued for 6 months foowing remission and reapses within 1 year). A Markov mode was used and heath states were a combination of patient's disease (i.e. depressed, response, in remission, no response, reapse, recurrence) and treatment status (i.e. acute treatment, continued treatment, switch treatment and no treatment). Cyce ength was 8 weeks (typica treatment duration) and patients were abe to discontinue treatment in a mode cyces. A patients entered the mode in the acute treatment heath state, from which they coud experience remission, response, no response or drop out/discontinue their therapy. Patients who remitted were at risk of reapse, regardess of treatment status. Patients who responded but did not remit and remained on treatment coud experience remission, response, no response or drop out in the subsequent cyce. Simiary, non-responders who did not discontinue coud experience remission, response or continued non-response. Patients were eigibe to switch treatments foowing recurrence of symptoms after initia treatment discontinuation, no response foowing previous response or remission, or continued non-response. Efficacy data The accumuation of QALYs is driven by the rate of response, remission and discontinuation throughout the acute and maintenance phases of treatment. The utiity vaues associated with response, remission and discontinuation were derived from the European Quaity of Life-5 Dimensions (EQ-5D) data of 300 European patients with MDD from the Ei Liy HMBU tria (data on fie). In moderate to severe patients, effectiveness data for acute treatment were synthesised from eight RCTs for duoxetine, two head-to-head trias for ER venafaxine and pubished meta-anayses for mirtazapine and SSRIs; a cass effect was assumed for SSRIs. Data for effectiveness of maintenance treatment in moderate to severe patients were taken from a head-to-head tria comparing duoxetine with ER venafaxine. However, the absence of maintenance effectiveness data for mirtazapine and the SSRIs ed to the appication of a weighted average of the effectiveness of duoxetine and venafaxine. In patients with severe MDD, the probabiities of response to acute treatment with duoxetine and ER venafaxine were taken from two head-to-head trias. Reative differences were used to cacuate the acute and maintenance probabiities for mirtazapine, and maintenance probabiities for duoxetine and ER venafaxine. Resource use and cost data The perspective of the anaysis was that of the Scottish NHS; therefore, direct costs are the main consideration. However, societa costs of ost productivity are accounted for in the sensitivity anaysis. The number of GP visits for menta heath reasons, psychiatrist visits, hospitaisations, and visits to accident and emergency departments were derived from expert panes in both primary and secondary care. The average 52 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 doses of SSRIs and mirtazapine were taken from UK market information. 97 The doses used in head-to-head trias of ER venafaxine and duoxetine were assumed to represent standard dosing of these drugs. The authors reported that unit costs were taken from pubished UK sources (no reference was provided). Summary of resuts Duoxetine was associated with ower costs than venafaxine in the severe popuation and ower costs than mirtazapine and venafaxine in the moderate to severe popuation. More QALYs were accrued with duoxetine than any other treatment in both patient popuations. However, the incrementa gain was sma (0.003 and 0.005 compared with venafaxine in patients with moderate to severe MDD and severe MDD, respectivey). Consequenty, duoxetine dominated (i.e. was more effective and ess costy) ER venafaxine in both patient popuations and dominated mirtazapine in the moderate to severe patient popuation. In the moderate to severe MDD, the ICERs for duoxetine compared with mirtazapine and the SSRIs were 2353 and 6304 per QALY, respectivey. The authors concuded that duoxetine dispays simiar efficacy and a different side effect profie to ER venafaxine and that it represents an important treatment option for patients with MDD in the UK. Critique Overa, the study was considered to be high quaity; the assessment of cost per QALY gained is appropriate to address the decision probem from the perspective of the Scottish NHS. A mode inputs are ceary described and the anaysis has severa strengths, incuding the use of head-to-head RCT data to inform the comparison of duoxetine and ER venafaxine and the use of (unpubished) utiity vaues derived from RCT data. The chosen time horizon of 48 weeks seems appropriate to gather the different short- and medium-term costs and outcomes associated with the treatments of MDD. The consideration of two different patient popuations is aso a strength, as patients are managed differenty according to their history and whether they are treated in primary or secondary care. 16 For the purposes of this review, the focus is on severe patients in secondary care who are assumed to have received mutipe ines of SSRI therapy. The effectiveness data used to inform this aspect of the anaysis are weaker than those forming the basis of the anaysis of first-ine primary care patients. Athough head-to-head RCT data are used to inform the acute outcomes of treatment with duoxetine and ER venafaxine in the acute phase, maintenance therapy is based on reative differences between first and second cyce probabiities (detais of cacuations were not provided in the paper). Resource-use data were coected from a pane consisting of two GPs and one psychiatrist in primary care and two psychiatrists in secondary care. No rationae for the choice of mode type was provided but a Markov mode may be appropriate given the cycica nature of remission and reapse often seen in depressed patients. 98 In addition, it is not cear to what extent the progress of patients is tracked through cyces of response, reapse, treatment switch and disease recurrence. In particuar, patient's progression foowing switch treatment is uncear. The authors were contacted to carify this; no response has been received at the time of writing the fina report. Simpson et a. 2009 Simpson et a. 89 present a cost-effectiveness anaysis of TMS in patients with moderate to severe unipoar non-psychotic MDD, defined as a MADRS score of 17. Patients were aso moderatey to severey pharmacoogicay treatment resistant [as measured by the antidepressant treatment history form (ATHF)], having received no cinica benefit from between one and four adequate antidepressant exposures. Anaysis was undertaken from a US payer and societa perspective. TMS was compared with sham TMS and pharmacotherapy as usua. Efficacy data for pharmacotherapy as usua were based on the cinica outcomes observed in the pubished resuts of the Sequenced Treatment Aternatives to Reieve Depression (STAR*D) tria. 99 The STAR*D tria 99 was a series of RCTs in outpatients with non-psychotic MDD who were eigibe for medication as first-ine treatment. Patients were evauated across sequentia ines of therapy (up to four ines Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 53

ASSESSMENT OF COST-EFFECTIVENESS of therapy) and guidance was provided on how to start therapy and how to proceed if initia treatment faied. 100 Mode description The authors used a hybrid mode, consisting of decision tree and Markov components. The decision tree component was used to simuate treatment outcomes foowing 6 weeks of acute treatment and a panned 3-week taper phase. Patients were cassified as we (MADRS score of 0 9) or midy (MADRS score of 10 17), moderatey (MADRS score of 18 27) or severey depressed (MADRS score of > 27). Foowing the acute treatment phase of the mode, patients were disaggregated into two separate Markov components. The first accounted for the progression of patients who were we or who had mid or moderate depression foowing acute treatment. The second Markov component accounted for the progression of patients who were severey depressed foowing acute treatment; these patients were assumed to have faied at east two antidepressants and TMS therapy. Within each Markov component patients coud move between the heath states of we or mid, moderate or severe depression in quartery (3-month) cyces over a 1-year time horizon. Effectiveness data The decision tree component of the economic mode was parameterised with acute treatment outcomes and severity-specific rates of reapse were used to drive the Markov components. A TMS and sham TMS cinica efficacy parameters were derived from anaysis of raw cinica effectiveness data, provided by the manufacturer of TMS (Neuronetics, Inc., Mavern, PA, USA). Patients with severe depression foowing acute TMS treatment were assumed to have the same potentia to benefit from further antidepressant therapy as patients in eves 3 and 4 of the STAR*D study 99 (i.e. patients who had faied on two or three previous therapies). Patients who were we or had mid-to-moderate depression foowing acute TMS treatment were assumed to have the same potentia to benefit from further antidepressant therapy as patients in eves 2 and 3 of the STAR*D study 99 (i.e. patients who had faied on one or two previous treatments). In addition, a subgroup of TMS patients who had faied ony one previous therapy in their current episode of depression were compared with eve 2 patients of the STAR*D tria. 99 QoL weights were obtained from a study by Revicki et a. 91 Resource use and cost data Heath-care resource utiisation was estimated for each heath state using data from the resuts of sef-report questionnaires used in the Neuronetics trias. The questionnaire covered ost productivity, heath-care utiisation and costs, and caregiver support. Unit cost data were taken from the 2004 Medicaid biing database for patients with depression or the Neuronetics studies; a costs were infated to 2006 vaues. The overa cost associated with each treatment regimen was dependent on the composition of the patient popuation, treatment efficacy and costing assumptions used. Anayses were carried out from both the payer and societa perspective. Resuts Mean annua costs (excuding ost productivity) of STAR*D patients varied from US$4379 to US$26,546 for patients who responded to initia therapy and non-responders, respectivey; non-responders often required hospitaisation and mutidrug treatment. Simiary, mean annua costs for TMS patients (excuding the cost of TMS) ranged from US$3683 for responders to US$26,599 for non-responders. The cost of TMS was estimated to be US$300 per treatment. Incrementa cost-effectiveness resuts were presented for the foowing comparisons: TMS vs. sham TMS (based on cinica data from a RCT) TMS vs. sham in patients with ATHF = 1 (based on subgroup anaysis of a RCT) TMS vs. pharmacotherapy as usua (based on an open-abe TMS study and the eve 2 and 3 outcomes of STAR*D 99 ) TMS vs. pharmacotherapy as usua in patients with ATHF = 1 (based on a subgroup of an open-abe TMS study and the eve 2 outcomes of STAR*D 99 ). 54 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 No resuts were presented for the more severe patient popuation. TMS resuted in cost savings (i.e. dominated) of US$746 and US$2243 compared with pharmacotherapy as usua in the eve 2/3 STAR*D patients and the eve 2 STAR*D patients, respectivey. These cost savings increased to US$7243 and US$9844 when ost productivity and increased caregiver costs, respectivey, were incuded in the anayses. The comparison of TMS with sham TMS (i.e. no treatment) resuted in ICERs of US$36,551 and US$29,556 in a randomised patients and patients with ATHF scores of 1, respectivey. The incusion of ost productivity and caregiver costs into the anayses resuted in an ICER of US$3544 for TMS compared with sham in a eigibe patients and cost savings of US$5092 in patients who have faied ony one previous antidepressant therapy. Critique The study presented by Simpson et a. 89 had many strengths, not east of which was the arge voume of acute treatment data derived from RCTs used to inform the mode transitions. Furthermore, the ength of foow-up of RCT patients was sufficient to address questions over the durabiity of treatment effect. However, the description of which method was used to extrapoate 6-month foow-up data to 1 year was insufficient to determine its methodoogica robustness. The comparison of TMS with the STAR*D study 99 was argey refective of cinica practice. However, the expected difference in costs and outcomes between TMS and distinct treatment regimens or sequences was not quantified. The use of individua patient surveys to coect heath-care resource utiisation data is another particuar strength of this anaysis. The use of QoL data from Revicki et a. 91 is consistent with many studies in this disease area. However, adverse events and treatment-reated QoL were not considered. The absence of adverse events is the main weakness of this study, aong with insufficient reporting of gains in QALYs and the inconsistent reporting of ICERs between the summary tabe and the text. Quaity-of-ife iterature review Introduction This iterature review was carried out to identify utiity vaues associated with depression heath states in a patient popuation with TRD. As in the review of the economic iterature, it was expected a priori that, if avaiabe, QoL iterature on TRD woud be imited. In addition, it was considered that the heath-reated quaity of ife (HRQoL) associated with TRD was unikey to be different to that associated with depression in genera. Therefore, a decision was made to expand the remit of the search to consider studies of depression in genera rather than imiting to studies in TRD. Hence, the primary objectives of this review were to identify heath-state utiity vaues (HSUVs) associated with different eves of depression. The secondary objective of this review was to identify any issues that positivey or negativey impact upon the QoL in depressed patients. The review of the QoL iterature identified five HSUV studies that addressed the primary objective of the review: to retrieve utiity vaues associated with different eves of depression. A further 12 studies 30,43,46 53,59 were identified that addressed the secondary objective: to understand the scope of issues that affect quaity of ife in the TRD patient popuation. The sections that foow provide an overview and quaity assessment of the HSUV studies identified as addressing the primary objective of this review; the rationae for the utiity vaues used in the economic mode; and an overview of the QoL studies identified as addressing the secondary objective of this review. The quaity assessment of HSUV studies is based on quaity assessment criteria outined by NICE's decision support unit: Technica Support Document 9 2011 (TSD 9 2011). 101 As recommended by the decision support unit, factors considered when evauating quaity incuded seection and recruitment of respondents, incusion and excusion criteria and whether or not the study incuded a description of the baseine characteristics of the popuation from which vaues were derived. Response rates of the measures used to derive the HSUVs and oss to foow-up were aso evauated. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 55

ASSESSMENT OF COST-EFFECTIVENESS Systematic iterature search and seection process As stated above, a systematic iterature search was carried out to identify utiity vaues associated with depression heath states and to identify any issues that positivey or negativey impact on the QoL of depressed patients. Literature search terms and strategies used to identify cost-effectiveness studies reevant to the decision probem were based on vaidated search strategies deveoped by Haynes et a. 84 and Dickersin et a. 85 Mutipe databases encompassing medica and economic iterature were searched to maximise the potentia of capturing reevant studies. Searches were carried out in the foowing databases: EMBASE (for the period 1988 to August 2011) MEDLINE, incuding MEDLINE In-Process & Other Non-Indexed Citations (for the period 1950 to August 2011) CENTRAL (from inception to August 2011) HTA database (from inception to August 2011) NHS EED accessed via Wiey Onine Library (for the period 1999 to August 2011) PsycINFO (from inception to August 2011). Detais of the search strategy are provided in Appendix 7. The search terms used covered condition, popuation and intervention; no country or anguage restrictions were appied. A references were exported to the Reference Manager bibiographic database and dedupicated. In addition, the reference ists of identified systematic reviews were hand-searched for additiona references. Incusion and excusion criteria specific to the QoL iterature review were deveoped (Tabe 8). The initia search identified 352 papers, of which 114 were dupicates. One reviewer (LN) carried out the first appraisa of the tite and abstract of the 238 studies (eve one screening) for potentia incusion. At this stage, 205 papers were excuded. Further assessment of incuded abstracts (eve two screening) was carried out by a second reviewer (NT) and 13 abstracts were excuded. The fu-text pubications of 29 studies identified as potentiay reevant by both reviewers were ordered (incuding nine studies identified through bibiographic hand-searching). The papers were independenty assessed for fina incusion by two reviewers (LN and NT) using the criteria presented in Tabe 8: 12 studies were excuded at this stage. Figure 12 summarises the seection process. Overview of incuded heath-state utiity vauation studies The five HSUV studies identified in the QoL iterature review reported utiity vaues for depression severity and/or treatment response. Two studies 102,103 reported utiity by depression severity, one study 104 reported utiity associated with treatment response and two studies 91,105 considered both depression severity and treatment response. Most studies were carried out in North America (three in Canada 91,102,103 and one in the USA; 104 the remaining study was carried out in France 105 ). The studies were pubished between 1998 and TABLE 8 Incusion and excusion criteria appied in the review of the QoL iterature Criteria Incusion Excusion Type of study Any study type None Popuation Peope with TRD, MDD or any severity of depression Peope with bipoar disorder or psychotic conditions Geographica ocation Pubications from any country None Interventions Any None Outcomes of interest Utiity scores, method of eicitation/vauation Genera QoL issues None 56 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Records identified through database searching (n = 352) Dupicates (n = 114) Records screened after removing dupicates at eve one screening (n = 238) Records excuded at eve one screening (n = 205) Records screened at eve two (n = 33) Records excuded at eve two screening (n = 13) Additiona papers identified through hand searching (n = 9) Fu-text reports assessed for eigibiity (n = 29) Fu-text reports excuded (n = 12) Incuded papers in the review (n = 17) Papers reporting utiity scores (n = 5) Papers reporting HRQoL in depression (n = 12) FIGURE 12 Fow diagram of studies in systematic review to identify HSUVs. 2004 and sampe sizes ranged from 58 to 250. The studies each used a different too to evauate heath-state utiity: the toos incuded the EQ-5D questionnaire, the sef-administered Quaity of We-Being Scae (QWB-SA), the SF-36, and the McSad instrument. The mean age of the study participants was < 50 years and more than two-thirds of study participants were women. Tabe 9 beow summarises the five incuded studies. 91,102 105 Narrative review of incuded studies (critique of identified quaity-of-ife studies) Revicki et a. 1998 Revicki et a. 91 used standard gambe (SG) techniques to eicit utiity vaues from 70 Canadian (57% of participants) and US (43% of participants) primary care patients. Patients had MDD (DSM-III-revised) with mean HAMD score of 11.65 (SD 8.2). Patients were either currenty receiving treatment or had competed an antidepressant treatment in the past 2 months. The mean age of participants was 42 years (SD 11 years) and 77% were femae. Heath status was measured using the SF-36. Patients were presented with 11 hypothetica depression-reated heath states based on vignettes for depression heath states and treatment. The vignettes were informed by iterature and three psychiatrists experienced in treating depression. Utiity was generated by appying a structured SG interview for each hypothetica heath state. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 57

ASSESSMENT OF COST-EFFECTIVENESS TABLE 9 Summary of utiity vaues reported in a incuded HSUV studies Author, year, country Sampe Instrument size Patient popuation a (vauation) Utiity resuts Pyne et a. (2003), 104 USA 58 Patients with MDD (DSM-IV criteria) QWB-SA Responders Baseine: 0.43 4 weeks: 0.54 4 months: 0.63 Non-responders Baseine: 0.41 4 weeks: 0.46 4 months: 0.43 Revicki et a. (1998), 91 Canada 70 Patients with unipoar depression seen in US and Canadian primary care The heath states were hypothetica, based on vignettes Mean (SD) Current heath state (foowing at east 8 weeks of antidepressant treatment) 0.74 (0.22) Severe depression untreated 0.30 (0.28) Moderate depression Nefazodone: 0.63 (0.23) Fuoxetine: 0.63 (0.19) Imipramine: 0.55 (0.03) Mid depression Nefazodone: 0.73 (0.21) Fuoxetine: 0.70 (0.20) Imipramine: 0.64 (0.20) Remission: maintenance treatment Nefazodone: 0.83 (0.13) Fuoxetine: 0.80 (0.15) Imipramine: 0.72 (0.17) Remission: no treatment 0.86 (0.16) Schaffer et a. (2002), 103 Canada 75 A mixture of peope with current depression, previous depression (according to MDD DSM-IV criteria) and heathy peope The heath states were hypothetica, based on vignettes Mean (SD) Currenty depressed patients Severe: 0.31 (0.31) Moderate: 0.51 (0.34) Mid: 0.59 (0.33) Previousy depressed patients Severe: 0.47 (0.34) Moderate: 0.67 (0.36) Mid: 0.79 (0.28) 58 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 9 Summary of utiity vaues reported in a incuded HSUV studies (continued) Author, year, country Sampe Instrument size Patient popuation a (vauation) Utiity resuts Heathy individuas Severe: 0.46 (0.28) Moderate: 0.69 (0.29) Mid: 0.80 (0.21) A participants Severe: 0.43 (0.31) Moderate: 0.64 (0.33) Mid: 0.75 (0.28) Bennett et a. (2000), 102 Canada 105 Patients currenty in remission, with history of unipoar depression in the past 2 years McSad (SG and VAS) Mean (95% CI) Sef-reported heath-state 0.79 (0.74 to 0.83) Temporary states (6 months' duration) Mid: 0.59 (0.55 to 0.62) Moderate: 0.32 (0.29 to 0.34) Severe: 0.09 (0.05 to 0.13) Chronic states (ifetime duration) Severe: 0.04 (0.009 to 0.07) Sapin et a. (2004), 105 France 250 Patients with new episodes of MDD (DSM-IV) seen in primary care EQ-5D (TTO) Mean (SD) Disease severity heath states (after 8 weeks of treatment) Mid: 0.74 (0.19) Moderate: 0.44 (0.27) Severe: 0.30 (0.27) Treatment response heath states (after 8 weeks of treatment) Remission: 0.85 (0.13) Response: 0.72 (0.20) Non-response: 0.58 (0.28) NDC, non-directive counseing; TTO, time trade-off; VAS, visua anaogue scae. a Patients enroed in the study. Detais of who provided the heath-state descriptions and who provided the vauations are provided in the narrative section beow. Heath states were a combination of depression severity (mid, moderate or severe) and antidepressant treatment [nefazodone (Serzone, Bristo-Myers Squibb), fuoxetine or imipramine]. Heath-state profies were framed as enduring for 1 month and each profie contained descriptions of symptom severity, functioning and we-being, and side effects from treatment. The authors concuded that there were significant differences between the mean utiity scores generated for the hypothetica heath states, depending on a patient's current severity of depression; more severey depressed patients (HAMD score of 15) and patients experiencing three or more side effects gave ower scores for hypothetica remission states than patients currenty in remission. Athough the study was of good quaity, despite reporting the number of patients with missing or incompete utiity data, the authors did not state how they had deat with missing or incompete data. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 59

ASSESSMENT OF COST-EFFECTIVENESS Bennett et a. 2000 Bennett et a. 102 used a disease-specific measure, the McSad instrument, to estimate utiity scores for a crosssectiona sampe of 105 Canadian patients. Patients were currenty in remission and had experienced at east one episode of major unipoar depression in the previous 2 years. The mean age of respondents was 41.7 years (SD 8.7 years) and 74% were femae. The authors reported that McSad is a direct utiity measure of depression heath states that uses a combination of rating scae and SG techniques to obtain utiity vaues. Based on the DSM-III-revised criteria for major unipoar depression, McSad assesses six dimensions: emotion; sef-appraisa; cognition; physioogy; behaviour; and roe function. Patients provided utiity vaues for four depression heath states, incuding three hypothetica cinica marker heath states of untreated depression (mid, moderate and severe) and current sef-reported heath. Utiity scores for the three cinica marker heath states were framed as enduring for 6 months, and the chronic states (sef-reported and severe depression) were measured assuming ifetime duration. The authors concuded that depression is associated with poor HRQoL ; moderate depression generated utiity vaues that were ower than those reported for patients who are bind, deaf and dumb. However, the authors aso stated that these findings may not be generaisabe, as the extent to which the sampe is representative is unknown. The study appears to be of good quaity, as it satisfies most of the NICE TSD quaity criteria. 101 Schaffer et a. 2002 Schaffer et a. 103 eicited utiity scores from a mixed popuation in Canada, consisting of 40 patients with MDD (meeting DSM-IV criteria) and 35 heathy peope. The mean age of participants was 42.4 years (SD 11.4 years) and 67.5% were femae. Utiity vaues were assigned using SG techniques. Participants were presented with 10 profies describing different symptoms of depression and three profies describing depression of a mid, moderate or severe nature. The individua symptom profies contained five descriptive statements, which were derived from depression rating scaes and interviews such as the Beck Depression Inventory (BDI), DSM-IV, HAMD and MADRS. Somatic symptoms (e.g. ow energy and decreased appetite or seep) were rated more highy than psychoogica symptoms (e.g. suicida ideation, depressed mood and anhedonia), with suicida ideation having the owest utiity score. Peope with current depression generay assigned ower utiity scores to profies than did heathy peope or participants with a history of depression. In addition, no significant difference in utiity was found between peope with a history of depression and heathy individuas. The authors suggested that the presence of current depression may affect utiity scores and that this coud be an important consideration in patients with primary diagnoses other than depression. However, the authors aso highighted the imitations to this study of sma sampe size and the use of SG techniques in depressed patients, who as a consequence of their iness may not be as risk averse as peope who are not depressed. Overa, the study was of good quaity, with respondent seection, recruitment, and incusion and excusion criteria ceary described. There was no discussion around patients ost to foow-up but the resuts suggest that a patients enroed were foowed up for the duration of the study. Pyne et a. 2003 Pyne et a. 104 used the QWB-SA in a prospective observationa study. Fifty-eight US patients with a current diagnosis of MDD were monitored over 16 weeks to assess the reationship between QWB-SA scores and depression severity. Consistency was assessed by comparing scores obtained with the QWB-SA and those obtained from the interviewer-administered version (interviewer-qwb). The mean age of participants was 45.7 years (SD 10.3 years), 83% of participants had unipoar depression and 17% had bipoar depression, and 78% were mae. The QWB-SA is a generic HRQoL instrument with five domains (symptoms, sef-care, mobiity, physica function and performance of usua activity). The output of QWB-SA is a quaity-adjusted index score of between 0 (death) and 1 (perfect heath) derived from scores in the five domains. 106 QWB-SA scores improved during foow-up for treatment responders (defined by a 50% reduction in HAMD-17 scores from baseine) but did not improve for non-responders. The authors concuded that QoL was associated with depression severity, with responders having a better QoL than non-responders. The study appears to be of good quaity with ceary described methods. 60 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Sapin et a. 2004 Sapin et a. 105 examined the impact of MDD on patients' HRQoL using data coected from a muticentre non-comparative prospective cohort study in patients with a new episode of MDD. Patients (n = 250) were recruited in the French primary care setting and foowed for 2 months. The mean age of participants was 44.2 years (SD 14.1 years) and 72% were femae. Patient preferences were eicited through the use of the EQ-5D questionnaire. The resutant heath states were assigned utiity vaues derived from a arge UK survey using time trade-off (TTO) techniques. 107 EQ-5D utiity scores were reported by depression severity [defined by the Cinica Goba Impression of Severity scae (CGI-S)], and by cinica response (defined by MADRS scores) at foow-up. Patients were cassed as remitters, responders or non-responders. Remitters were patients with MADRS scores of 12, and responders were patients with a reduction in their MADRS score of 50% from baseine score; a other patients were cassified as non-responders. The difference in utiity scores among remitters, responders and non-responders was statisticay significant (p < 0.001) from baseine at 4 and at 8 weeks. In addition, the authors aso found that sex and age did not infuence utiity vaue: there was no statisticay significant difference in utiity scores between men and women and no difference by age. Overa the study was we described and of good quaity, and satisfies the quaity assessment criteria outined in the TSD by NICE's decision support unit. 101 Utiity data used in the mode There is ongoing debate around the optimum approach for measuring patient outcomes for use in economic evauation and decision making. 101 Areas of uncertainty incude which heath status instrument to use [EQ-5D, Heath Utiities Index (HUI) or Short Form questionnaire-6 Dimensions (SF-6D)]; which vauation technique [e.g. TTO, SG or visua anaogue scae (VAS)] to appy; and whose preferences (patients, cinicians or the genera pubic) to consider. However, NICE's methods guide 108 recommends that: (a) Heath status shoud be reported by the patients experiencing the condition. (b) The vaues paced on changes in heath shoud come from the UK genera popuation using a choice-based method. (c) EQ-5D is the preferred measure of HRQoL in aduts. In ine with these recommendations, one of the HSUV studies identified in the review of the QoL iterature, Sapin et a. (2004), 105 met the above criteria and was considered as the source of utiity for the mode. Utiity was reported by eve of treatment response (remission, response and non-response) and by severity of depression (mid, moderate and severe). The study was carried out in French patients but used UK genera pubic vauation scores. The mode assumes that at baseine everyone had the utiity of severe depression, as a patients in this mode have TRD. Overview of incuded quaity-of-ife studies The incusion criteria for this review were broad, and studies examining HRQoL in depression of any severity (not imited to TRD) were considered in addition to HSUV studies. This was to identify any issues that positivey or negativey impact on the QoL of depressed patients (these issues were considered ikey to be simiar in depressed patients, regardess of eve of treatment resistance). Tweve studies considering HRQoL were identified that covered a wide range of issues reevant to depression, incuding the impact on HRQoL of treatment (drug dosage and duration of treatment), physica iness, exercise and depression severity. Ten of the 12 studies were conducted in the USA. Of these 10 studies, four measured heath status using the SF-36, 109 112 three used the Quaity of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q), 67,69,113 one used both SF-36 and Q-LES-Q, 114 and one used the Word Heath Organization Quaity of Life Assessment (WHOQOL-BREF). 115 Finay, one study used both the SF-36 and the Late-Life Function and Disabiity Instrument (LLFDI). 116 A further study using the WHOQOL-BREF was carried out in Brazi 117 and a study using the Sickness Impact Profie (SIP) was carried out in the Netherands. 118 The patient popuations examined varied across studies; however, two studies 69,110 incuded patients with TRD as defined in this review (faied on at east two adequate antidepressant treatments in the current Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 61

ASSESSMENT OF COST-EFFECTIVENESS episode of depression). Two studies 109,117 assessed outcomes in patients with unipoar depression. The remaining studies 67,111 118 incuded patients meeting DSM-IV criteria for MDD. A range of interventions were considered in the identified studies: eight studies 69,110,111,113,115 118 considered antidepressant treatment (SSRI/SNRI/tricycic antidepressant); two studies 67,112 considered augmentation of antidepressant therapy with risperidone; one study 114 considered the augmentation of antidepressant therapy with physica exercise; and the remaining study 109 considered any therapeutic regimen that had been agreed by patients and their psychiatrist. A studies reported an improvement in QoL with treatment. There is some evidence that the impact of treatment on QoL is sustained and a onger duration of treatment resuts in further improvement. A 2-year maintenance study by Trivedi et a. 113 assessed the psychosocia outcomes in patients with recurrent MDD. The patients were responders to ER venafaxine during acute and continuation phase and at 1- and 2-year maintenance periods. The authors found that the onger a patient continues treatment, the higher their QoL. The study found that peope who responded to treatment and switched to pacebo had worse QoL than those who remained on treatment. The authors aso found that patients with greater severity of depression have worse QoL across mutipe domains and concuded that improvement in depressive symptoms woud be refected in assessment of functioning. Another study by Karp et a. 116 assessed the correation between disabiity depression severity and QoL in patients with or without depression. The study used the LLFDI, which is defined by the authors as a measure of instrumenta activity of daiy iving, persona roe and socia roe function. The authors found that there was a correation between disabiity and depression severity measured on the HAMD-17 scae and that antidepressant treatment improved functiona abiity. The study aso demonstrated that those patients who stayed onger on treatment continued to improve. In addition, a prospective muticentre observationa study by Dunner et a. 110 in peope with TRD who were severey depressed and receiving treatment as usua [any therapeutic regimen agreed to by the treating physician and the patient (i.e. drugs or ECT)] found that, after 12 months of treatment, 48% of the patients reported that their QoL had not changed. After 2 years of treatment, the proportion of patients reporting no change in QoL fe sighty to 42%. The proportion of patients reporting an improvement in QoL was 30% and 36% at 12 months and 24 months' foow-up, respectivey. Therefore, based on the resuts from the studies by Trivedi et a. 113 and Dunner et a. 110 it coud be inferred that patients shoud be kept on treatment as ong as possibe to maximise the HRQoL gains. There is, however, some evidence from reapse prevention studies in depression that indicates a higher rate of reapse in patients who cease therapy foowing response. 15 It may be that patients in Trivedi et a. 113 and Dunner et a. 110 who ceased therapy or switched to pacebo experienced a rebound effect that resuted in a ower QoL than in patients who remained on therapy. About haf of the studies reported that the physica domain of QoL for different instruments (SF-36, SIP, Q-LES-Q and WHOQOL-BREF) did not improve with treatment (antidepressants or antidepressant pus risperidone). 109,110,112,113,118 However, one study by Carta et a. 115 assessing physica activity as an adjunctive therapy in women with MDD found that the physica component of the WHOQOL-BREF improved significanty in women who did physica activity compared with women who did not. There were no differences in other domains (reationships, environment and psychoogica) between those who did physica activities and those who did not. These findings suggest that physica activity may be used as adjunctive therapy to antidepressants to improve the overa QoL of peope suffering from MDD. A study by Sma et a. 109 in edery patients (age > 60 years) with unipoar depression assessed the impact of physica iness on QoL. The indicators for physica iness were number of current chronic and historica inesses and these indicators were found to infuence QoL measures. QoL was measured using the SF-36, and depression eves were the same between patients with physica iness and those who did not have physica iness. The commony reported chronic inesses were joint diseases, cardiovascuar disease, aergies and gastrointestina disease. Frequenty reported historica inesses were surgica procedures, gastrointestina disorder, accidenta injuries and cardiovascuar disease. The authors concuded that current and previous 62 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 physica iness is associated with poor QoL, especiay physica and menta functioning in oder patients with depression. Therefore, based on the findings of the studies by Carta et a. 115 and Sma et a., 109 physica and psychoogica heath may be considered as reativey independent factors of QoL. In concusion, there is evidence to suggest that treatments for depression are effective in improving the QoL of patients, particuary if treatment is sustained. In addition, the QoL evidence base indicates that pharmacoogica treatment of depression does not resut in any improvement of physica outcomes, and physica therapies do not resut in any improvement of psychoogica symptoms. This, in turn, suggests that the physica and psychoogica domains of QoL are independent of each other in determining the overa QoL of a depressed patient. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 63

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 5 De novo economic anaysis Introduction As many as two-thirds of patients with major depression wi either not respond to, or have a suboptima response to, first-ine treatment with antidepressants. 16 After an inadequate response to at east one antidepressant treatment, current NICE depression guidance 16 recommends that: If a person with depression is informed about, and prepared to toerate, the increased side effect burden, consider combining or augmenting an antidepressant with ithium or an AAP, such as aripiprazoe, oanzapine, quetiapine or risperidone or another antidepressant, such as mirtazapine or mianserin. The focus of this report is the comparative cinica effectiveness and cost-effectiveness of augmentation of SSRI therapy with ithium or with an AAP in a TRD popuation. For the purposes of this report, TRD is defined as those patients with inadequate response to two or more adequate trias of antidepressant therapies, as specified in the NIHR HTA fina protoco. 24 To faciitate the effective and efficient aocation of heath-care resources, it is necessary to quantify and compare the economic benefits of the two key augmentation strategies of ithium and AAPs. A review of the cost-effectiveness iterature carried out as part of this report (see Chapter 4, Economic iterature review) did not identify any anayses comparing the two augmentation strategies in a TRD popuation. Therefore, a de novo economic anaysis was deveoped to estimate the costs, consequences and reative cost-effectiveness of each augmentation strategy over a 1-year time horizon. The mode was constructed in Microsoft Exce (Microsoft Corporation, Redmond, WA, USA) and used probabiistic anaysis of a cohort of 2000 TRD patients to estimate the expected cost-effectiveness of each strategy. Mode overview A hybrid economic mode was constructed to simuate the cinica and economic consequences of augmenting an SSRI with either ithium or an AAP in the treatment of TRD. The mode considered outcomes from the perspective of the NHS. This modeing approach was inspired by the mode deveoped by Simpson et a. 89 discussed in the review of the cost-effectiveness iterature in Chapter 4 (see Narrative review of incuded studies). The hybrid mode faciitates capturing the granuarity of the acute treatment phase whie simutaneousy accounting for patient progression within 1 year (i.e. discontinuation, reapse, remission and treatment response). It is ikey that each treatment considered wi have a differentia impact on costs and effects over a onger period of time than 1 year. However, as a resut of the paucity of ong-term foow-up data (the effectiveness data identified in the cinica review had foow-up of, at most, 8 weeks) and, in ine with pubished iterature, 86 the mode adopted a pragmatic time horizon of 1 year. Extrapoation beyond 1 year woud increase mode uncertainty. The economic mode consists of two distinct components: 1. decision tree 2. Markov mode. A decision tree was used to simuate treatment outcomes during the acute treatment phase, which, in accordance with avaiabe cinica data, is defined as 8 weeks of treatment. A patients wi be initiated on ithium or an AAP (as augmentation of SSRI therapy). The outcomes of the acute treatment phase (with or Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 65

DE NOVO ECONOMIC ANALYSIS without discontinuation) are response, non-response and remission. The schematic representation of the acute treatment phase of the mode is shown in Figure 13. Foowing on from the acute treatment phase, a patients are transitioned into the Markov component of the mode. The Markov component of the mode captures the maintenance phase, which incudes 6 months of maintenance therapy and 4 months of foow-up (where patients may remain on or cease therapy depending on the eve of response achieved). The heath states captured in the maintenance phase of the mode were response (with and without discontinuation), remission (with and without discontinuation), non-response (a patients in non-response are assumed to have discontinued therapy) and reapse (Figure 14). As a resut of the ack of ong-term comparative cinica data, transitions between heath states within the Markov component of the mode are assumed to be independent of treatment. Patients move between heath states for 10 months, in cyces of 2 months, chosen to represent the usua ength of a treatment course. The cyce ength was considered to be sma enough not to require a haf-cyce correction. The absence of a haf-cyce correction may resut in the underestimation of the amount of reapse, discontinuation and response experienced by patients. However, given that the Markov phase of the mode is treatment independent this is unikey to bias the resuts of the mode. To ensure the mode refects the management of unipoar depression in the UK, two cinica advisors (both of whom are practising psychiatrists with experience of managing TRD in either primary or secondary care) Response Continue Non-response Remission TRD patients Response Discontinue Non-response Remission FIGURE 13 The decision tree component of the mode for the cost-effectiveness of augmentation of SSRIs with ithium or an AAP during the acute treatment of TRD. Response Non-response Response discontinued Remission Reapse Remission discontinued FIGURE 14 The Markov component of the mode for the cost-effectiveness of augmentation of SSRIs with ithium or an AAP during maintenance-phase treatment in TRD. 66 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 were consuted to vaidate mode assumptions throughout the mode deveopment process. Information from experts was obtained through teeconferences and standardised questionnaires. The mode considers direct treatment costs incurred by the NHS. These incude costs of the pharmacoogica treatments, HCPs, hospitaisation and monitoring. Societa costs were not incuded within the mode. A costs used were 2011 costs: no discounting was appied because the time horizon was no onger than 1 year. Effectiveness was measured in terms of QALYs as per the NICE reference case. 108 The mode did not consider dose titration/escaation as, foowing advice from cinica experts, it was assumed that patients were on maximum toerated doses. In addition, cinica experts advised that adverse events are unikey to be treated; patients are more ikey to discontinue their therapy. Therefore, the mode does not expicity consider the costs of treating adverse effects or the disutiities associated with adverse effects. Mode structure Patients entered the mode foowing initiation of augmentation therapy (ithium or an AAP) and progressed through the decision tree component of the mode, representing the acute phase of treatment. Foowing initiation, patients were exposed to the risk of discontinuation. Discontinuation can occur for any reason; however, ack of efficacy and treatment-reated adverse effects are frequenty reported as the main causes of discontinuation. 119 Expert opinion suggests that discontinuation due to side effects is usuay instantaneous, whereas discontinuation owing to ack of efficacy may take pace between 2 and 6 weeks after initiation of augmentation treatment. For simpicity, and owing to the way discontinuation was reported in the studies, the mode does not distinguish between discontinuation due to ack of efficacy and discontinuation due to treatment-reated adverse effects. Those who discontinue their treatment are expected to do so at week 4 in accordance with discontinuation owing to ack of efficacy; this is a simpifying assumption agreed by cinica experts. Foowing 8 weeks of treatment, patients entered one of the foowing heath states: remission, response or non-response. The review of the evidence avaiabe on cinica effectiveness (see Chapter 3, Quaity assessment) indicated that symptom severity was predominanty assessed using the MADRS. Therefore, in the economic mode, the heath states of response and remission were defined using this scae. Response, which can be thought of as a period during which an improvement of sufficient magnitude is observed such that the individua is no onger fuy symptomatic, 120 was defined as the proportion of patients who, at 8 weeks, had a reduction in their MADRS score of at east 50% (from baseine). Remission is a period during which an improvement of sufficient magnitude is observed that the individua is asymptomatic. 120 Remission was defined differenty in the identified trias, with some trias defining remission as a tota MADRS score of 8 for two consecutive visits and others defining it as a tota MADRS score of 10 at the end of the study. Based on discussions with cinica experts, remission in the mode is defined as a MADRS tota score of 10 at the end of the acute treatment phase. Patients who competed acute therapy but did not enter the heath states of remission or response moved into the heath state of non-response, defined as those patients not achieving a 50% reduction in MADRS score at the end of the acute treatment phase. Of the patients who do not compete acute therapy owing to discontinuation, a proportion is assumed to cinicay improve. Therefore, amongst peope who discontinue, some wi respond and some wi enter remission (cinica expert opinion). Simiar to patients who compete acute therapy, patients who discontinue and do not enter the heath states of remission or response enter the heath state of non-response. On competion of the acute treatment phase, a patients, irrespective of whether they have responded or discontinued their acute therapy, are moved into the maintenance phase of the mode. The maintenance phase is assessed using the Markov component of the mode and incudes 6 months of maintenance therapy and 4 months of foow-up (on which patients may remain or cease treatment, depending on the eve of response achieved). Therefore, depending on their heath state at the end of the acute treatment phase, Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 67

DE NOVO ECONOMIC ANALYSIS patients enter the Markov component of the mode in one of five heath states: remission, remission discontinue, response, response discontinue or non-response. In addition to these heath states, the Markov mode incudes the heath state of reapse. A patient in the fuy symptomatic heath state (MADRS score of > 15) after having achieved remission or response is considered to have reapsed. Thus, a reapse represents the return of the symptoms of a sti ongoing but symptomaticay suppressed episode. 120 Within the Markov mode, patients may remain in state or move to other heath states, depending on the risks associated with their current heath state. Patients who are in remission are exposed to the risk of reapse or discontinuation (of their maintenance therapy). Patients in the response heath state may transition to the heath states of remission or response discontinuation. Patients in the heath state of remission discontinuation may reapse, whereas those in response discontinuation may transition into reapse or remission discontinuation. Patients who have not responded to their acute treatment (irrespective of whether or not an acute therapy course was competed) are assumed to receive no further benefit from augmentation therapy and are assigned to a standard package of care (fu detais of the standard package of care are given beow; see Costs associated with each heath state). However, a proportion of non-responders are assumed to cinicay improve, i.e. some wi respond and some wi enter remission (cinica expert opinion). Therefore, patients in the heath state of non-response may move to the heath states of remission discontinuation or response discontinuation. Simiar to the state of non-response, peope who reapse are assumed to receive no further benefit from augmentation therapy and are assigned a standard package of care. This simpifying assumption was used in pace of expicity modeing foow-on patient care pathways, as these are diverse and there is a paucity of data to inform them. However, unike the heath state of non-response, the heath state of reapse is an absorbing state, as patients entering this heath state do not experience any further transitions and remain in state for the duration of the mode. In contrast to non-response, reapse is assumed to be the return of symptoms that had previousy been suppressed. Therefore, the probabiity of entering response or remission foowing reapse is ikey to be consideraby different from the probabiity of entering response or remission from a state of non-response. No data were identified in the iterature to inform these transitions. However, the Markov component of the mode uses transition probabiities that are independent of treatment and it is, therefore, unikey that assuming reapse is an absorbing state woud bias the mode over the short time horizon considered. Effectiveness data Acute treatment phase Patient outcomes foowing the decision tree component of the mode are dictated by the cinica efficacy of and the proportion of patients competing their acute augmentation therapy. In this section, discussion focuses on the: acute efficacy associated with each augmentation therapy rate of discontinuation associated with each therapy ikeihood of cinica improvement of symptoms after discontinuation from acute therapy. Acute efficacy The acute efficacy of augmentation therapy with ithium or with an AAP used in the mode was based on data derived from a MTC carried out as part of this review (see Chapter 3, Assessment of effectiveness). A MTC was performed because no head-to-head studies were identified comparing SSRI augmentation with ithium and augmentation with a AAP. SSRI was used as the baseine treatment for the comparison. The studies incuded in the MTC used fuoxetine as the SSRI and oanzapine as the AAP, as no other studies were identified in a reevant popuation (see Chapter 3, Assessment of effectiveness). Consequenty, there was 68 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 an absence of evidence to inform comparisons of different SSRI and AAP treatment strategies in a cost-effectiveness anaysis. Therefore, based on the absence of evidence to suggest different efficacies associated with individua SSRIs or AAPs, a cass effect was assumed for the SSRIs and the AAPs in the cost-effectiveness anayses. The robustness of this assumption was investigated in a sensitivity anaysis that used data from a second MTC carried out as part of the cinica effectiveness anaysis (see Chapter 3, Assessment of effectiveness). The second MTC incuded trias in which ithium, oanzapine, aripiprazoe or quetiapine were used to augment a patient popuation treated with a mixture of SSRI or venafaxine therapy. This sensitivity anaysis does not expicity reax the assumption of cass effect (as no data were avaiabe). However, it does provide an estimate of the extent to which the cost-effectiveness resuts are sensitive to the use of additiona data to inform of the cass effect of AAP augmentation. The probabiities of remission and response associated with each augmentation strategy were required to inform the decision tree component of the mode, representing the acute phase of treatment. From the MTC, data were avaiabe for the outcome of response for both augmentation therapies. However, data on remission rate were avaiabe ony for augmentation with an AAP. As a consequence of the imited data avaiabe to inform the acute treatment efficacy, the decision was taken to use an aternative approach to generate the required probabiities. The approach used invoved samping the treatment effect of each augmentation strategy (from a distribution of possibe effects) and cacuating the proportion of patients (in a cohort of 1000 for each treatment arm) that woud achieve remission or response. The detais of this aternative approach are described here and summarised in Figure 15. Step 1 Baseine MADRS score = 30.0 (before continuation of SSRI therapy or augmentation of SSRI therapy) Step 2 Norma distribution fitted to mean change in MADRS score from baseine foowing treatment with SSRI pus ithium (mean = 12.58, SE = 10.42) Norma distribution fitted to mean change in MADRS score from baseine foowing treatment with SSRI pus an AAP (mean = 11.22, SE = 9.65) Step 3 Mean change in score of augmentation with ithium samped for 1000 patients Mean change in score of augmentation with AAPs samped for 1000 patients Step 4 Fina MADRS score foowing augmentation with ithium cacuated for 1000 patients Fina MADRS score foowing augmentation with AAPs cacuated for 1000 patients Step 5 Proportion of patients in remission, response and non-response foowing augmentation therapy with ithium cacuated based on fina MADRS score Proportion of patients in remission, response and non-response foowing augmentation therapy with AAPs cacuated based on fina MADRS score Step 6 Probabiity of remission and probabiity of response for patients augmented with ithium cacuated Probabiity of remission and probabiity of response for patients augmented with an AAP cacuated FIGURE 15 Cacuating the treatment-specific probabiity of remission and response for the decision tree component of the mode, representing the acute phase of treatment. SE, standard error. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 69

DE NOVO ECONOMIC ANALYSIS Cacuating the treatment-specific probabiity of remission and response First, a baseine MADRS score of 30.0, which was considered to be representative of the TRD popuation (as defined in the scope of this review), was chosen and entered into the mode. This baseine MADRS score was assumed to appy to a patients entering the mode (i.e. before initiation of augmentation therapy). The average MADRS score of two studies pooed in an anaysis by Thase et a. 53 identified as part of the cinica review was seected to provide the baseine MADRS score: the popuations in these studies were considered to be the most representative of a TRD popuation compared with other studies identified as part of the cinica review (for more detais of the cinica review see Chapter 3, Assessment of effectiveness). Second, the absoute changes in MADRS score from baseine with SSRI + ithium and with SSRI + AAP were estimated from the MTC carried out as part of the cinica effectiveness anaysis (see Chapter 3, Assessment of effectiveness). Lithium was associated with an absoute mean change in MADRS score of 12.58 [standard error (SE) 10.42] and the AAPs were associated with an absoute mean change in MADRS score of 11.22 (SE 9.65). The mean and SE of each treatment effect were used to fit a norma distribution from which the change in MADRS score associated with each augmentation therapy coud be samped. A patients in the cohort (1000 patients for each treatment arm) were then assigned a fina MADRS score foowing augmentation of SSRI therapy with ithium or with an AAP (fina MADRS score = baseine MADRS score + samped change in MADRS score with augmentation treatment). In cinica practice, the baseine MADRS score of the patient popuation woud be variabe. However, for the purposes of impementing the samping methodoogy, the baseine MADRS score was assumed to be fixed, as the absoute treatment effect used to inform the sampe distributions is not independent of baseine MADRS score. No data were avaiabe to account for the correation between baseine MADRS score and the absoute effect of each augmentation treatment. Therefore, the decision was made to assume a fixed baseine MADRS score to avoid an inappropriate increase of the uncertainty associated with the acute treatment efficacy. The fina MADRS score of each patient foowing augmentation therapy was assessed against the definitions of response and remission outined in Chapter 3 (see Assessment of effectiveness). Therefore, patients who achieved a fina MADRS score of < 10 were cassified as remitters and patients who achieved a 50% or more reduction in their MADRS score from baseine (but did not achieve remission) were cassified as responders. Patients who did not meet remission or the response criterion were cassified as non-responders, defined as those patients not achieving a 50% reduction in MADRS score at the end of the acute treatment phase. The proportion of patients achieving remission, response and non-response for each augmentation therapy were cacuated. These proportions were used to inform the probabiity of entering remission, response and non-response in the decision tree component of the mode after 8 weeks of acute therapy. Vaidation of acute efficacy estimation The proportions of patients estimated to achieve response or remission by the samping method described above were compared with the rates of response reported in the iterature. Pooed anaysis of two studies reported by Thase et a. 53 (seected to provide the baseine MADRS score used in the mode) and the singe ithium tria 59 identified for incusion in the MTC were seected to provide this comparison. Tabe 10 summarises the proportion of patients estimated (based on 2000 runs of 1000 sampes for each treatment arm) to achieve remission and response and those reported in the iterature. The mode estimates of response with AAP augmentation therapy and ithium augmentation therapy are approximatey 5% ower than those reported in the iterature. 53,59 However, the baseine MADRS score of patients augmented with ithium in Katona et a. 59 is 26.06 (SD 4.93), rather than 30.0, as assumed in the mode. In addition, the patient popuation of Katona et a. 59 had faied ony one antidepressant treatment in their current episode of depression (for fu detais see Chapter 3, Assessment of effectiveness) and may be ess resistant to treatment than the modeed patient popuation. Therefore, the difference in response rate between the mode and Katona et a. 59 may be as a resut of the difference in baseine depression severity. Furthermore, the efficacy estimate for augmentation with an AAP, used to inform the mode, is based 70 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 10 Comparison between response and remission probabiities estimated in the mode and those reported in the iterature Outcome Treatment Mode estimate (%) Tria estimate (%) Difference (mode estimate tria estimate): (%) Tria reference Response a Lithium 43.0 47.1 4.1 Katona et a. (1995) 59 AAPs 35.0 40.4 5.4 Thase et a. (2007) 53 Remission Lithium 24.3 NR N/A N/A AAPs 18.7 27.3 8.65 Thase et a. (2007) 53 Non-response b Lithium 57.1 N/A N/A N/A AAPs 65.0 N/A N/A N/A N/A, not appicabe; NR, not reported. a Incuding remission. b Probabiity of non-response was cacuated as 1 (the probabiity of response pus probabiity of remission). on a MTC which incuded data from the two studies by Thase et a. 53 However, it is important to note that MD in MADRS score reported in Thase et a. 53 was much arger than that reported in the other trias incuded in the MTC. Therefore, estimates of response derived from the MTC may be expected to be ower than those reported in Thase et a. 53 The difference may be due to samping error or heterogeneity in the studies incuded in the MTC. Acute therapy discontinuation rates Discontinuation may occur for a variety of reasons, athough patients typicay discontinue their acute treatment because of ack of efficacy or drug-reated side effects. However, as discussed above (see Mode structure), the mode does not distinguish between reasons for discontinuation, and discontinuation is assumed to occur at week 4 of acute therapy. Rates of discontinuation (for any reason) were taken from the same MTC that provided acute efficacy estimates, carried out as part of the cinica efficacy anayses described in Chapter 3 (see Assessment of effectiveness). Augmentation with an AAP was associated with a higher rate of discontinuation compared with no augmentation (OR 1.27; 95% CrI 0.90 to 1.75). However, augmentation with ithium was associated with a ower rate of discontinuation compared with no augmentation (OR 0.92; 95% CrI 0.13 to 3.32), abeit with wider 95% CrIs. These estimates were based on data from five AAP trias assessing a tota of 1017 patients and a singe ithium tria in 33 patients (reported in Chapter 3). As a consequence of the imited data avaiabe, the estimate of the OR of discontinuation with ithium augmentation is ikey to be ess reiabe than that of AAP augmentation. This is demonstrated by the wider 95% CrIs associated with ithium augmentation. To enabe the use of the ORs estimated from the MTC in the economic mode, the ORs were converted into reative risks (RRs) using the foowing formua: OR RR¼ ð1 r b Þþðr b ORÞ ð1þ where r b = baseine risk of discontinuation (i.e. that associated with SSRI therapy aone) reported in the pooed anaysis by Thase et a. 53 Cinica improvement foowing discontinuation from acute therapy The mode assumed that patients discontinuing their acute treatment at week 4 woud not receive the benefit of treatment experienced by patients who compete 8 weeks of therapy. However, no studies were identified that reported response and remission rates for patients who did not compete their acute Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 71

DE NOVO ECONOMIC ANALYSIS treatment. Therefore, cinica experts were consuted to estimate, based on their experience, the ikeihood of cinica improvement in patients who have discontinued from therapy. Experts agreed that overa approximatey 5% and 15% of non-competers woud remit and respond, respectivey. These estimates were assumed to represent the annua probabiity of cinica improvement and standard formuae 121 were used to convert them into 2-monthy probabiities for use in the mode. r ¼ ½nð1 pþš=t p 2 - ¼1 exp r 2- month 6 ð2þ ð3þ where p = annua probabiity, r = instantaneous rate and p 2-month = 2-month probabiity. Tabe 11 summarises the probabiities used to inform the decision tree component of the mode. Maintenance phase The maintenance phase of the mode incudes 6 months of maintenance treatment and 4 months of foow-up (where patients may remain on or cease therapy depending on the eve of response achieved). An absence of comparative ong-term data for maintenance augmentation treatment ed to the decision to assume that the maintenance phase of the mode is predominanty treatment independent. Therefore, the parameters used in the Markov component of the mode, representing the maintenance phase, are non-drug specific. Tabe 12 summarises the probabiity associated with each possibe transition for patients in the Markov mode. Response to remission A ongitudina study conducted in the UK by Fekadu et a. 122 foowed 118 patients with TRD who had been discharged from speciaist inpatient care (median foow-up of 3 years). Patient outcomes were reported by post-treatment status at discharge: remission, partia remission or sti in episode. The authors reported that patients had received individuaised treatment packages consisting of mainy pharmacotherapy TABLE 11 A parameters used in the decision tree component Treatment-specific probabiity (%) Parameter Lithium AAP Sources Remission a 24.3 18.7 Samping based on resuts of MTC Response a 18.7 16.3 Non-response a 57.0 65.0 Discontinuation b 18.1 23.4 Thase et a. (2007) 53 MTC Remission foowing discontinuation 3 3 Expert cinica opinion Response foowing discontinuation 1 1 Non-response foowing discontinuation 96 96 a Cacuated from the PSA (1000 patients samped 2000 times and the proportions of remission, response and non-response averaged). b The rate of discontinuation associated with SSRI therapy reported in Thase et a. 53 is adjusted by the RRs of 0.93 and 1.21 for ithium and AAPs, respectivey, estimated from the MTC. 72 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 12 Transition probabiities (per 2-month cyce) used in the Markov mode Transition to Transition from Remission (%) Remission discontinue (%) Response (%) Response discontinue (%) Nonresponse (%) Reapse (%) Remission 70.3 a 20.8 b 0 c 0 c 0 c 8.9 d Response 3.8 e 0 c 54.3 a 20.8 b 0 c 21.2 d Remission discontinue 0 c 91.1 a 0 c 0 c 0 c 8.9 d Response discontinue 0 c 3.8 d 0 c 75.1 a 0 c 21.2 d Non-response 0 c 0.9 f 0 c 2.7 f 96.4 a 0 c Reapse 0 c 0 c 0 c 0 c 0 c 1 c a Figure = 1 sum of a other transition probabiities from this heath state. b Assumed to be equa to the average of acute therapy discontinuation. c Assumption. d STAR*D tria. 99 e Fekadu et a. 122 f Expert opinion. (mosty using medication combination), physica therapy and psychoogica therapy as indicated. The outcomes were assessed using the seven-point-scae Longitudina Interva Foow-up Evauation (LIFE) chart, which is a foow-up evauation scae that aows retrospective rating of a patient's symptomatic state. 122 The scores range from an asymptomatic state (score 1) to severe episode (score 7). Remission was defined as a score of 1 or 2, a score of 3 or 4 represented partia remission, and scores 5 7 were defined as a depressive episode. The authors reported that for those that were discharged in partia remission, 50% achieved remission over the 3-year foow-up period. This 3-year probabiity was converted into a 2-month probabiity using standard formuae. 121 This probabiity was then used to represent the probabiity of improvement to remission for patients in response and for improvement to remission discontinuation for patients in response discontinuation. As acknowedged by the authors, the generaisabiity of the findings from Fekadu et a. 122 may be imited because the patients incuded in the study received intensive speciaist inpatient care at a speciaist tertiary centre. Furthermore, patients who remain on therapy may be more ikey to improve to the eve required for remission than patients who have discontinued. However, in the absence of data to inform these different transitions in an appropriate popuation, data from Fekadu et a. 122 were assumed to be appicabe to the probabiity of remitting in the modeed popuation regardess of treatment status. This assumption is tested in the one-way sensitivity anaysis (OWSA) (see Resuts). Reapse Reapse rates were based on the 1-year foow-up resuts reported for eve 3 patients in the STAR*D tria. 99 The study evauated feasibe treatment strategies to improve cinica outcomes for rea-word patients with TRD. Patients in eve 3 had faied to benefit from at east two antidepressant treatment exposures. Reapse rates in this foow-up study were reported by treatment status, i.e. whether or not the patients had remitted prior to foow-up. Fewer patients who were in remission at the beginning of the foow-up period reapsed than those who were not in remission (42.9% vs. 76%). These annua probabiities were converted using standard formuae 121 into 2-month probabiities for use in the mode. The probabiity of reapse was assumed to be the same regardess of treatment status. Therefore, patients in response discontinue and remission discontinue were assigned the same probabiity of reapse as patients in response and remission, respectivey. This assumption is tested in the OWSA (see Resuts). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 73

DE NOVO ECONOMIC ANALYSIS Non-response to remission discontinuation or response discontinuation No study was identified to inform the probabiity of moving from a state of non-response to response or remission. However, expert cinica opinion sought during the construction of the acute treatment phase of the mode was that approximatey 5% and 15% of non-competers woud remit and respond, respectivey (see Acute treatment phase, above). Lack of efficacy is acknowedged to be a main reason for discontinuation. Therefore, it was considered reasonabe to assume that the ikeihood of entering response or remission foowing discontinuation from acute therapy is equivaent to the ikeihood of remitting or responding for patients in non-response. The estimates obtained from expert cinica opinion (5% and 15% for remission and response, respectivey) were assumed to represent the annua probabiity of remission or response and were converted into 2-monthy probabiities using standard formuae. 121 Discontinuation No studies were identified reporting discontinuation rates of patients on maintenance augmentation therapy. Therefore, discontinuation rates obtained for ithium augmentation and augmentation with an AAP from the MTC (18.1% and 23.4%, respectivey) were averaged and assumed to appy in the maintenance phase, which resuted in a discontinuation rate of 20.8%. Costs The mode is constructed from the perspective of the NHS and considers the costs of drug treatment, HCPs (e.g. GPs, and CMHTs), hospitaisation and treatment-reated monitoring (e.g. aboratory tests). Socia care costs, such as those reated to residentia care, were not incuded in the anaysis because these costs were beieved not to vary widey among the interventions assessed. Other societa costs, incuding socia benefit payments, costs associated with ega system services and productivity osses of patients and carers were not estimated as they were beyond the scope of the anaysis. Drug costs As discussed above (see Acute treatment phase), a cass effect was assumed for the SSRIs and AAPs. Therefore, the cost of SSRI used in the mode is a weighted average of the commony prescribed SSRIs. The weights are based on cinica opinion rather than prescription cost anaysis, as prescription cost anaysis encompasses depression of any severity rather than focusing on TRD. The four most commony prescribed SSRIs are citaopram (20%), escitaopram (20%), fuoxetine (30%) and sertraine (30%). However, cinica experts acknowedge that citaopram use is decreasing because of concerns over side effects of QT interva proongation, particuary with respect to augmentation. The most commony prescribed AAPs are aripiprazoe, oanzapine, quetiapine and risperidone; experts estimated that the proportion of use for each of these woud be 30%, 30%, 20% and 20%, respectivey. Anaogous to the assumptions made for SSRIs, a weighted average cost was cacuated for AAPs. A drug costs were taken from the BNF 63. 19 Tabe 13 summarises the weighted drug costs used in the mode. The maximum icensed dose of each SSRI is used in the cost cacuations, as it is assumed that patients eigibe for augmentation wi be receiving the maximum dose of their current SSRI therapy. The dose assumed in the AAP cost cacuation is the usua maintenance dose stated in the BNF. 19 Lithium was assumed to be given in tabet form as the cost of tabets was ower than that of ithium injections (Priade, 800 mg per day). A threshod anaysis around the cost of AAPs is carried out as part of the sensitivity anaysis (see Sensitivity anaysis, beow). The cost of fuoxetine was aso varied in sensitivity anaysis using 3 20-mg tabets, which cost 4 per month, as the base case assumed the use of 1 60-mg fuoxetine tabet, which cost 53 per month. Heath-care professiona costs For the purpose of costing the patient pathway, the anaysis assumes that during the acute phase of the disease a proportion of patients wi be managed in the community and the remainder wi be managed in hospita. For those patients managed in the community, a fraction wi be seen by CRHTTs, who provide intensive home-based support, whereas the remainder wi receive usua care from their GP and CMHT. Cinica experts estimate that 70% of patients with TRD wi receive usua care and 20% wi be seen by 74 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 13 Commony prescribed drugs and the weighted costs used in the mode (SSRIs, AAPs and ithium) Drug SSRIs No. in pack Cost/pack ( ) Cost/day ( ) Cost/month ( ) Weighted drug use (%) Weighted costs used in the mode ( ) Acute Maintenance Citaopram 40 mg Sertraine 100 mg 28 1.37 0.05 1 20 0.60 1.79 28 1.80 0.13 4 30 2.35 7.04 Fuoxetine 60 mg 30 52.54 1.75 53 30 31.96 95.89 Escitaopram 10 mg 28 25.20 0.90 27 20 10.95 32.85 Tota cost ( ) 45.85 137.56 AAPs Quetiapine 300 mg 60 170.00 2.83 86 30 51.71 155.13 Oanzapine 5 mg 28 43.70 1.56 47 20 18.99 56.97 Aripiprazoe 10 mg 28 95.74 3.42 104 30 62.40 187.21 Risperidone 3 mg 60 2.71 0.05 1 20 0.55 1.65 Tota cost ( ) 133.65 400.95 Lithium Priade 400 mg 100 3.35 0.07 2 100 4.08 12.23 the CRHHT. The remaining 10% of patients woud receive inpatient care. These estimates are in ine with those cited in the NICE Depression Guideine, 16 as we as the King's Fund Report. 22 Foowing acute treatment, a patients are assumed to receive care in the community, with the amount of care received depending on the eve of response achieved. Care in the community Usua care Usua care in this mode refers to a package of care administered by the GP and the CMHT; patients receiving usua care are assumed to have no contact with a speciaist psychiatrist. During the acute phase of treatment, it is assumed that patients receiving usua care wi be seen by their GP and CMHT twice in the first month and once in the second month. During the maintenance phase of the mode, patients in remission or response are assumed to see their GP and CMHT ess frequenty. Based on expert cinica opinion, patients who remit are assumed to have contact with their GP and CMHT five and three times, respectivey, over the whoe maintenance phase. Patients who respond but do not achieve remission are assumed to have five visits with their GP and CMHT over the whoe maintenance phase. Crisis Resoution and Home Treatment Teams Crisis Resoution and Home Treatment Teams wi visit about 20% of patients with TRD. CRHTTs work with patients in their own homes who, without this support, woud need to be admitted to hospita. This ensures that patients recover at home in the environment in which they wi have to function in the community rather than in an inpatient environment that bears itte resembance to norma iving conditions. 123 Foowing a crisis assessment, a care pan is agreed with the individua patient, which aims to meet their current menta heath needs. For the purpose of this mode, cinica experts suggested that during Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 75

DE NOVO ECONOMIC ANALYSIS the acute treatment phase patients wi have contact with a CRHTT 16 times, i.e. he/she wi be seen twice a week during the acute treatment phase. Foowing the acute treatment phase, patients who have not responded to treatment are assumed to be cared for by a CRHTT (one visit every 2 months pus one visit every 2 months to their GP). This assumption was made to represent the increased eve (and cost) of care that is ikey to be required by patients who do not respond to treatment. In addition, patients who reapse during the maintenance phase are assumed to be cared for by a CRHTT (one visit every 2 months pus one visit every 2 months to their GP). The costs of care by a CMHT and a CRHTT were taken from the NHS reference costs for 2010 11, 124 and the cost of a visit to the GP was taken from the Unit Costs for Heath and Socia Care 2011. 125 A community care costs are summarised in Tabe 14. Inpatient care costs The mode assumed that 10% of patients with TRD woud be managed as inpatients. Data from the Hospita Episode Statistics (HES) 126 showed that the mean ength of hospitaisation during acute depression for the year 2010 11 [F30 F39 Mood (affective) disorders] was 40 days (median 19 days). Therefore, the mode assumed an inpatient ength of stay of 40 days. The cost of inpatient menta heath care was taken from the NHS reference costs for 2010 11. 124 The mean inpatient cost per day was cacuated as a weighted average of costs for the foowing individua components: adut intensive care, acute care, rehabiitation and care of the edery (codes MHIPA1 3 and MHIPE1). As summarised in Tabe 15, the cost of each component was weighted by the reported eve of activity for that component to cacuate the unit cost of inpatient care. After discharge from hospita, patients were assumed to receive usua care in the community by their GP and CMHT. Monitoring costs During initiation and maintenance of treatment, patients incurred the aboratory costs of bood testing required for monitoring purposes, depending on the type of ong-term medication they received. The type and frequency of aboratory tests incuded in the mode were provided by two cinica experts and checked for consistency with pubished cinica guideines. 16,123 TABLE 14 Community care costs used in the mode Average unit cost per visit Lower quartie Upper quartie HCPs used in the mode ( ) unit cost ( ) unit cost ( ) CRHTT 179.00 136 207 CMHT 136.00 109 160 GP 36.00 30 40 TABLE 15 Inpatient costs used in the mode Currency description (NHS reference cost code) Activity Nationa average costs ( ) Weight Weighted mean costs ( ) Weighted ower quartie ( ) Weighted upper quartie ( ) Adut: Intensive Care (MHIPA1) 201,557 613 0.031566 19 16 21 Adut: Acute Care (MHIPA2) 2,994,811 304 0.469019 143 131 150 Adut: Rehabiitation (MHIPA3) 886,705 274 0.138867 38 31 44 Edery (MHIPE1) 2,156,019 310 0.337656 105 94 116 Tota weighted costs ( ) 312 278 339 76 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Patients receiving ithium have their ithium pasma eves monitored twice a month during the acute treatment phase then at 3-monthy intervas during maintenance treatment. Rena function tests are performed once at initiation and subsequenty at 3-monthy intervas for the duration of treatment. Thyroid function tests are carried out once at initiation and biannuay thereafter. Patients wi aso have eectrocardiography, body mass index (BMI) and weight measured, and a fu bood count at east once a year. Patients receiving an AAP have a fu bood count, urea and eectroyte eves measured, and a iver function test once a year. Fasting bood gucose and ipids profie are measured once at initiation and biannuay thereafter. Weight shoud be measured at east once a month in the acute phase and then at 3-monthy intervas for the remaining duration of treatment. The costs of aboratory tests and monitoring were taken from the NHS reference costs and iterature. 124 Tabes 16 and 17 summarise the frequency and type of test carried out during the acute and maintenance phases of treatment for those patients receiving augmentation with ithium or an AAP, respectivey. Costs associated with each heath state Acute treatment phase The medication cost incurred by patients in the acute treatment phase depends on whether a patient competes their acute therapy or discontinues. As discussed in Mode structure, above, patients who discontinue their acute treatment are assumed to do so after 4 weeks of therapy. Therefore, those patients who do discontinue wi incur the cost of 4 weeks rather than 8 weeks of acute treatment. However, TABLE 16 Frequency of tests and tota costs for patients augmented with ithium during the acute and maintenance phases Laboratory costs Unit cost ( ) Source Acute treatment phase Maintenance treatment phase Foow-up phase (if treatment is continued) Comment BMI 6.17 Meads et a. (2008) 127 1 0 0 Once a year ECG 34.94 NHS ref costs 2010 11 124 1 0 0 Once a year Fu bood count 2.35 a Bipoar 1 0 0 Once a year guideines 123 egfr 1.03 a Bipoar guideines 123 1 2 1 Once during initiation, and then at 3-monthy intervas for the duration of treatment Creatinine (to monitor rena function) Serum ithium concentration (ithium pasma eves) Thyroid function 1.03 2.82 16.08 a Bipoar guideines 123 1 2 1 Once during initiation, and then at 3-monthy intervas for the duration of treatment a Bipoar guideines 123 4 2 1 Once a month during initiation then at 3-monthy intervas for the duration of treatment a Bipoar 1 1 1 Once during initiation and guideines 123 then biannuay Cost appied in the mode ( ) 72.85 25.82 20.95 egfr, estimated gomeruar fitration rate. a Infated to 2011 prices using the Hospita and Community Heath Services infation indices Curtis 2010. 125 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 77

DE NOVO ECONOMIC ANALYSIS TABLE 17 Frequency of tests and tota costs for patients augmented with an AAP during the acute and maintenance phases Laboratory costs Unit cost ( ) Source Acute treatment phase Maintenance treatment phase Foow-up phase (if treatment is continued) Comment 2 2 1 Once a month during BMI 6.17 Meads 2008 127 initiation then at 3-monthy intervas for the duration of treatment ECG 34.94 NHS reference costs 2010 11 124 1 0 0 Once a year Fu bood count 2.35 Bipoar guideines a 1 0 0 Once a year egfr 1.03 Bipoar guideines a 1 0 0 Once a year Creatinine (to monitor rena function) 1.03 Bipoar guideines a 1 0 0 Once a year Gucose test 1.03 Bipoar 1 1 1 Once during initiation guideines a and then biannuay Lipid profie test 2.21 Bipoar guideines a 1 1 1 Once during initiation and then biannuay Cost appied in the mode ( ) 54.91 18.80 9.40 egfr, estimated gomeruar fitration rate. a Infated to 2011 prices using the Hospita and Community Heath Services infation indices, Curtis 2010. 125 monitoring costs in the acute phase are assumed to remain the same regardess of treatment status. Furthermore, a patients in the acute treatment phase wi receive the same cost of acute patient care regardess of which heath state they enter. As discussed in Heath-care professiona costs, above, some patients receive care in the community whie others are cared for as inpatients. The cost of patient care in the acute treatment phase is cacuated as a weighted average of community and inpatient care (Tabe 18). Tabe 19 summarises the tota cost appied to patients in the acute treatment phase. Maintenance phase The costs accrued in the maintenance phase of the mode depend on a patient's heath state. Tabe 20 summarises the costs associated with each heath state in the maintenance phase. As a simpification, the cost per cyce is cacuated as a proportion of the tota cost had the patient remained in state throughout the maintenance phase. For exampe, the cost per cyce of patients in remission is cacuated as one-fifth of the tota cost of entering the maintenance phase in remission and remaining in remission for the duration of the mode. This assumption was considered reasonabe, as the costs associated with HCPs, drugs and monitoring are ikey to be eveny distributed across the 10 months of the maintenance phase. Costs for patients who remit Patients who remit are assumed to be under the care of HCPs for a tota of 6 months in the maintenance phase. At the end of maintenance treatment, patients in remission are assumed to be discharged from the care of their CMHT and wi subsequenty be cared for by their GP aone. For the remaining 4 months of the mode, foowing the competion of maintenance therapy, remitters are assumed to visit their GP once every 2 months for a genera check-up. Therefore, the costs accrued by a patient in remission during 78 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 18 Cost of patient care appied in the acute treatment phase (HCPs) Setting of care Proportion of patients (%) Cost component Cost ( ) Resource use during acute treatment phase Tota cost ( ) Community (CRHTT) 20 CRHTT visit 179 16 visits (16 179) = 2864 Community (usua care) 70 GP visit 36 2 visits (2 36) + (2 136) = CMHT visit 136 2 visits 344 Inpatient 10 Menta heath inpatient cost 312 40 days (40 311.96) + 136 + 36 = 12,650.40 GP visit 36 1 visit CMHT visit 136 1 visit Overa weighted cost ( ) (0.2 2864)+(0.7 344)+(0.1 12,650.4)=2078.64 TABLE 19 Tota costs appied to patients in the acute treatment phase Cost ( ) Tota cost ( ) Treatment status Cost component Lithium AAP Lithium AAP Competed acute therapy Eight weeks of acute treatment 49.93 179.50 2201.42 2313.05 Acute monitoring costs 72.85 54.91 Eight weeks of acute patient care 2078.64 2078.64 Discontinued acute therapy Four weeks of acute treatment 24.96 89.75 2176.46 2223.30 Acute monitoring costs 72.85 54.91 Eight weeks of acute patient care 2078.64 2078.64 the Markov component of the mode woud incude 6 months of maintenance treatment; maintenance monitoring; three visits with the GP and a CMHT during maintenance treatment (one visit every 2 months); and two further visits to his/her GP. Costs for patients who respond In the Markov component of the mode, responding patients are assumed to remain on maintenance therapy for the fu mode time horizon. In addition, responding patients are assumed to continue to receive usua care by their GP and CMHT. Therefore, responders incur 10 months of maintenance treatment, monitoring for the maintenance and foow-up phase, and five visits with their GP and CMHT (one visit every 2 months). Costs of non-responding patients Patients who do not respond to acute treatment enter the non-response heath state. These patients are taken off their augmentation therapy and are assigned a standard package of care that incudes: 10 months of non-specific therapy (cacuated by averaging the cost of augmentation with ithium and with an AAP) patient care provided by a CRHTT and GP (one visit every 2 months by CRHTT and GP) monitoring costs (averaged of ithium and AAP monitoring costs). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 79

DE NOVO ECONOMIC ANALYSIS TABLE 20 Cost per cyce associated with each heath state in the maintenance phase by treatment arm Cost per cyce ( ) over the 10 months of the Markov mode Heath state Cost components Lithium AAP Remission Six months of maintenance treatment 127.27 190.89 Three CMHT visits during maintenance treatment phase Five GP visits (three during maintenance treatment phase and two during foow-up) Maintenance monitoring Response Ten months of maintenance treatment 192.74 297.62 Five visits with a CMHT and GP (three during the maintenance treatment phase and two during foow-up) Maintenance and foow-up monitoring Reapse Standard package of care: 281.01 281.01 Ten months of non-specific therapy a Patient care provided by a CRHTT and GP (one visit every 2 months) Monitoring costs (maintenance and foow-up) Discontinuation remission Three CMHT visits during maintenance treatment phase 98.00 98.00 Five GP visits (three during maintenance treatment phase and two during foow-up) Two GP visits during foow-up Discontinuation response Non-response Five visits with a CMHT and GP (three during the maintenance treatment phase and two during foow-up) Standard package of care: 143.33 143.33 281.01 281.01 Ten months of non-specific therapy a Patient care provided by a CRHTT and GP (one visit every 2 months) Monitoring costs (maintenance and foow-up) a Cost assumed to be equa to the average cost of augmentation therapy. Costs for patients who reapse A patient in the fuy symptomatic heath state (MADRS score of > 15) after having achieved remission or response is considered to be in the reapse state. Patients who reapse during the maintenance phase of the mode (either during maintenance treatment or in the 4 months foowing maintenance treatment) are assumed to receive the standard package of care received by patients in non-response at the end of the acute treatment phase for the remainder of the mode. Costs for patients who discontinue foowing remission or response Patients who discontinue maintenance therapy after experiencing remission or response are assumed to receive the same eve of patient care as their counterparts who remain on therapy. Therefore, patients who discontinue foowing remission woud incur the cost of 6 months of usua care (GP pus CMHT) and two further visits to their GP. Patients who discontinue foowing response to treatment woud accrue the cost of 10 months of usua care. 80 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Sensitivity anaysis Probabiistic sensitivity anaysis was the standard method of anaysis, and mean costs and QALYs from the PSA were used to inform the base-case resuts. In addition, severa sensitivity anayses were carried out in support of the economic evauation of augmentation of SSRI therapy with ithium compared with an AAP. These were: OWSA of cost, acute efficacy and acute discontinuation, to assess the univariate sensitivity of the deterministic mode to changes in individua parameters a threshod anaysis on the cost of AAPs scenario anaysis, to assess the sensitivity of the mode to the assumption of cass effects. Probabiistic sensitivity anayses To assess the simutaneous effect of parameter uncertainty on outcomes of cost-effectiveness, the mode was buit to be fuy probabiistic. This was achieved by assigning appropriate distributions to each parameter from which estimated vaues were repeatedy samped. Sampes of 1000, 2000 and 5000 were used in the base-case mode and the stabiity of the mode was assessed with respect to non-inearity between the deterministic and probabiistic resuts. A sampe size of 2000 was chosen, as this gave a more stabe probabiistic estimate than a sampe size of 1000 and was more efficient than a sampe size of 5000. Tabe 21 dispays the distributions assigned to each mode parameter. Uncertainty in the acute treatment phase Uncertainty around the effect of acute augmentation treatment was captured by assigning a norma distribution to the absoute treatment effect estimated from the MTC carried out as part of the cinica effectiveness anaysis (see Chapter 3, Assessment of effectiveness). However, uncertainty around patients' baseine MADRS scores has not been captured, as expained above (see Acute treatment phase). This is because it was considered that the potentia correation between baseine MADRS score and treatment effect woud compromise the assumption of parameter independence necessary for a PSA unadjusted for correation as described above (see Acute treatment phase). The probabiity of patients responding or remitting foowing discontinuation of acute treatment was informed by expert cinica opinion. Therefore, the uncertainty associated with these estimates was not readiy quantifiabe in terms of a SE or CI. Consequenty, the uncertainty around these estimates was approximated by examining the effect of varying the estimates within a boundary of ± 25% using a og-norma distribution. Discontinuation in the acute treatment phase was informed by the MTC carried out as part of the cinica effectiveness review (see Chapter 3, Assessment of effectiveness). Uncertainty around these estimates was captured by assigning a og-norma distribution to the ORs obtained from the MTC. Uncertainty in the maintenance phase The maintenance phase of the mode, which incudes 6 months of maintenance treatment pus 4 months of foow-up (where patients may remain on or cease treatment, depending on the eve of response achieved) is parameterised by the foowing probabiities: reapse for patients currenty in remission (or remission discontinuation ) reapse for patients currenty in response (or response discontinuation ) remission for patients currenty in response discontinuation remission for patients currenty in non-response (these patients wi enter the heath state of remission discontinuation ) response for patients currenty in non-response (these patients wi enter the heath state of remission discontinuation ). Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 81

DE NOVO ECONOMIC ANALYSIS TABLE 21 Summary of parameters and distributions impemented in the PSA Parameter Mean vaue SE/ variation Distribution used Acute efficacy Acute efficacy of ithium augmentation therapy (change in MADRS score) 12.58 10.42 Norma Acute efficacy AAP augmentation therapy (change in MADRS score) 11.22 9.65 Norma Probabiity of cinica improvement for non-competers/non-responders Annua probabiity of remission 5% ± 25% Log-norma Annua probabiity of response 15% ± 25% Log-norma Acute discontinuation OR for discontinuation in patients treated with ithium + SSRI vs. SSRI aone OR for discontinuation in patients augmented with an AAP + SSRI vs. SSRI aone Maintenance-phase parameters 0.92 0.85 Log-norma 1.27 0.17 Log-norma Three-year probabiity of moving from response to remission 0.500 0.064 Beta Annua probabiity of reapse for patients who have responded to treatment 0.760 0.052 Beta Annua probabiity of reapse for patients who are in remission 0.439 0.082 Beta Patient care costs ( ) CRHTTs 179.00 179.00 Gamma CMHTs 136.00 136.00 Gamma Menta heath inpatients 318.00 318.00 Gamma Outpatient setting: first attendance 228.00 228.00 Gamma Outpatient setting: foow-up 160.00 160.00 Gamma Menta Heath Secure Units 533.45 533.45 Gamma GP 36.00 36.00 Gamma Monitoring costs ( ) BMI 6.17 6.17 Gamma ECG 34.94 34.94 Gamma Fu bood count 2.35 2.35 Gamma egfr 1.03 1.03 Gamma Creatinine 1.03 1.03 Gamma Serum ithium concentration 2.82 2.82 Gamma Thyroid function 16.08 16.08 Gamma Gucose test (Biochemistry EAP 841) 1.03 1.03 Gamma Lipid profie test 2.21 2.21 Gamma Utiity vaues Remitters 0.8500 0.008 Beta Responders 0.7200 0.013 Beta Non-responders a 0.5800 0.018 Beta egfr, estimated gomeruar fitration rate. a Assumed to be the same at baseine as patients with severe depression. 82 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Beta distributions were assigned to the probabiities derived from iterature sources, and the SEs associated with these probabiities were cacuated from the data reported in the reevant iterature source. Therefore, the uncertainty around the probabiity of remission (for patients currenty in response or response discontinuation ) and of reapse was cacuated from data provided in the studies of Fekadu et a. 122 and Rush et a., 99 respectivey. The og-norma distribution was used to assess the uncertainty around probabiities estimated from expert cinica opinion, with arbitrary variation of ± 25% assumed. This is because the sampe size of the expert pane was considered to be too sma (n = 2) to cacuate the apha and beta vaues required to impement the beta distribution. Discontinuation in the maintenance phase of the mode is assumed to be an average of the treatment-specific discontinuation rates of the acute treatment phase. Therefore, uncertainty around discontinuation in the maintenance phase is accounted for by assessing the uncertainty around discontinuation in the acute treatment phase. Uncertainty around cost and quaity of ife It is assumed that there is no uncertainty surrounding the current cost of drugs to the NHS; therefore, drug costs were not varied in the sensitivity anaysis. The impact of uncertainty surrounding the costs of monitoring and patient care was assessed by assigning gamma distributions to these costs (see Tabe 21). The gamma distribution was chosen as cost data are often highy skewed and the gamma distribution is constrained to be non-negative. 121 However, as SEs were not avaiabe (and coud not be derived) for these costs, the SE of each cost was assumed to be equa to the mean cost. 121 Uncertainty around the vaues used for utiity was assessed using the beta distribution (see Tabe 21) based on SEs reported in the study by Sapin et a. 105 The anaysis was run over 2000 iterations, and the expected costs and QALYs generated from each run were averaged to determine the probabiistic ICER reported beow (see Resuts) for the base-case and scenario anayses. One-way sensitivity anayses A range of OWSA was undertaken on key mode inputs to investigate the impact the independent variation of each input had on the incrementa cost, incrementa benefit and ICERs. Each key parameter was aternatey assigned a ow and high vaue and the deterministic cost-effectiveness resuts using this vaue recorded. The deterministic mode was used for the OWSA as a pragmatic measure because the probabiistic mode was computationay intensive (average run time 10 hours). In the OWSA, it was assumed that there is no uncertainty with regards to current drug costs in the NHS; therefore, drug costs were not varied in these anayses. The interquartie ranges reported in the Nationa Schedue of Reference Costs were used as the ow and high vaues of HCP costs. However, the variation associated with the cost of monitoring was not reported; therefore, the vaues used in the OWSA were assumed to be ± 25% of the mode's base-case vaue. To assess the impact of univariate changes in acute treatment efficacy, 95% CIs were cacuated from the means and SEs reported in the MTC (used to inform the norma distribution from which treatment effect is samped in the base case) for each treatment. Then for each treatment, in turn, the upper and ower threshods of the CI associated with that treatment were aternativey used to represent the mean change in MADRS score. The mean change in MADRS score was then used to inform the norma distribution from which acute treatment efficacy is samped. The upper and ower vaues used in the assessment of the mode's sensitivity to acute discontinuation were derived from the 95% CI associated with the og-norma distribution assigned to acute discontinuation (see discussion of PSA). The uncertainty surrounding the assumption that the probabiities of reapse and of improvement (for patients in response) are independent of treatment status was assessed using assumed variance of ± 25%. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 83

DE NOVO ECONOMIC ANALYSIS Tabe 22 summarises the upper and ower vaues used in each univariate sensitivity anaysis. The resuts of this and other sensitivity anayses are presented beow (see Resuts). In addition, any parameters identified as being significant in the OWSA based on the deterministic mode were assessed in the probabiistic mode. TABLE 22 Upper and ower vaues used to inform univariate sensitivity anayses Parameter HCP costs ( ) Base-case vaue Lower vaue Upper vaue Source of variation CRHTTs 179.00 136 207 Interquartie ranges reported CMHTs 136.00 109 160 in Nationa Schedue of Reference Costs GP 36.00 30 40 Menta heath inpatient costs 312 278 339 Monitoring costs ( ) BMI 6.17 4.62 7.71 Assumed to be 25% ECG 34.94 26.20 43.67 Fu bood count 2.35 1.76 2.94 egfr 1.03 0.77 1.28 Creatinine 1.03 0.77 1.28 Serum ithium concentration 2.82 2.11 3.52 Thyroid function 16.08 12.06 20.10 Gucose test (Biochemistry EAP 841) 1.03 0.77 1.28 Lipid profie test 2.21 1.66 2.76 Acute treatment efficacy Change in MADRS score foowing treatment with SSRI pus ithium Samped from norma distribution, mean 12.58 Samped from norma distribution, mean 33.00 Samped from norma distribution, mean 7.84 95% CI estimated from resuts of MTC Change in MADRS score foowing treatment with SSRI pus an AAP Samped from norma distribution, mean 11.22 Samped from norma distribution, mean 30.13 Samped from norma distribution, mean 7.69 Acute discontinuation Acute ithium discontinuation OR 0.92 OR 0.13 OR 1.75 95% CI associated with Acute AAP discontinuation OR 1.27 OR 0.90 OR 3.32 og-norma distribution (fitted to ORs as part of PSA) Maintenance-phase probabiities Remitting for patients in response discontinuation Reapse for patients in response discontinuation Reapse for patients in remission discontinuation 3.8% 2.9% 4.8% Assumed to be ± 25% 21.2% 15.9% 26.5% Assumed to be ± 25% 8.9% 6.7% 11.1% Assumed to be ± 25% egfr, estimated gomeruar fitration rate. 84 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Threshod anaysis on the cost of atypica antipsychotic drugs At present, there is significant uncertainty surrounding the future cost of AAPs to the NHS. This is because at the time of writing this report the patents hed by Ei Liy for oanzapine and AstraZeneca for quetiapine were cose to expiration. (Note that by the time of pubication of this report these patents had expired.) In addition, the patent for aripiprazoe hed by Otsuka wi expire in 2015 16. 128 Therefore, the price of AAPs is ikey to decrease as the manufacturers face competition from the generic market. However, there is uncertainty over the extent of the price reduction expected. A threshod anaysis on the price of aripiprazoe, oanzapine and quetiapine used in the base case was carried out to expore the impact of various price reductions on the base-case mode resuts. The resuts of this are dispayed beow (see Resuts). Scenario anayses In the base case, the mode assumes a cass effect for the SSRIs and the AAPs. This assumption was based on an absence of evidence of a difference of effect, rather than evidence of no difference in effect. Therefore, there is a degree of uncertainty surrounding this assumption. Moreover, data were not avaiabe to reax this assumption. However, a scenario anaysis was carried out to assess the sensitivity of the mode to the use of additiona data to inform the cass effect of AAP augmentation. In the scenario anaysis, data were used from a MTC carried out as part of the cinica effectiveness sensitivity anaysis (see Chapter 3, Assessment of effectiveness). The MTC incuded trias in which ithium, oanzapine, aripiprazoe or quetiapine were used to augment a patient popuation treated with a mixture of SSRI or venafaxine therapy. Tabe 23 summarises the acute efficacy and discontinuation derived from the MTC and used in this scenario anaysis. The resuts of this anaysis are presented beow (see Resuts). Resuts Base-case resuts The probabiistic and deterministic base-case resuts of the comparison between augmentation of SSRI therapy with ithium and augmentation with an AAP are dispayed in Tabe 24. Deterministic resuts are reported to faciitate understanding on the OWSA resuts. The cost-effectiveness pane and cost-effectiveness acceptabiity curve (CEAC) associated with the probabiistic base case are dispayed in Figures 16 and 17, respectivey. The base-case resuts indicate that augmentation of SSRI therapy with an AAP is dominated by augmentation of a SSRI with ithium. The cost-effectiveness pane (see Figure 16) represents the incrementa costs and TABLE 23 Acute efficacy and discontinuation used in the scenario anaysis of cass effect Treatment-specific probabiity (%) Parameter Lithium AAP Sources Remission a 16.7 20.7 Samping based on resuts Response a 17.3 16.4 of MTC Non-response a 66.0 62.9 Discontinuation b 18.0 23.0 Thase et a. (2007) 53 MTC a Cacuated from the PSA (1000 patients samped 2000 times and the proportions of remission, response and non-response averaged). b The rate of discontinuation associated with SSRI therapy reported in Thase et a. 53 is adjusted by the RRs of 0.93 and 1.18 for ithium and AAPs, respectivey, estimated from the MTC. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 85

DE NOVO ECONOMIC ANALYSIS TABLE 24 Base-case probabiistic and deterministic resuts of the comparison between augmentation of SSRI therapy with ithium and augmentation with an AAP Augmentation treatment Cost ( ) QALYs Incrementa cost ( ) Incrementa QALYs ICER ( /QALY) Probabiistic resuts a Lithium 4739 1.253 AAP 5644 1.225 905 0.03 AAP is dominated by ithium Deterministic resuts Lithium 4702 1.258 AAP 5643 1.226 941 0.03 AAP is dominated by ithium a Probabiistic resuts were generated by averaging the resuts of 2000 probabiistic runs. 2000 1500 1000 Incrementa costs ( ) 500 0 0.080 0.060 0.040 0.020 0.000 0.020 0.040 0.060 0.080 500 1000 1500 2000 Incrementa QALYs FIGURE 16 Cost-effectiveness pane associated with base-case probabiistic anaysis of augmentation of SSRI therapy with ithium vs. augmentation with an AAP. QALYs associated with AAP augmentation (vs. ithium augmentation) as points on the graph. This scatter pot suggests that there is no uncertainty about the dominance resut and it is important to note that AAP augmentation did not provide more benefit than ithium augmentation in any of the probabiistic runs. One-way sensitivity anaysis One-way sensitivity anaysis of cost, acute efficacy and acute discontinuation were carried out on the deterministic base-case resut, the resuts of which are presented in Tabe 25 and the tornado diagram dispayed in Figure 18. The resuts of the OWSA indicated that the key drivers of the cost-effectiveness resuts are acute efficacy and discontinuation. Changes in assumptions around the probabiity of reapse and the probabiity of remission in 86 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 AAPs Lithium 0.2 0.1 0.0 0 800 1600 2400 3200 4000 4800 5600 6400 7200 8000 8800 9600 10,400 11,200 12,000 12,800 13,600 14,400 15,200 16,000 16,800 17,600 18,400 19,200 20,000 20,800 21,600 22,400 23,200 24,000 24,800 25,600 26,400 27,200 28,000 28,800 29,600 Probabiity of being cost-effective Wiingness to pay per QALY gained ( ) FIGURE 17 Cost-effectiveness acceptabiity curve associated with the probabiistic base case of augmentation of SSRI therapy with ithium vs. augmentation with an AAP. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 87

DE NOVO ECONOMIC ANALYSIS TABLE 25 Resuts of OWSAs Parameter Incrementa costs ( ) Incrementa QALY Cost ( )/ QALY (ICER) Base case 941 0.03 Lithium dominates Discontinuation, AAPs Lower 95% CI 964 0.03 Lithium dominates Upper 95% CI 917 0.04 Discontinuation, ithium Lower 95% CI 1079 0.05 Lithium dominates Upper 95% CI 700 0.01 113,596 a Change in MADRS score, AAPs Lower 95% CI 917 0.20 4672 b Upper 95% CI 815 0.11 Lithium dominates Change in MADRS score, ithium Lower 95% CI 2026 0.26 Lithium dominates Upper 95% CI 169 0.08 1996 b CRHTT Lower costs 836 0.03 Lithium dominates Upper costs 992 0.03 CMHT Lower costs 941 0.03 Lithium dominates Upper costs 941 0.03 GP Lower costs 936 0.03 Lithium dominates Upper costs 941 0.03 BMI Lower costs 941 0.03 Lithium dominates Upper costs 944 0.03 ECG Lower costs 941 0.03 Lithium dominates Upper costs 943 0.03 Fu bood count Lower costs 941 0.03 Lithium dominates Upper costs 941 0.03 Creatinine Lower costs 941 0.03 Lithium dominates Upper costs 941 0.03 Serum ithium concentration Lower costs 941 0.03 Lithium dominates Upper costs 938 0.03 88 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 25 Resuts of OWSAs (continued) Parameter Incrementa costs ( ) Incrementa QALY Cost ( )/ QALY (ICER) Thyroid function Lower costs 941 0.03 Lithium dominates Upper costs 937 0.03 Gucose test Lower costs 941 0.03 Lithium dominates Upper costs 942 0.03 Lipid profie test Lower costs 941 0.03 Lithium dominates Upper costs 942 0.03 egfr Lower costs 941 0.03 Lithium dominates Upper costs 941 0.03 Inpatients costs Lower costs 941 0.03 Lithium dominates Upper costs 969 0.03 Response to remission rates in patients sti on maintenance therapy (ower vaue) Response to remission rates in patients sti on maintenance therapy (higher vaue) Reapse rates in response patients who have discontinued therapy (ower vaue) Reapse rates in response patients who have discontinued therapy (higher vaue) Reapse rates in remission patients who have discontinued therapy (ower vaue) Reapse rates in remission patients who have discontinued therapy (higher vaue) 941 0.03 Lithium dominates 941 0.03 Lithium dominates 942 0.03 Lithium dominates 940 0.03 Lithium dominates 941 0.03 Lithium dominates 941 0.03 Lithium dominates egfr, estimated gomeruar fitration rate. a AAPs have an ICER of > 20,000/QALY. b AAPs have an ICER of < 20,000/QALY. the maintenance phase had minima impact on the overa resuts. Simiary, changes in costs did not change the overa resut of ithium dominance. However, changes in acute efficacy or discontinuation reversed the direction of the cost-effectiveness resuts. For exampe, when a ow eve of acute efficacy (upper 95% CI) for ithium was assumed, the direction of cost-effectiveness reversed and AAP augmentation had ICERs of < 20,000/QALY compared with ithium augmentation, suggesting that AAPs wi be the preferred strategy. A simiar resut was observed when a high eve of acute efficacy for the AAPs (ower 95% CI) was assumed. When higher eves of discontinuations were assumed for ithium, AAP augmentation resuted in more heath benefits, abeit with an estimated ICER of > 100,000/QALY. Therefore, the impact of univariate changes in these key parameters (acute efficacy and discontinuation) on the resuts of the probabiistic mode were examined. The resuts of these anayses are dispayed in Tabe 26 and the tornado diagram in Figure 19. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 89

DE NOVO ECONOMIC ANALYSIS Reapse rates in remission patients who have discontinued therapy Reapse rates in response patients who have discontinued therapy Response to remission rates in patients sti on maintenance therapy Inpatients costs egfr upper costs Lipid profie test upper costs Gucose test upper costs Thyroid function upper costs Serum ithium concentration upper costs Creatinine upper costs Fu bood count upper costs ECG upper costs Parameter BMI upper costs GP upper costs CMHT upper costs CRHTT upper costs Change in MADRS score upper ithium 95% CI Change in MADRS score AAPs upper 95% CI Discontinuation upper ithium 95% CI Discontinuation AAPs upper 95% CI Base case 60 40 20 0 20 40 60 80 100 120 Cost per QALY UK ( thousand) FIGURE 18 Resuts of the OWSAs on the deterministic base case of augmentation of SSRI therapy with ithium vs. augmentation with an AAP. egfr, estimated gomeruar fitration rate. 90 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 TABLE 26 Impact of OWSAs (response and discontinuation) on the resuts of the probabiistic mode of augmentation of SSRI therapy with ithium vs. augmentation with an AAP Parameter Incrementa costs ( ) Incrementa QALY Cost ( )/QALY (ICER) Base case 905 0.03 32,650 Change in MADRS score (response) Change in MADRS score, AAPs ower 95% CI 916 0.20 4688 Change in MADRS score, upper ithium 95% CI 174 0.09 1976 Discontinuations Discontinuation, upper ithium 95% CI 676 0.01 85,600 Discontinuation upper ithium 95% CI Parameter Change in MADRS score upper ithium 95% CI Change in MADRS score AAPs ower 95% CI Base case 40 20 0 20 40 60 80 100 Cost/QALY ( thousand) FIGURE 19 Impact of the OWSAs (response and discontinuation) on the resuts of the probabiistic mode of augmentation of SSRI therapy with ithium vs. augmentation with an AAP. The mean expected vaues from the probabiistic mode compare we with the deterministic mode, indicating that the resuts are robust to pausibe changes in parameter estimates. Threshod anaysis As discussed above (see Sensitivity anaysis), a threshod anaysis on the cost of AAPs to the NHS was carried out to assess the uncertainty around these costs, in the current cimate of patent expiry. The resuts of this sensitivity anaysis are presented in Tabe 27. The resuts of this threshod anaysis further highight the robustness of the mode resuts to changes in costs. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 91

DE NOVO ECONOMIC ANALYSIS TABLE 27 Resuts of the threshod anaysis of AAP drug cost Parameter Incrementa costs ( ) Incrementa QALY Cost ( )/QALY (ICER) Base case 941.06 0.03 30,206 AAPs fa in price (%) 60 786.47 0.03 25,244 80 734.94 0.03 23,590 90 709.18 0.03 22,763 Fuoxetine (non-proprietary) 20 mg 919 0.03 29,512 Scenario anayses As discussed above (see Sensitivity anayses), a scenario anaysis was carried out in support of the economic evauation of augmentation of SSRI therapy with ithium compared with an AAP. In this scenario anaysis, additiona data from a sensitivity anaysis carried out as part of the cinica effectiveness section were used to inform the cass effect of AAP augmentation. The cinica sensitivity anaysis was a MTC incuding trias in which ithium, oanzapine, aripiprazoe or quetiapine was used to augment a patient popuation treated with a mixture of SSRI or venafaxine therapy. The probabiistic and deterministic resuts of this are dispayed in Tabe 28, with the cost-effectiveness pane and CEAC associated with the probabiistic resut dispayed in Figures 20 and 21, respectivey. Compared with the base case, the scenario anaysis into cass effect resuts in a reduction in the incrementa benefit of augmentation therapy with ithium. Heath economics discussion The base-case resuts indicated that augmentation of SSRI therapy with an AAP is dominated by augmentation of a SSRI with ithium in patients with TRD. Furthermore, PSA suggested that the benefit obtained with AAP augmentation was consistenty dominated by the benefit obtained with ithium augmentation; none of the probabiistic runs resuted in an incrementa QALY gain for AAP augmentation over ithium augmentation. However, OWSA reveaed that the mode is highy sensitive to the reative eve of discontinuation assumed and changes in the distributions used to sampe the acute efficacy of treatment. TABLE 28 Probabiistic and deterministic resuts of the scenario anaysis extending the assumption of cass effect based on data from a cinica sensitivity anaysis Augmentation treatment Cost ( ) QALYs Probabiistic resuts a Incrementa cost ( ) Incrementa QALYs ICER ( /QALY) Lithium 4875 1.22809 AAP 5651 1.23307 776 0.005 155,828 Deterministic resuts Lithium 4876 1.22671 AAP 5647 1.23399 771 0.01 105,850 a Probabiistic resuts were generated by averaging the resuts of 2000 probabiistic runs. 92 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 2500 2000 1500 Incrementa costs ( ) 1000 500 0 0.050 0.040 0.030 0.020 0.010 0.000 0.010 0.020 0.030 0.040 0.050 500 1000 1500 2000 2500 Incrementa QALYs FIGURE 20 Cost-effectiveness pane associated with the probabiistic resut of the scenario anaysis extending the assumption of cass effect based on data from a cinica sensitivity anaysis. In the absence of direct comparison in a RCT, the acute efficacy of treatment was estimated from a MTC carried out as part of the cinica effectiveness review. However, there is a paucity of RCTs considering ithium as an augmentation therapy, and no tria in the required popuation (faied on two or more antidepressant therapies in the current episode of depression) was identified in the cinica effectiveness review. Consequenty, a RCT that matched the prespecified incusion criteria as cosey as possibe was used as a surrogate. The RCT (n = 33) considered ithium augmentation in a popuation of patients who had faied on one or more previous antidepressant regimes. It is important to note that this tria recruited ony 33 patients (17 in the ithium arm and 16 in the SSRI-aone arm) and, as might be expected with such a sma sampe size, the resuts are uncertain; this uncertainty is propagated through to the MTC. In addition, it is important to note that the incusion of some patients with ony one previous faiure of an antidepressant regime may resut in an overa bias in favour of ithium. However, as the reative effect of ithium + SSRI (compared with SSRI aone) was used to inform the MTC, which may be a consistent benefit over antidepressant therapy aone irrespective of number of previous antidepressants, the difference in study popuations may have itte effect on the resuts. Nevertheess, the presence or extent of this potentia bias remains unknown and it is important to consider the impact of this uncertainty when interpreting the cost-effectiveness resuts. The mode required data on the outcomes of remission, response and discontinuation. However, data on remission were not reported and the definitions of response varied across the trias incuded in the MTC. Therefore, a samping method was used to generate the required probabiities; the treatment effect (change in MADRS score from baseine) with each augmentation therapy was samped from a distribution of possibe effects. The 95% CrIs for the distributions used for each augmentation therapy overapped (ithium 95% CrI 33.00 to 7.84; AAP 95% CrI 30.13 to 7.69) suggesting a non-significant difference in the effect of each treatment. However, the consequence of wider CrIs for ithium ed to a higher proportion of patients in the ithium arm achieving remission, a phenomenon that was maintained throughout the probabiistic anaysis. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 93

DE NOVO ECONOMIC ANALYSIS 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 AAPs Lithium Probabiity of being cost-effective 0.20 0.10 0.00 0 800 1600 2400 3200 4000 4800 5600 6400 7200 8000 8800 9600 10,400 11,200 12,000 12,800 13,600 14,400 15,200 16,000 16,800 17,600 18,400 19,200 20,000 20,800 21,600 22,400 23,200 24,000 24,800 25,600 26,400 27,200 28,000 28,800 29,600 Wiingness to pay per QALY gained ( ) FIGURE 21 Cost-effectiveness acceptabiity curve associated with the probabiistic resut of the scenario anaysis extending the assumption of cass effect based on data from a cinica sensitivity anaysis. 94 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Discontinuation was a key driver of the mode, as patients who discontinued their acute therapy were far ess ikey to experience response or remission. The ORs of discontinuation (vs. SSRI therapy aone) used to inform the economic mode were based on the same MTC used to derive the distributions of acute efficacy. Simiary, the 95% CrIs for the ORs associated with each augmentation therapy overapped (ithium 95% CrI 0.13 to 3.32; AAP 95% CrI 0.90 to 1.75) suggesting a non-significant difference in the eve of discontinuation. Moreover, the mean OR for discontinuation with ithium (0.92) was amost equa to the ower 95% CrI of the OR for discontinuation with AAP (0.90), resuting in generay more favourabe eves of discontinuation with ithium. The OWSA indicated that the acute efficacy (change in MADRS score) and eve of discontinuation associated with each treatment are key drivers of the mode resuts. Therefore, it is important to consider the combined effect of the uncertainty surrounding these inputs when interpreting the mode's resuts. Furthermore, in the absence of evidence to inform separate comparisons of each augmentation strategy, the mode assumed a cass effect for the SSRIs and AAPs. There is evidence of a cass effect with the SSRIs from a study by Kroenke et a. 128,129 However, no evidence of a cass effect for the AAPs was identified and the resuts of the cinica sensitivity anaysis of cass effect were inconcusive. Therefore, this additiona uncertainty around the vaidity of a cass effect shoud be considered when impementing any recommendations based on the economic evauation presented as part of this review. Concusion The economic evauation presented as part of this review indicated that, in patients with TRD, augmentation of SSRI therapy with an AAP may be dominated by augmentation with ithium. However, the cost-effectiveness resuts are predominanty driven by estimates of acute efficacy and discontinuation derived from the assessment of cinica effectiveness. The resuts of the cinica effectiveness anayses indicate that there is no statisticay significant difference between the two augmentation strategies. In addition, the genera paucity of RCTs avaiabe for augmentation therapy using SSRI as background treatment in patients with TRD resuts in a high eve of uncertainty in both the cinica and economic anayses. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 95

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 6 Overa discussion This section summarises the principa findings of the cinica effectiveness and cost-effectiveness review of the iterature surrounding the augmentation of SSRI therapy with either ithium or an AAP. The key studies identified as part of the cinica effectiveness review and the resuts of anayses based on these are discussed. Consideration is then given to the interpretation and significance of these resuts. The resuts of the cost-effectiveness and QoL iterature reviews are aso discussed and the deveopment of the de novo mode summarised. Any impications of the findings from the de novo economic evauation are considered in conjunction with any apparent uncertainty. Statement of principa findings Description of the identified cinica studies Initiay, 11 RCTs 30,43,46 53 were identified that considered the augmentation of SSRI therapy in the reevant popuation (patients who had faied to respond to two or more antidepressant regimens). However, 10 of these RCTs considered the augmentation of SSRI therapy with an AAP. 43,46 53 The remaining RCT 30 considered the augmentation of SSRI or SNRI (venafaxine) therapy with ithium compared with AAP. None of the identified studies considered augmentation of SSRI therapy with ithium. The primary anaysis did not assume a cass effect for either the SSRIs or AAPs. Therefore, the singe tria identified that considered ithium was ineigibe for incusion in the primary anaysis. Consequenty, to enabe a primary anaysis of the reative treatment effect of augmentation with ithium compared with augmentation with an AAP, a surrogate ithium tria was incuded. The patient popuation of this tria had faied on one or more antidepressant therapy. Summary of cinica findings Primary and secondary anayses of the reative cinica effectiveness were carried out based on the RCTs identified as part of the cinica effectiveness review for the foowing comparisons: SSRI + AAP vs. SSRI aone SSRI + ithium vs. SSRI aone SSRI + AAP vs. SSRI + ithium. The primary anayses assumed that the SSRIs and AAPs were associated with different efficacy profies, whereas the secondary anayses assumed a cass effect for both casses of drug. Seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor aone Of the 12 RCTs identified, 10 were used to inform the comparison of SSRI + AAP with SSRI aone (six in the primary anaysis and a 10 in the secondary anaysis). The six trias in the primary anaysis were a comparisons of fuoxetine (SSRI) pus oanzapine (AAP) with fuoxetine aone. Not a incuded trias reported data for each outcome assessed. To summarise, six reported response based on the MADRS or HAMD scae, five reported remission based on the MADRS or HAMD scae, four reported MD in MADRS score from baseine and five reported a-cause withdrawa. For the outcome of response, the use of fixed-effects meta-anaysis demonstrated a statisticay significant benefit of fuoxetine pus oanzapine compared with fuoxetine aone (OR 1.48; 95% CI 1.13 to 1.94), with moderate heterogeneity (I 2 = 53%; p = 0.07). The significance of this benefit was maintained when a random-effects meta-anaysis was used to synthesise the data (OR 1.60; 95% CI 1.01 to 2.53). The resuts of a fixed-effects meta-anaysis for the outcome of remission demonstrated a statisticay significant increase in remission for patients treated with fuoxetine pus oanzapine compared with fuoxetine aone (OR 1.77; 95% CI 1.27 to 2.47), with no statistica heterogeneity (I 2 =0%;p = 0.75). The outcome of response was the Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 97

OVERALL DISCUSSION ony outcome considered in the secondary anaysis assuming a cass effect. The resuts of both the fixed- and random-effects meta-anayses used to assess this outcome were consistent with the primary anaysis; a statisticay significant improvement in response was seen with AAP augmentation of baseine therapy. Meta-anaysis of the change in MADRS score from baseine resuted in a statisticay significant MD of 2.04 (95% CI 3.25 to 0.82) in favour of treatment with fuoxetine pus oanzapine. However, the eve of heterogeneity associated with this anaysis was high (I 2 = 73%; p = 0.01). It was noted that one of the studies reporting change in MADRS score from baseine (the pooed anaysis by Thase et a. 53 ) reported a much arger MD compared with other trias incuded in the meta-anaysis. An exporatory meta-anaysis excuding data from Thase et a. 53 resuted in a non-statisticay significant MD of 1.15 (95% CI 2.49 to 0.19) in favour of treatment with fuoxetine pus oanzapine, with no statistica heterogeneity (I 2 = 0%; p = 0.38). The resuts of a fixed-effects meta-anaysis of a-cause withdrawa suggested a statisticay non-significant reduction in discontinuations with fuoxetine aone compared with fuoxetine pus oanzapine (OR 1.25; 95% CI 0.91 to 1.71), with no statistica heterogeneity (I 2 =0%; p = 0.51). Seective serotonin reuptake inhibitor pus ithium compared with seective serotonin reuptake inhibitor aone Ony one tria comparing SSRI pus ithium with SSRI aone was identified in the cinica effectiveness review. However, two definitions of response were used, neither of which was consistent with the definition of response used in this review ( 50% reduction in MADRS score from baseine). The resuts of this tria regarding response indicate a non-significant trend towards improved response with fuoxetine pus ithium compared with fuoxetine aone (a priori definition of response, OR 1.48; 95% CI 0.37 to 5.95; post hoc definition of response, OR 3.85; 95% CI 0.80 to 18.62). The MD in change in MADRS score from baseine between fuoxetine + ithium and fuoxetine aone was 3.79 (95% CI 11.25 to 3.67). This represents a non-significant improvement from baseine score with fuoxetine pus ithium compared with fuoxetine aone. Furthermore, discontinuations were sighty ower with fuoxetine pus ithium (4/17) than with fuoxetine aone (5/16). Seective serotonin reuptake inhibitor pus atypica antipsychotic compared with seective serotonin reuptake inhibitor pus ithium Of the 12 trias identified in the cinica effectiveness review, 10 were used to inform the comparison of SSRI + AAP compared with SSRI + ithium (seven in the primary anaysis and a 10 in the secondary anaysis). Of the seven trias incuded in the primary anaysis, six were comparisons of fuoxetine pus oanzapine with fuoxetine aone and one ooked at fuoxetine pus ithium compared with ithium aone. Not a incuded trias reported data for each outcome assessed. To summarise, seven trias reported data for response based on MADRS or HAMD score, five trias reported data for MD in MADRS score from baseine and six trias reported data on a-cause withdrawa. For the outcome of response (using data from the post hoc definition of response from the ithium tria), a statisticay non-significant trend in favour of augmentation with ithium was detected (OR 4.15; 95% CrI 0.25 to 20.34). However, it is important to note that the definition of response used in the tria comparing ithium pus fuoxetine with fuoxetine aone differed from that used in the trias comparing fuoxetine pus oanzapine with fuoxetine aone. The MD in the change in MADRS score from baseine determined from the MTC aso suggested a non-significant trend in favour of augmentation with ithium (OR 1.47; 95% CrI 9.10 to 6.41). Furthermore, the resuts of the MTC with respect to a-cause withdrawa indicated a non-significant benefit for patients augmented with ithium (OR 0.74; 95% CrI 0.10 to 2.66). Discussion of cinica findings The avaiabe cinica effectiveness data informing the comparison of SSRI + AAP with SSRI + ithium in the primary anaysis was based on fuoxetine + oanzapine 43,49,51 53 compared with fuoxetine + ithium. 59 The resuts from the MTC of the star-shaped network demonstrate no significant differences between treatment regimens for any of the outcomes assessed. A non-significant trend in benefit was observed for the 98 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 ithium-based augmentation strategy compared with fuoxetine + oanzapine for response, mean change in MADRS from baseine, and fewer discontinuations. The resuts of the MTC aso demonstrated a non-significant trend in favour of the oanzapine-based augmentation strategy compared with fuoxetine + ithium for fewer adverse events. However, care shoud be taken when interpreting non-significant resuts. When the resuts of the MTC are compared with the individua resuts for the pairwise meta-anayses, there is genera agreement with the resuts obtained when SSRI is used as the baseine. This suggests a reasonabe fit of the modeing approach used within the MTC. The radiating star shape of the network means that ony the trias providing the resuts from the pair-wise meta-anayses are providing the resuts within the MTC. The resuts for the ithium-based augmentation strategy compared with SSRI aone in the MTC tend to have wider 95% CrIs than the 95% CIs provided from the singe tria informing this comparison. 59 This is ikey to be due to the random-effects mode tending to be the preferred mode for the MTC outcomes assessed. The resuts for both the pair-wise meta-anaysis and MTC estimates of SSRI + AAP compared with SSRI aone showed a statisticay significant benefit in favour of augmentation with AAP for the outcomes of response and mean change in MADRS score. The equivaent resuts for SSRI + ithium compared with SSRI aone showed a statisticay non-significant trend in favour of augmentation with ithium. The resuts for ithium augmentation coud be considered inconcusive, athough it shoud be noted that they are based on data from ony a sma subgroup of patients in one RCT 59 and other pubications have reported resuts demonstrating ithium to be an effective augmentation strategy. 60,65,66 A recent meta-anaysis by Crossey et a. 60 incuded 10 RCTs and demonstrated a statisticay significant benefit in terms of response rate with ithium augmentation compared with pacebo (OR 3.11; 95% CI 1.80 to 5.37). This meta-anaysis is not entirey comparabe with the patient popuation under review in the current research because it incuded patients with bipoar disorder, patients on various antidepressants and patients with a minimum of one previous antidepressant faiure. However, the Crossey et a. 60 meta-anaysis does provide strong evidence to suggest that ithium is an effective augmentation agent in TRD and thus is supportive of the resuts from the MTC. Reapse rates and mortaity were prespecified outcomes of interest for this review; however, none of the trias incuded reported comparabe mortaity or reapse rate data. In addition, extremey imited subgroup data, if any, were reported for the trias incuded in the cinica effectiveness review, and no tria reported suitabe subgroup data for the prespecified subgroup anayses. The trias incuded in this review were vaidated against the trias incuded in the NICE cinica guideine for depression in aduts (CG90) 16 and those incuded in two separate systematic reviews; one for the comparison of AAP augmentation with pacebo in MDD 11 and the other for ithium augmentation compared with pacebo in TRD. 60 A three pubications incuded additiona trias compared with this review, athough none of the additiona trias were found to be suitabe for incusion. Moreover, each tria was excuded for mutipe reasons based on the incusion criteria for this review. Mosty, the additiona RCTs did not meet the foowing criteria: the augmentation of SSRI as baseine therapy; 67 81 a 4-week minimum duration of treatment; 74 80 or two or more faiures of antidepressant therapy in the current episode of depression 67 73 (it is uncear how many trias in Crossey et a. 60 are affected by this criterion). The resuts of both the Crossey et a. 60 meta-anaysis evauating ithium compared with pacebo augmentation and the Neson et a. 11 review of AAP compared with pacebo augmentation demonstrated that the respective augmentation agents were statisticay significanty more effective than pacebo at achieving treatment response in peope with MDD or TRD. The sensitivity anaysis based on a cass effect (sensitivity anaysis 1) aowed the incusion of two additiona AAPs (aripiprazoe and quetiapine) in the MTC. For the outcomes assessed (response) the non-significant trend favouring ithium augmentation observed in the primary anaysis is diminished (mean OR changes from 4.15 to 1.07) and remains non-significant. This coud be supportive evidence for no cinicay meaningfu Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 99

OVERALL DISCUSSION difference between the two augmentation strategies, a refection of the ack of information avaiabe for the comparison, or an indication that the assumption of a cass effect is fawed and that there is a difference between the individua treatments within a cass. With regards to the ast concern, the tria providing information on quetiapine in the MTC 30 demonstrated a non-significant trend in favour of a quetiapinebased augmentation strategy compared with ithium-based augmentation (OR 1.25; 95% CI 0.74 to 2.12). The resut of this tria is thus in contrast with the resut of the MTC and therefore suggests a difference in treatment effect between the different AAPs. Summary of the cost-effectiveness review No economic evauations were identified that considered the reative cost-effectiveness of augmentation of SSRI therapy with ithium or with an AAP. However, four economic evauations were identified that incuded a popuation of peope with depression who were treatment resistant. Each study highighted issues surrounding the economic evauation of TRD, particuary around the ong-term consequences of treatment and the paucity of data avaiabe in this area. A variety of approaches were taken to assess the costeffectiveness of various interventions (mosty SSRI therapy). The hybrid approach used by Simpson et a. 89 was considered to be most reevant to the current decision probem and was used to inform the structure of the de novo economic mode. Summary of quaity-of-ife review A systematic review of the QoL iterature was carried out to identify: HSUV to inform the economic mode any issues that positivey or negativey impact on the QoL of depressed patients. Five studies reporting utiity vaues 91,102 105 by severity of depression or eve of response were identified. Of these, ony one study by Sapin et a. 105 met the criteria for eicitation and vauation outined in the NICE methods guide. 108 The study was carried out in France but used UK genera pubic vauation scores. Sapin et a. 105 reported utiity by eve of treatment response (remission, response and non-response) and by severity of depression (mid, moderate and severe). Therefore, it was these utiity vaues that were used to inform the QALY cacuations in the de novo economic mode. A further 12 studies were identified that considered the overa QoL in depressed patients. 30,43,46 53,59 There is evidence from these studies to suggest that treatments for depression are effective in improving the QoL of patients, particuary if treatment is sustained. In addition, the QoL evidence base indicates that pharmacoogica treatment of depression does not resut in any improvement of physica outcomes, and physica therapies do not resut in any improvement of psychoogica symptoms. This, in turn, suggests that the physica and psychoogica domains of QoL are independent of each other in determining the overa QoL of a depressed patient. Summary of economic evauation A hybrid economic mode was constructed to simuate the cinica and economic consequences of augmenting an SSRI with either ithium or an AAP in the treatment of TRD. The mode considered outcomes from the perspective of the NHS over a 1-year time horizon. The mode consists of two distinct components: a decision tree (8-week time horizon) and a Markov component (10-month time horizon). A hybrid mode was chosen, as this faciitated capturing the granuarity of the acute treatment phase and aso accounted for the patient progression within 1 year. The mode was constructed around the eve of response to acute treatment (assessed by changes in MADRS score) and the extent to which this response (or otherwise) is maintained. The acute efficacy and discontinuation associated with each treatment was obtained from a MTC carried out as part of the cinica effectiveness review. Discontinuation was a key driver of the mode, as patients who discontinued their acute therapy were far ess ikey to experience response or remission. 100 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Furthermore, as a resut of the paucity of ong-term data, progression within the Markov component of the mode was assumed to be independent of treatment regimen. The progression of patients who did not respond to acute treatment (and patients who reapsed during the Markov component of the mode) was simpified such that these patients were assumed not to benefit from further augmentation treatment and received a standard package of care. It is acknowedged that, in reaity, patients who do not respond to (or reapse foowing) acute augmentation therapy are ikey to experience some benefit from further treatment (e.g. change in augmentation treatment, ECT, etc.). However, these foow-on patient pathways are diverse and there is a paucity of data to inform them, and, therefore, the decision was made to simpify this aspect of patient care. In addition, whie non-responding patients were exposed to the probabiity of spontaneous remission or response, in accordance with expert cinica opinion, patients who reapsed were assumed to enter an absorbing heath state. Summary of economic findings The base-case resuts indicated that augmentation of SSRI therapy with an AAP is dominated by augmentation of an SSRI with ithium in patients with TRD. Furthermore, PSA suggested that the overa heath benefit obtained with AAP augmentation was consistenty ess than the overa heath benefit obtained with ithium augmentation; none of the probabiistic runs resuted in an incrementa QALY gain for AAP augmentation over ithium augmentation. However, the cost-effectiveness resuts are predominanty driven by estimates of acute efficacy and discontinuation derived from the assessment of cinica effectiveness. The resuts of the cinica effectiveness anayses indicate that there is no statisticay significant difference between the two augmentation strategies. In addition, the genera paucity of RCTs considering the augmentation of SSRI therapy in a TRD popuation resuts in a high eve of uncertainty in the cinica effectiveness resuts. In particuar, there are a imited number of robust data avaiabe to inform the cinica effectiveness of ithium augmentation. This uncertainty is carried through to the economic anaysis and is potentiay perpetuated by the structura assumptions made. Athough every effort has been made to ensure that the mode structure is in ine with current cinica expectations, it is acknowedged that the natura history of TRD remains somewhat unpredictabe. Strengths and imitations and uncertainty of the assessment The main strengths of this review were the: systematic identification of studies used to inform the cinica effectiveness and cost-effectiveness anayses robust methodoogy used to synthesise the cinica efficacy evidence in meta-anaysis and MTC consutation with cinica experts throughout the deveopment process of the de novo economic evauation use of samping to synthesise the cinica effectiveness evidence within the de novo economic mode. To systematicay identify cinica effectiveness and cost-effectiveness evidence, a reevant databases were searched from inception. In addition, the CCDAN was contacted for access to their study registries. The website CinicaTrias.gov was searched to identify reevant ongoing cinica trias and cinica experts in the reevant therapy areas were contacted for detais of pubished or unpubished trias. This evauation of the avaiabe iterature constitutes the most comprehensive research into augmentation therapy for TRD. Athough few RCTs were identified, the identified RCTs represent the best avaiabe data on which to base a comparison of augmentation of SSRI with ithium or AAP. In the absence of a direct comparison in a RCT, use of a MTC has been shown to be one of the most reiabe methods for indirecty Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 101

OVERALL DISCUSSION comparing treatments estimates. 39 By using a network of connected trias, in which a new tria has to incude a comparator aready existing in the network, vaid estimates of reative efficacy can be obtained. 33 35 The key criterion for a vaid network is to be popuated with a set of RCTs that are as simiar as possibe. A strength of this review is the rigour with which the incusion/excusion criteria were appied to obtain as homogeneous as possibe set of trias. It is unfortunate that, for the ithium comparator, the nearest tria identified to the type of tria required was a ess than perfect match for one incusion criterion. 59 However, the researchers (VH, SB, SJE) agreed that it was the best approximation of the required tria avaiabe. In addition, it was unfortunate the RCTs identified for the primary anaysis ony provided sufficient data for a stepwise indirect comparison of the data, i.e. A compared with B compared with C, without the possibiity of drawing further strength and cohesion in the network that a tria of A compared with C woud have provided. To ensure the mode refected the management of unipoar depression in the UK, two cinica advisors (both of whom are practising psychiatrists with experience of managing TRD in either primary or secondary care) were consuted to vaidate mode assumptions throughout the mode deveopment process. Information from experts was obtained through teeconferences and standardised questionnaires. The use of samping to estimate the acute efficacy of each augmentation strategy was necessary because of the paucity of usabe information from the identified cinica effectiveness studies. However, the use of this technique circumvented the need for a consistent definition of response or remission across studies. The most significant imitation to this review was the absence of a direct RCT comparing augmentation of SSRI therapy with ithium to augmentation with an AAP. In addition, the genera paucity of RCTs considering each individua comparator in the required popuation was a significant imitation, particuary with respect to ithium augmentation for which no RCTs were avaiabe in the required popuation, which necessitated the incusion of a surrogate tria. 59 The singe RCT identified comparing ithium augmentation of SSRI to SSRI aone was the main source of uncertainty in the primary cinica anaysis. This is because the sampe size of this tria was sma, with ony 33 patients randomised. In addition, this tria was the ony ink for the stepwise indirect comparison of AAP augmentation and ithium augmentation. Furthermore, the observed withdrawa profie obtained from this sma RCT was counter to prior cinica expectation, in that fewer withdrawas were seen with SSRI + ithium than with SSRI aone. However, this may be a refection of the sma size of the tria; in a sma tria, one or two patients having a different outcome in a study arm can have a dramatic effect on the OR between two study groups. Restriction of the scope of background therapy for which augmentation treatment was considered was a further imitation of this review: consideration of other background therapies may have provided additiona inking studies to enabe the comparison of the treatments of interest. Regarding the economic component to this review, the use of non-standard methodoogy that has not been previousy vaidated was a imitation. This imitation was further enhanced by the uncertainty in the cinica data, which resuted in counterintuitive cost-effectiveness resuts; the treatment associated with the most uncertainty dispayed the most benefit under the samping methodoogy. This is because the threshod of benefit (in terms of MADRS score) is fixed. Therefore, treatments associated with more uncertainty (which are more ikey to achieve a wider range of scores) are more ikey to achieve higher rates of remission and response as a resut of random chance. However, treatments associated with more uncertainty are aso more ikey to have more severe eves of non-response. This uncertainty was not captured in the methodoogy used, which sampes treatment effect and categorises the sampes around a fixed threshod of benefit. 102 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Finay, the paucity of data to inform ong-term patient progression imited the economic modeing of medium to ong-term consequences of each individua therapy. The overa resuts of this review are highy uncertain with regards to which augmentation strategy is most cinicay effective and so which is the most cost-effective. However, the review does highight the requirement for a direct head-to-head cinica tria to assess SSRI + ithium and SSRI + AAP in patients with TRD to enabe more definitive concusions to be drawn. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 103

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Chapter 7 Concusions The resuts of this review support the concusion that augmentation of SSRIs with ithium or an AAP is ikey to be beneficia in peope with TRD, defined as faiure to respond to two or more antidepressants in the current episode of depression. However, based on the imited number of RCTs identified in this research, the cinica evauation suggests there is no statisticay significant difference between the two augmentation strategies. There is a genera paucity of tria data avaiabe in patients with TRD for SSRI + ithium and SSRI + AAP. The cost-effectiveness resuts suggest that augmentation with ithium is cheaper and more effective than augmenting with AAP. However, the resuts are not definitive because the mode is sensitive to the cinica effectiveness parameters of discontinuation and treatment response. The cost-effectiveness of SSRI + ithium and SSRI + AAP wi need to be reconsidered if data from a direct comparison in a RCT or additiona tria data that woud inform the MTC become avaiabe. Impications for service provision The resuts of this research suggest that augmentation of SSRIs with ithium or an AAP is ikey to be beneficia in peope with TRD, defined as faiure to respond to two or more antidepressants in the current episode of depression. However, the overa resuts of this research are highy uncertain with regards to which augmentation strategy is most cinicay effective and so which is ikey to be cost-effective. This report highights the requirement for further research to assess SSRI + ithium and SSRI + AAP in patients with TRD to enabe a more definitive concusion to be drawn. Suggested research priorities A RCT is needed, comparing SSRI + ithium and SSRI + AAP in patients with TRD. The research shoud coect a reevant outcomes, which are response, remission, and discontinuation rates (especiay treatment reated) in the short term, as we as coecting ong-term outcome (e.g. reapse rate) data during maintenance treatment. Adverse events and QoL data shoud aso be prioritised for coection as part of the research. In addition, as part of this research, it has become cear that other antidepressants are aso used as baseine therapy for augmentation in TRD, such as tricycic antidepressants and SNRIs. The current research coud be expanded to incude these treatments. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 105

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Acknowedgements Thanks to the foowing peope for their assistance in the production of this report: Samantha Barton, Heath Technoogy Assessment Anayst, BMJ Technoogy Assessment Group, London, UK. Lii Ng, Deveoper, BMJ Technoogy, London, UK. Professor Phiip J Cowen, MRC Cinica Scientist and Professor of Psychopharmacoogy, Warneford Hospita, Oxford, UK. Dr Luiz Dratcu, Consutant Psychiatrist, South London & Maudsey NHS Foundation Trust, London, UK. Dr Hamish McAister-Wiiams, Reader in Cinica Psychopharmacoogy & Honorary Consutant Psychiatrist, Newcaste University, Newcaste upon Tyne, UK. Dr Andrea L Maizia, Consutant Senior Lecturer (Cinica Psychopharmacoogy and Neurostimuation), University of Bristo, Bristo, UK. Ifigeneia Mavranezoui, Senior Heath Economist, Nationa Coaborating Centre for Menta Heath, University Coege London, London, UK. The Cochrane Coaboration Depression, Anxiety and Neurosis Review Group (CCDAN), University of Bristo, Bristo, UK. Thanks aso to those manufacturers and authors of key trias who responded to requests for additiona data. Contributions of authors Steve Edwards Project ead: supervised the production of the fina report; data vaidation and vaidation of anayses; critica appraisa of the cinica evidence; and critica appraisa of the economic evidence. Victoria Hamiton Critica appraisa of the cinica evidence and drafted the cinica sections. Leo Nherera Critica appraisa of the economic evidence and drafted the economic sections and scientific summary. Nicoa Trevor Critica appraisa of the economic evidence and drafted part of the economic sections, the discussion and overa concusions. A authors read and commented on draft versions of the report. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 107

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 References 1. Jenkins E, Godner EM. Approaches to understanding and addressing treatment-resistant depression: a scoping review. Depres Res Treat 2012. URL: www.hindawi.com/journas/drt/2012/ 469680/ (accessed 23 Apri 2013). 2. Tierney JG. Treatment-resistant depression: managed care considerations. J Manag Care Pharm 2007;13(Supp.):2 7. 3. Nationa Institute for Heath and Care Exceence (NICE). Depression: the treatment and management of depression in aduts. CG23. London: NICE; 2004. 4. Berim MT, Turecki G. What is the meaning of treatment resistant/refractory depression (TRD)? A systematic review of randomized trias. Eur Neuropsychopharmaco 2007;17:696 707. http://dx.doi.org/10.1016/j.euroneuro.2007.03.009 5. Bauer M, Whybrow PC, Angst J, Versiani M, Möer HJ. Word Federation of Societies of Bioogica Psychiatry (WFSBP) guideines for bioogica treatment of unipoar depressive disorders, part 1: acute and continuation treatment of major depressive disorder. Word J Bio Psychiatry 2002;3:5 43. http://dx.doi.org/10.3109/15622970209150599 6. Kupfer DJ. Long-term treatment of depression. J Cin Psychiatry 1991;52(Supp. 5):28 34. http://dx.doi.org/10.1016/s0165-0327(00)00357-8 7. Musseman DL, Evans DL, Nemeroff CB. The reationship of depression to cardiovascuar disease. Arch Gen Psychiatry 1998;55:580 92. http://dx.doi.org/10.1001/archpsyc.55.7.580 8. Bostwick JM, Pankratz VS. Affective disorder and suicide risk: a re-examination. Am J Psychiatry 2000;157:1925 32. http://dx.doi.org/10.1176/appi.ajp.157.12.1925 9. American Psychiatric Association. Diagnostic and Statistica Manua of Menta Disorders (DSM-IV- TR). 4th edn, text revision. Arington State, VA: American Psychiatric Association; 1994. URL: http://dsm.psychiatryonine.org/book.aspx?bookid=22 (accessed 31 May 2012). 10. Word Heath Organization (WHO). Internationa cassification of diseases. URL: www.who.int/cassifications/icd/en/ (accessed 31 May 2012). 11. Neson JC, Papakostas GI. Atypica antipsychotic augmentation in major depressive disorder: a meta-anaysis of pacebo-controed randomized trias. Am J Psychiatry 2009;166:980 91. http://dx.doi.org/10.1176/appi.ajp.2009.09030312 12. Rush AJ, Kraemer HC, Sackeim HA, Fava M, Trivedi MH, Frank E, et a. Report by the ACNP Task Force on response and remission in major depressive disorder. Neuropsychopharmacoogy 2006;31:1841 53. http://dx.doi.org/10.1038/sj.npp.1301131 13. Hamiton M. A rating scae for depression. J Neuro Neurosurg Psychiatry 1960;23:56 62. http://dx.doi.org/10.1136/jnnp.23.1.56 14. Montgomery SA, Åsberg M. A new depression scae designed to be sensitive to change. Br J Psychiatry 1979;134:382 9. http://dx.doi.org/10.1192/bjp.134.4.382 15. Zimmerman M, Posternak MA, Cheminski I. Defining remission on the Montgomery Åsberg Depression Rating Scae. J Cin Psychiatry 2004;65:163 8. http://dx.doi.org/10.4088/jcp.v65n0204 16. Nationa Institute for Heath and Care Exceence (NICE). Depression: the treatment and management of depression in aduts (update). CG90. London: NICE; 2009. 17. Singeton N, Lee A, Metzer H. Psychiatric morbidity among aduts iving in private househods. Technica Report. London: Office for Nationa Statistics; 2007. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 109

REFERENCES 18. Thomas CM, Morris S. Cost of depression among aduts in Engand in 2000. Br J Psychiatry 2003;183:514 19. http://dx.doi.org/10.1192/bjp.183.6.514 19. British Medica Association and Roya Pharmaceutica Society of Great Britain. British nationa formuary. No. 63, March 2012. London: BMA and RPS; 2012. 20. Anderson IM, Ferrier IN, Badwin RC, Cowen PJ, Howard L, Lewis G, et a. Evidence-based guideines for treating depressive disorders with antidepressants: a revision of the 2000 British Association for Psychopharmacoogy guideines. J Psychopharmaco 2008;22:343 96. http://dx.doi.org/10.1177/0269881107088441 21. Greenberg P, Corey-Lise PK, Birnbaum H, Marynchenko M, Caxton A. Economic impications of treatment-resistant depression among empoyees. Pharmacoeconomics 2004;22:363 73. http://dx.doi.org/10.2165/00019053-200422060-00003 22. McCrone P, Dhanasiri S, Pate A, Knapp M, Lawton-Smith S. Paying the price: the cost of menta heath care in Engand to 2026. London: The King s Fund; 2008. 23. Menta Heath Foundation. New Government figures revea the economic burden of depression has risen to 9bn a year. URL: www.mentaheath.org.uk/our-news/news-archive/2010/2010 11 22/? view=standard (accessed May 2012). 24. The NIHR Evauation, Trias and Studies Coordinating Centre (NETSCC), HTA. Management of treatment-resistant depression. www.hta.ac.uk/protocos/201000300001.pdf (accessed May 2012). 25. Edwards S, Hamiton V, Nherera L, Barton S, Trevor N. Lithium or an atypica anti-psychotic in the management of treatment resistant depression: systematic review and economic evauation. PROSPERO 2011:CRD42011001464. URL: www.crd.york.ac.uk/prospero/dispay_record.asp? ID=CRD42011001464 (accessed 31 May 2012). 26. Schuz KF, Atman DG, Moher D, for the CONSORT Group. CONSORT 2010 Statement: updated guideines for reporting parae group randomised trias. Int J Surg 2011;9:672 7. http://dx.doi.org/10.1016/j.ijsu.2011.09.004 27. Higgins JPT, Green S, editors. Cochrane handbook for systematic reviews of interventions. Version 5.1.0 (updated March 2011). The Cochrane Coaboration; 2011. URL: www.cochranehandbook.org (accessed 31 May 2012). 28. Mante N, Haensze W. Statistica aspects of the anaysis of data from retrospective studies of disease. J Nat Cancer Inst 1959;22:719 48. 29. DerSimonian R, Laird N. Meta-anaysis in cinica trias. Contro Cin Trias 1986;7:177 88. http://dx.doi.org/10.1016/0197-2456(86)90046-2 30. Franco MM, Figueira ML, Petrov P, Rihmer Z, Vavrusov L, Dencker-Vansvik E, et a. Quetiapine XR monotherapy and quetiapine XR + ongoing antidepressant vs ithium + ongoing AD for Stage II treatment-resistant major depressive disorder. Eur Neuropsychopharmaco 2010;20(Supp. 3):347. 31. Bauer M, DeOsso L, Kasper S, Pitchot W, Dencker-Vansvik E, Jörgensen L, et a. Quetiapine XR monotherapy, quetiapine XR + ongoing antidepressants and ithium + ongoing antidepressants in patients with treatment-resistant major depressive disorder. Int J Neuropsychopharmaco 2010;13(Supp. 1):143. 32. Montgomery S, Bauer M, De Osso L, Kasper S, Pitchot W, Dencker-Vansvik E, et a. Quetiapine XL monotherapy, quetiapine XL + ongoing antidepressants and ithium + ongoing antidepressants in patients with treatment-resistant major depressive disorder (MDD): a UK perspective. Poster presented at the British Association for Psychopharmacoogy, Harrogate, North Yorkshire, UK, 25 28 Juy 2010. 110 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 33. Ades AE. A chain of evidence with mixed comparisons: modes for mutiparameter synthesis and consistency of evidence. Stat Med 2003;22:2995 3016. 34. Lu G, Ades AE. Combination of direct and indirect evidence in mixed treatment comparisons. Stat Med 2004;23:3105 24. http://dx.doi.org/10.1002/sim.1875 35. Cadwe DM, Ades AE, Higgins JPT. Simutaneous comparison of mutipe treatments: combining direct and indirect evidence. BMJ 2005;331:897 900. http://dx.doi.org/10.1136/bmj.331.7521.897 36. Spiegehater DJ, Best NG, Carin BP, Van Der Linde A. Bayesian measures of mode compexity and fit. J Roy Statist Soc B 2002;64:583 639. http://dx.doi.org/10.1111/1467-9868.00353 37. Dempster AP. The direct use of ikeihood for significance testing. Stat Comp 1997;7:247 52. 38. Higgins JP, Thompson SG, Deeks JJ, Atman DG. Measuring inconsistency in meta-anayses. BMJ 2003;327:557 60. http://dx.doi.org/10.1136/bmj.327.7414.557 39. Edwards SJ, Carke MJ, Wordsworth S, Borri J. Indirect comparisons of treatments based on systematic reviews of randomised controed trias. Int J Cin Pract 2009;63:841 54. http://dx.doi.org/10.1111/j.1742-1241.2009.02072.x 40. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-anaysis detected by a simpe, graphica test. BMJ 1997;315:629 34. http://dx.doi.org/10.1136/bmj.315.7109.629 41. Sterne JAC, Egger M. Funne pots for detecting bias in meta-anaysis: guideines on choice of axis. J Cin Epidemio 2001;54:1046 55. 42. Bin W, Mei H, Po L. Fuoxetine and oanzapine in the treatment of refractory depression. Shandong Arch Psychiatry 2006;19:87 9. 43. Feng H, Meifang C, Yunhai T. Efficacy of oanzapine and fuoxetine combination therapy in treatment-resistant depression. Zheijang Pract Med 2008;13:117 18. 44. Li H, Zhang Y, Zhang Y. Doube-bind study of fuoxetine augmented with oanzapine in the treatment of treatment-resistant depression. Shandong Arch Psychiatry 2006;19:85 6. 45. Wang XH, Guo X. Effectiveness and safety of oanzapine combined with fuoxetine for refractory depression: a systematic review. Chin J Evid Med 2010;10:1102 9. 46. Mattingy G, Iivicky H, Canae J, Anderson R. Quetiapine combination for treatment-resistant depression. Annua Meeting of the American Psychiatric Association, New Research Abstracts (Toronto, Canada, 20 25 May, 2006). Washington, DC: APA; 2006. 47. Berman RM, Marcus RN, Swanink R, McQuade RD, Carson WH, Corey-Lise PK, et a. The efficacy and safety of aripiprazoe as adjunctive therapy in major depressive disorder: a muticenter, randomized, doube-bind, pacebo-controed study. J Cin Psychiatry 2007;68:843 53. http://dx.doi.org/10.4088/jcp.v68n0604 48. Berman RM, Fava M, Thase ME, Trivedi MH, Swanink R, McQuade RD, et a. Aripiprazoe augmentation in major depressive disorder: a doube-bind, pacebo-controed study in patients with inadequate response to antidepressants. CNS Spectr 2009;14:197 206. 49. Corya SA, Wiiamson D, Sanger TM, Briggs SD, Case M, Toefson G, et a. A randomized, doubebind comparison of oanzapine/fuoxetine combination, oanzapine, fuoxetine, and venafaxine in treatment-resistant depression. Depress Anxiety 2006;23:364 72. http://dx.doi.org/10.1002/ da.20130 50. Marcus RN, McQuade RD, Carson WH, Hennicken D, Fava M, Simon JS, et a. The efficacy and safety of aripiprazoe as adjunctive therapy in major depressive disorder: a second muticenter, randomized, doube-bind, pacebo-controed study. J Cin Psychopharmaco 2008;28:156 65. http://dx.doi.org/10.1097/jcp.0b013e31816774f9 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 111

REFERENCES 51. Sheton RC, Toefson GD, Tohen M, Stah S, Gannon KS, Jacobs TG, et a. A nove augmentation strategy for treating resistant major depression. Am J Psychiatry 2001;158:131 4. http://dx.doi.org/ 10.1176/appi.ajp.158.1.131 52. Sheton RC, Wiiamson DJ, Corya SA, Sanger TM, Van Campen LE, Case M, et a. Oanzapine/ fuoxetine combination for treatment-resistant depression: a controed study of SSRI and nortriptyine resistance. J Cin Psychiatry 2005;66:1289 97. http://dx.doi.org/10.4088/ JCP.v66n1012 53. Thase ME, Corya SA, Osuntokun O, Case M, Heney DB, Snager TM, et a. A randomized, doube-bind comparison of oanzapine/fuoxetine combination, oanzapine, and fuoxetine in treatment-resistant major depressive disorder. J Cin Psychiatry 2007;68:224 36. http://dx.doi.org/ 10.4088/JCP.v68n0207 54. Sheton RC, Tohen M, Stah S, Jacobs T, Gannon KS, Toefson GD. The study of oanzapine pus fuoxetine in treatment-resistant major depressive disorder without psychotic features. Schizophr Res 2000;1:Specia Issue. http://dx.doi.org/10.1016/s0920-9964(00)90772-2 55. Tohen M, Sheton R, Toefson GD, Stah S, Jacobs T, Gannon KS, et a. Oanzapine pus fuoxetine: doube-bind and open-abe resuts in treatment-resistant major depressive disorder. Eur Neuropsychopharmaco 1999;9(Supp. 5):S246. http://dx.doi.org/10.1016/s0924-977x(99)80224-7 56. Toefson GD, Sheton R, Tohen M, Stah S, Jacobs T, Buras W, et a. Efficacy of oanzapine, fuoxetine and combination therapy in treatment-resistant major depressive disorder without psychotic features. 11th European Coege of Neuropsychopharmacoogy Congress, Paris, France, 31 October to 4 November 1998. 57. Dunner DL, Amsterdam JD, Sheton RC, Loebe A, Romano SJ. Efficacy and toerabiity of adjunctive ziprasidone in treatment-resistant depression: a randomized, open-abe, piot study. J Cin Psychiatry 2007;68:1071 7. http://dx.doi.org/10.4088/jcp.v68n0714 58. CinicaTrias.gov. Ziprasidone augmentation of SSRIs for patients with major depressive disorder (MDD) that do not sufficienty respond to treatment with SSRIs. URL: http://cinicatrias.gov/ct2/ show/nct00633399 (accessed May 2012). 59. Katona CL, Abou-Saeh MT, Harrison DA, Nairac BA, Edwards DR, Lock T, et a. Pacebo-controed tria of ithium augmentation of fuoxetine and ofepramine. Br J Psychiatry 1995;166:80 6. http://dx.doi.org/10.1192/bjp.166.1.80 60. Crossey NA, Bauer M. Acceeration and augmentation of antidepressants with ithium for depressive disorders: two meta-anayses of randomized, pacebo-controed trias. J Cin Psychiatry 2007;68:935 40. http://dx.doi.org/10.4088/jcp.v68n0617 61. Soares HP, Danies S, Kumar A, Carke M, Scott C, Swann S, et a. Bad reporting does not mean bad methods for randomised trias: observationa study of randomised controed trias performed by the Radiation Therapy Oncoogy Group. BMJ 2004;328:22 5. 62. Thase ME, Trivedi MH, Neson JC, Fava M, Swanick R, Tran QV, et a. Examining the efficacy of adjunctive aripiprazoe in major depressive disorder: a pooed anaysis of 2 studies. J Cin Psychiatry 2008;10:440 7. 63. Abrams K. Interpreting tau: OR Scae. Presented at the Indirect and Mixed Treatment Comparisons Course, University of Leicester, UK, 19 21 September 2007. 64. Zimmerman M, Posternak A, Cheminski I. Derivation of a definition of remission on the Montgomery Åsberg depression rating scae corresponding to the definition of remission on the Hamiton rating scae for depression. J Psychiatr Res 2004;38:577 82. http://dx.doi.org/10.1016/ j.jpsychires.2004.03.007 112 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 65. Bauer M, Dopfmer S. Lithium augmentation in treatment-resistant depression: meta-anaysis of pacebo-controed studies. J Cin Psychopharmaco 1999;19:427 34. http://dx.doi.org/10.1097/ 00004714-199910000-00006 66. Bauer M, Adi M, Bschor T, Pihatsch M, Pfennig A, Sasse J, et a. Lithium s emerging roe in the treatment of refractory major depressive episodes: augmentation of antidepressants. Neuropsychobioogy 2010;62:36 42. http://dx.doi.org/10.1159/000314308 67. Mahmoud RA, Pandina G, Turkoz I, Kosik-Gonzaez C, Canuso CM, Kujawa MJ, et a. Risperidone for treatment-refractory major depressive disorder. Ann Intern Med 2007;147:593 602. http://dx.doi.org/10.7326/0003-4819-147-9-200711060-00003 68. Reeves H, Batra S, May RS, Zhang R, Dah DC, Li X. Efficacy of risperidone augmentation to antidepressant in the management of suicidaity in major depressive disorder: a randomized, doube-bind, pacebo controed piot study. J Cin Psychiatry 2008;69:1228 336. 69. Keitner GI, Garow SJ, Ryan CE, Ninan PT, Soomon DA, Nemeroff CB, et a. A randomized, pacebo-controed tria of risperidone augmentation for patients with difficut-to-treat unipoar, non-psychotic major depression. J Psychiatr Res 2009;43:205 14. http://dx.doi.org/10.1016/j. jpsychires.2008.05.003 70. Khuar A, Chokka P, Fuerton D, McKenna S, Backman A. A doube-bind, randomized, pacebocontroed study of quetiapine as augmentation therapy to SSRI/SNRI agents in the treatment of non-psychotic unipoar depression with residua symptoms, in American Psychiatric Association 2006 Annua Meeting: New Research Abstracts (Toronto, Canada, May 20 25, 2006). Washington, DC: APA; 2006. 71. McIntyre A, Gendron A, McIntyre A. Quetiapine adjunct to seective serotonin reuptake inhibitors or venafaxine in patients with major depression, comorbid anxiety, and residua depressive symptoms: a randomized, pacebo-controed piot study. Depress Anxiety 2007;24:487 94. http://dx. doi.org/10.1002/da.20275 72. Earey W, McIntyre A, Bauer M, Pretorius HW, Sheton R, Lindgren P, et a. Efficacy and toerabiity of extended reease quetiapine fumarate (quetiapine extended reease) as add-on to antidepressants in patients with major depressive disorder (MDD): resuts from a doube-bind, randomized, phase III study. American Coege of Neuropsychopharmacoogy 2007 Annua Meeting Abstracts (Boca Raton, FL, 9 13 December 2007). Nashvie, TN: ACNP; 2007. 73. E-Khaii N, Joyce M, Atkinson S, Buynak R, Datto C, Lindgren P, et a. Adjunctive extended-reease quetiapine fumarate (quetiapine-extended reease) in patients with major depressive disorder and inadequate antidepressant response. American Psychiatric Association 2008 Annua Meeting: New Research Abstracts (Washington, DC, 3 8 May 2008). Washington, DC: APA; 2008. 74. Heninger GR, Charney DS, Sternberg DE. Lithium carbonate augmentation of antidepressant treatment. An effective prescription for treatment-refractory depression. Arch Gen Psychiatry 1983;40:1335 42. http://dx.doi.org/10.1001/archpsyc.1983.01790110077013 75. Kantor D, McNevin S, Leichner P, Harper D, Krenn M. The benefit of ithium carbonate adjunct in refractory depression: fact or fiction? Can J Psychiatry 1986;31:416 18. 76. Zusky PM, Biederman J, Rosenbaum JF, Manschreck TC, Gross CC, Weiberg JB, et a. Adjunct ow-dose ithium carbonate in treatment-resistant depression: a pacebo-controed study. J Cin Psychopharmaco 1988;8:120 4. http://dx.doi.org/10.1097/00004714-198804000-00007 77. Schöpf J, Baumann P, Lemarchand T, Rey M. Treatment of endogenous depressions resistant to tricycic antidepressants or reated drugs by ithium addition. Resuts of a pacebo-controed doubebind study. Pharmacopsychiatry 1989;22:183 7. http://dx.doi.org/10.1055/s-2007-1014603 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 113

REFERENCES 78. Browne M, Lapierre YD, Hrdina PD, Horn E. Lithium as an adjunct in the treatment of major depression. Int Cin Psychopharmaco 1990;5:103 10. http://dx.doi.org/10.1097/00004850-199004000-00004 79. Joffe RT, Singer W, Levitt AJ, MacDonad C. A pacebo-controed comparison of ithium and triiodothyronine augmentation of tricycic antidepressants in unipoar refractory depression. Arch Gen Psychiatry 1993;50:387 93. http://dx.doi.org/10.1001/archpsyc.1993.01820170065008 80. Stein G, Bernadt M. Lithium augmentation therapy in tricycic-resistant depression. A controed tria using ithium in ow and norma doses. Br J Psychiatry 1993;162:634 40. http://dx.doi.org/10.1192/ bjp.162.5.634 81. Nierenberg AA, Papakostas GI, Petersen T, Montoya HD, Worthington JJ, Tedow J, et a. Lithium augmentation of nortriptyine for subjects resistant to mutipe antidepressants. J Cin Psychopharmaco 2003;23:92 5. http://dx.doi.org/10.1097/00004714-200302000-00013 82. Baumann P, Ni R, Souche A, Montadi S, Baettig D, Lambert S, et a. A doube-bind, pacebocontroed study of citaopram with and without ithium in the treatment of therapy-resistant depressive patients: a cinica, pharmacokinetic, and pharmacogenetic investigation. J Cin Psychopharmaco 1996;16:307 14. http://dx.doi.org/10.1097/00004714-199608000-00006 83. Duru G, Fantino B. The cinica reevance of changes in the Montgomery Åsberg Depression Rating Scae using the minimum cinicay important difference approach. Curr Med Res Opin 2008;24:1329 35. http://dx.doi.org/10.1185/030079908x291958 84. Haynes RB, Wiczynski N, McKibbon KA, Waker CJ, Sincair JC. Designing optima search strategies for detecting cinicay sound studies in MEDLINE. J Am Med Inform Assoc 1994;1:447 59. 85. Dickersin K, Scherer R, Lefebvre C. Identifying reevant studies for systematic reviews. BMJ 1994;309:1286 91. http://dx.doi.org/10.1136/bmj.309.6964.1286 86. Benedict AA. Economic evauation of duoxetine vs serotonin seective reuptake inhibitors and venafaxine XR in treating major depressive disorder in Scotand. Cost-utiity comparison of escitaopram and sertraine in the treatment of major depressive disorder. J Affect Disord 2010;120:94 104. 87. Xie F, Despiege N, Danchenko N, Hansen K. Cost effectiveness anaysis of escitaopram compared to venafaxine and fuvoxamine in treatment of major depressive disorder. Int J Psychiatry Cin Pract 2009;13:59 69. http://dx.doi.org/10.1080/13651500802450506 88. Leeahanaj T. The cost-effectiveness of aripiprazoe as adjunctive therapy in major depressive disorder: Thai economic mode. J Med Assoc Thai 2010;93(Supp. 6):43 50. 89. Simpson KN, Wech MJ, Koze FA, Demitrack MA, Nahas Z. Cost-effectiveness of transcrania magnetic stimuation in the treatment of major depression: a heath economics anaysis. Adv Ther 2009;26:346 68. http://dx.doi.org/10.1007/s12325-009-0013-x 90. Phiips Z, Ginney L, Scupher M, Caxton K, Goder S, Riemsma R, et a. Review of guideines for good practice in decision-anaytic modeing in heath technoogy assessment. Heath Techno Assess 2004;8(36). http://dx.doi.org/10.2165/00019053-200624040-00006 91. Revicki AD, Wood M. Patient-assigned heath state utiities for depression-reated outcomes: differences by depression severity and antidepressant medications. J Affect Dis 1998;48:25 36. http://dx.doi.org/10.1016/s0165-0327(97)00117-1 92. Ministry of Heath. Cinica practice guideines: depression. Singapore: Ministry of Heath; 2004. 93. Montgomery SA, Andersen HF. Escitaopram vs venafaxine XR in the treatment of depression. Int Cin Psychopharmaco 2006;21:297 309. 114 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 94. Lancon C, Verpiat P, Annemans L, Despiege N, Francois C. Escitaopram in major depressive disorder: cinica benefits and cost effectiveness vs citaopram. Int J Psychiatry Cin Pract 2007;11:44 52. 95. Posternak MA, Zimmerman M. Switching vs augmentation, a perspective naturaist comparison in depressed treatment resistant patients. J Cin Psychiatry 2001;62:135 42. 96. Adi M, Bschor T, Lucka C, Lewitza U, Ising M, Uhr M, et a. Long-term outcome after ithium augmentation in unipoar depression: focus on HPA system activity. Neuropsychoogy 2009; 60:23 30. http://dx.doi.org/10.1159/000234814 97. Cegedim Strategic Data UK. March 2005. Psychotrak, HMSL Ltd, 2005. 98. House NA, Sommi RW. Defining response and attaining therapeutic goas for depression. J Pharm Pract 2001:14:453 7. http://dx.doi.org/10.1177/089719001129040946 99. Rush JA, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, et a. Acute and onger-term outcomes in depressed outpatients requiring one or severa treatment steps: a STAR*D report. Am J Psychiatry 2006;163:1905 17. 100. Gaynes BN, Rush AJ, Trivedi MH, Wisniewski SR, Spencer D, Fava M. The STAR*D study: treating depression in the rea word. Ceve Cin J Med 2008;75:57 66. 101. Papaioannou D, Brazier JE, Paisey S. NICE DSU technica support document 9. The identification, review and synthesis of heath state utiity vaues from the iterature. 2011. URL: www. nicedsu.org.uk (accessed 23 Apri 2013). 102. Bennett KJ, Torrance GW, Boye MH, Guscott R. Cost-utiity anaysis in depression: the McSad utiity measure for depression heath states. Psychiatr Serv 2000;51:1171 6. http://dx.doi.org/10.1176/ appi.ps.51.9.1171 103. Schaffer A, Levitt AJ, Hershkop SK, Oh P, MacDonad C, Lanctot K. Utiity scores of symptom profies in major depression. Psychiatry Res 2002;110:189 97. http://dx.doi.org/10.1016/s0165-1781(02)00097-5 104. Pyne J, Sieber W, David K, Kapan RM, Hyman Rapaport M, Keith Wiiams D. Use of the quaity of we-being sef-administered version (QWB-SA) in assessing heath-reated quaity of ife in depressed patients. J Affect Disord 2003;76:237 47. http://dx.doi.org/10.1016/s0165-0327(03) 00106-X 105. Sapin C, Fantino B, Nowicki ML, Kind P. Usefuness of EQ-5D in assessing heath status in primary care patients with major depressive disorder. Heath Qua Life Outcomes 2004;2:20. 106. Andresen E, Rothenberg R, Kapan R. Performance of a sef-administered maied version of the Quaity of We-Being (QWB-SA) questionnaire among oder aduts. Med Care 1998;36:1349 60. http://dx.doi.org/10.1097/00005650-199809000-00007 107. Doan P, Gudex C, Kind P, Wiiams A. The time trade-off method: resuts from a genera popuation study. Heath Econ 1996;5:141 54. http://dx.doi.org/10.1002/(sici)1099-1050(199603)5:2<141:: AID-HEC189>3.0.CO;2-N 108. Nationa Institute for Heath and Care Exceence (NICE). Guide to the methods of technoogy appraisa. June 2008. URL: www.nice.org.uk/media/b52/a7/tamethodsguideupdatedjune2008. pdf (accessed Apri 2012). 109. Sma GW, Birkett M, Meyers BS, Koran LM, Bystitsky A, Nemeroff CB. Impact of physica iness on quaity of ife and antidepressant response in geriatric major depression. J Am Geriatr Soc 1996;4:1220 5. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 115

REFERENCES 110. Dunner DL, Rush AJ, Russe JM, Burke M, Woodard S, Wingard P, et a. Prospective, ong-term, muticenter study of the naturaistic outcomes of patients with treatment-resistant depression. J Cin Psychiatry 2006;67:688 95. 111. Bradey RH, Barkin RL, Jerome J, DeYoung K, Dodge CW. Efficacy of venafaxine for the ong term treatment of chronic pain with associated major depressive disorder. Am J Ther 2003;10:318 23. http://dx.doi.org/10.1097/00045391-200309000-00003 112. Doraiswamy PM, Khan ZM, Donahue RM, Richard NE. Quaity of ife in geriatric depression: a comparison of remitters, partia responders, and nonresponders. Am J Geriatr Psychiatry 2001;9:423 8. http://dx.doi.org/10.1176/appi.ajgp.9.4.423 113. Trivedi MH, Pigotti TA, Perera P, Diingham KE, Carfango ML, Pitts CD. Effectiveness of ow doses of paroxetine controed reease in the treatment of major depressive disorder. J Cin Psychiatry 2004;65:1356 64. http://dx.doi.org/10.4088/jcp.v65n1010 114. Trivedi MH, Dunner DL, Kornstein SG, Thase ME, Zajecka JM, Rothschid AJ, et a. Psychosocia outcomes in patients with recurrent major depressive disorder during 2 years of maintenance treatment with venafaxine extended reease. J Affect Disord 2010;126:420 9. 115. Carta MG, Hardoy MC, Piu A, Sorba M, Foris AL, Mannu FA, et a. Improving physica quaity of ife with group physica activity in the adjunctive treatment of major depressive disorder. Cin Pract Epidemio Menta Heath 2008;4:1. http://dx.doi.org/10.1186/1745-0179-4-1 116. Karp JF, Skidmore E, Lotz M, Lenze E, Dew MA, Raynods CF. Use of the ate-ife function and disabiity instrument to assess disabiity in major depression. J Am Geriatr Soc 2009;57:1612 19. http://dx.doi.org/10.1111/j.1532-5415.2009.02398.x 117. Feck MPM. Efficacy of minacipran in outpatients experiencing major depression non respondent to SSRIs: a 12-week open study. Rev Psiquiatr Cin 2010;37:241 50. 118. Barge-Schaapved DQ, Nicoson NA, van der Hoop RG, De Vries MW. Changes in daiy ife experience associated with cinica improvement in depression. J Affect Disord 1995;34:139 54. http://dx.doi.org/10.1016/0165-0327(95)00012-c 119. Mitche AJ, Semes T. Why don t patients take their medicine? Reasons and soutions in psychiatry. Adv Psychiatr Treat 2007;13:336 46. http://dx.doi.org/10.1192/apt.bp.106.003194 120. Frank E, Prien RF, Jarret RB, Keer MB, Kupfer DJ, Lavori PW, et a. Conceptuaization and rationae for consensus definitions of terms in major depressive disorder. Arch Gen Psychiatry 1991;48:851 5. http://dx.doi.org/10.1001/archpsyc.1991.01810330075011 121. Briggs AHE, Caxton K, Scupher M. Decision modeing for heath economic evauation. Oxford: Oxford University Press; 2006. 122. Fekadu A, Wooderson SC, Rane LJ, Markopouou K, Poon L, Ceare AJ. Long-term impact of residua symptoms in treatment-resistant depression. Can J Psychiatry 2011;56:549 57. 123. Nationa Institute for Heath and Care Exceence (NICE). Bipoar disorder: the management of bipoar disorder in aduts, chidren and adoescents, in primary and secondary care. London: NICE; 2006. 124. Department of Heath (DoH). Reference costs 2010 11. London. URL: www.dh.gov.uk/en/ Pubicationsandstatistics/Pubications/PubicationsPoicyAndGuidance/DH_123459 (accessed Apri 2012). 125. Curtis L, Netten A. Unit costs of heath and socia care 2011. University of Kent, Canterbury: PSSRU; 2010. 126. Hospita Episode Statistics (HES) URL: www.hesonine.nhs.uk/ease/servet/contentserver? siteid=1937&categoryid=202 (accessed 31 May 2012). 116 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 127. Meads CA, Cnossen JS, Meher S, Juarez-Garcia A, ter Riet G, Duey L, et a. Methods of prediction and prevention of pre-ecampsia: systematic reviews of accuracy and effectiveness iterature with economic modeing. Heath Techno Assess 2008;12(6). 128. Drug Patent Watch. ABILIFY (aripiprazoe). Otsuka Pharmaceuticas. URL: www.drugpatentwatch. com/utimate/preview/tradename/index.php?query=abilify (accessed August 2012). 129. Kroenke K, West SL, Swinde R, Gisenan A, Eckert GJ, Door R, et a. Simiar effectiveness of paroxetine, fuoxetine, and sertraine in primary care: a randomized tria. JAMA 2001;286:2947 55. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 117

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 1 Fina protoco HTA no. 10/301: Management of treatment resistant depression, PROTOCOL, June 2011 1. Tite of the project Lithium or an atypica antipsychotic in the management of treatment resistant depression: systematic review and economic evauation 2. Name of TAR team and project ead BMJ Technoogy Assessment Group (BMJ-TAG), BMJ Evidence Centre, BMJ Group, London Dr Steven J Edwards Head of Heath Technoogy Assessment BMJ Technoogy Assessment Group (BMJ-TAG) BMJ Evidence Centre BMJ Group, BMA House Tavistock Square London WC1H 9JP Te: +44 (0) 207 383 6112 Mob: +44 (0) 776 823 7218 Fax: +44 (0) 207 383 6242 Emai: SEdwards@BMJGroup.com 3. Pain Engish summary Depression is a common menta disorder that presents with depressed mood, oss of interest or peasure, feeings of guit or ow sef-worth, disturbed seep or appetite, ow energy and poor concentration. 1 Depression can be categorised into two broad categories; unipoar depression and bipoar depression. Peope with unipoar depression suffer with ony episodes of depression, whereas peope with bipoar depression suffer with episodes of ow mood, and abnormay eevated mood (aso known as mania). The most common mood disorder is unipoar depression and because the pharmacoogica treatment of unipoar and bipoar depression are somewhat different we wi be focusing on the peope with unipoar depression in this report. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 119

APPENDIX 1 Depression may be treated with medication known as antidepressants, various kinds of psychoogica treatments or sef-hep measures. There are ots of different antidepressant medications avaiabe and so if someone does not get better with their first treatment a different one may be tried. This report wi focus on peope who have unipoar depression and who have not responded to treatment with at east two previous antidepressant medications; we refer to these peope as having treatment resistant unipoar depression. In peope with treatment resistant depression it is thought that the addition of another medication such as ithium or an atypica antipsychotic drug coud offer some benefit; however there is imited evidence directy comparing ithium and atypica antipsychotics in peope with treatment resistant unipoar depression. 2 The aim of this report is to identify how effective adding either ithium or an atypica antipsychotic medication to an antidepressant is at managing peope with treatment resistant unipoar depression. We aso aim to perform an economic anaysis to see how cost-effective these medications are when used to treat depression. 4. Decision probem Background Depression is a common menta disorder affecting about 121 miion peope wordwide and is among the eading causes of disabiity. 1 Peope presenting with depression may compain of depressed mood, oss of interest or peasure, feeings of guit or ow sef-worth, disturbed seep or appetite, ow energy and poor concentration. Depression can be diagnosed cinicay using different criteria, the most commony used of which are the DSM IV criteria as pubished by the American Psychiatric Association and the ICD 10 criteria deveoped by the Word Heath Organisation. 3,4 Up to two thirds of patients with major depression wi either not respond to or wi have a sub-optima response to first-ine treatment with antidepressants (i.e. they may respond but not enter remission which is the reative absence of cinica symptomatoogy). There are severa potentia pharmacoogica treatment options for patients not achieving sufficient response with antidepressants, one of which is to augment the antidepressant with an agent not approved for use as monotherapy in major depressive disorder. 5 Current NICE guidance 2 for the sequencing of treatments in depression after an inadequate response to at east one antidepressant recommends that peope who are informed about and prepared to toerate the increased side-effect burden, shoud be considered for treatment with the combination or augmentation of an antidepressant with ithium or an antipsychotic such as aripiprazoe, oanzapine, quetiapine or risperidone or another antidepressant such as mirtazapine or mianserin. Objective This report aims to determine the cinica effectiveness and cost-effectiveness of SSRI antidepressant therapy with either ithium or an atypica antipsychotic in the management of peope with treatment resistant unipoar depression. For this review, treatment resistant depression wi be defined as faiure to respond to at east two previous antidepressant medications. We wi not impose restrictions on the maximum number of previous antidepressant drugs aowed so as not to reduce the amount of data avaiabe for anaysis as we aware that 120 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 there wi be imited reevant SSRI RCT data avaiabe. This assumes that there is a consistent reative treatment effect independent of ine of therapy, i.e. addition of an atypica or ithium has the same reative benefit whether given with third-ine SSRI or fourth ine SSRI, etc. However, a sensitivity anaysis wi be conducted to assess the impact of this assumption. PICO criteria The panned PICO is as foows: Popuation: Aduts with treatment resistant unipoar depression defined as faiure to respond to at east two previous antidepressants in the current episode of depression ony. Intervention: An SSRI (seective serotonin reuptake inhibitor) (defined as either Citaopram (Ciprami), Escitaopram (Cipraex), Fuoxetine (Prozac, Feicium, Prozep, Prozit), Fuvoxamine (Faverin), Paroxetine (Seroxat) or Sertraine (Lustra)), PLUS An atypica antipsychotic drug (defined as either Amisupride (Soian), Aripiprazoe (Abiify), Cozapine (Cozari, Denzapine, Zaponex), Oanzapine (Zyprexa, Zypadhera), Paiperidone (Invega), Quetiapine (Seroque), Risperidone (Risperda) or Ziprasidone (Geodon)) Comparator: An SSRI (defined as either Citaopram (Ciprami), Escitaopram (Cipraex), Fuoxetine (Prozac, Feicium, Prozep, Prozit), Fuvoxamine (Faverin), Paroxetine (Seroxat) or Sertraine (Lustra)) PLUS Lithium (Lithium carbonate (Camcoit, Liskonum, Priade) or Lithium citrate (Li-Liquid, Priade) or Lithium (Litarex, Lithonate, Phasa)) Outcomes: Disease severity Quaity of ife Adverse effects Withdrawas (a cause) as a surrogate outcome for adherence to medication Reapse rate Mortaity Cost-effectiveness Subgroup anayses The panned subgroup anayses are as foows: Different durations of depression (i.e. time since first onset of current episode of depression) Casses of previous antidepressants (e.g. SSRI or tricycic antidepressant) Sex (i.e. maes and femaes) Age (i.e. those < 75 years and those 75 years od) Peope with different severities of depression (i.e. based on tria entry Hamiton Depression Rating Scae rating) Objectives The key areas that we pan to address in this report are: To identify and review the existing evidence reating to the cinica outcomes as pre-specified above To report the cost-effectiveness of these treatments To identify what the potentia areas for future research might be Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 121

APPENDIX 1 5. Report methods for synthesis of evidence of cinica effectiveness A review of the evidence for cinica effectiveness wi be undertaken systematicay foowing the genera principes recommended in the PRISMA statement (formery the QUOROM statement). 6 Search strategy The search strategy wi comprise the foowing main eements: 1. Searching of eectronic bibiographic databases 2. Contact with cinica experts in the fied 3. Review of the reference ists of retrieved papers 1. The eectronic databases that wi be searched are EMBASE, MEDLINE, PsycINFO and the Cochrane Controed Trias Register. We wi aso search the CinicaTrias.gov website to identify reevant ongoing cinica trias that when competed may have an impact on the resuts of this review, to assist us in drawing up our fina recommendations. 2. We wi contact cinica experts in the reevant therapy areas to request detais of trias (pubished and unpubished) of which they may be aware. We wi aow the experts 1 caendar month to provide an initia response, with any additiona time aowed being dependent on whether we have reached the data anaysis stage of the review. 3. The references from any reevant review papers or randomised controed trias (RCTs) uncovered in the search wi aso be examined for additiona references potentiay reevant to the review. Abstract appraisa Tites and abstracts of studies identified by the search process wi be assessed independenty by two reviewers (VH and SB) for incusion. In cases where the reviewers are unabe to reach a consensus as to whether the fu text shoud be obtained for further appraisa, the fu text wi be obtained. When potentiay reevant data are avaiabe in ony an abstract format then we wi attempt to contact the corresponding author in order to obtain the fu pubication; however, there wi be a pre-specified deadine of 1 caendar month by which they wi need to have contacted us, but we may aow additiona time for them to suppy the data requested depending on where we are in the review process. Any information suppied after the deadine wi be incuded in ony the discussion section of the review report. Incusion criteria For the review of cinica effectiveness, ony RCTs wi be incuded Aduts 18 years Peope with unipoar depression ony Treatment resistant depression defined as faiure to respond to at east two previous antidepressants in the current episode of depression ony SSRI (seective serotonin reuptake inhibitor) given as baseine treatment and patient randomised to either ithium or an atypica antipsychotic Minimum duration of 4 weeks treatment with study medication for the current episode of depression Studies reporting on one or more of the foowing outcomes: Disease severity Quaity of ife 122 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Adverse effects Adherence to medication or withdrawas (a cause) Reapse rate Mortaity Cost-effectiveness Excusion criteria Non-randomised studies Narrative reviews, editorias, opinions Studies performed in animas Studies not focusing on the treatment of the acute phase of depression (i.e. those ony focusing soey on maintenance therapy) Bipoar depression or bipoar disorder diagnosis prior to study entry Underying medica condition or another substantia co-morbid psychiatric condition Trias reporting ony post-crossover resuts Study incusion assessment Two reviewers (VH and SB) wi independenty assess for incusion the fu text of the trias identified during the abstract assessment stage and any differences in opinion wi be arbitrated by a third reviewer (SJE). Data extraction strategy Data wi be extracted by one reviewer (VH) using a standardised data extraction form (for draft copy of data coection form, pease see appendix 10.2) and vaidated by second reviewer (SB). A pragmatic decision for data vaidation wi be made depending on the number of trias identified owing to the time constraints for competing this review. If a arge number of trias are identified then a data wi be vaidated (checked) by a second reviewer, with a sampe being fuy independenty data extracted. This sampe wi be 25% or a minimum of 5 papers (whichever is arger). The Data Extraction Form wi be piot tested on a sampe of three papers by the reviewers and a fina version agreed. Discrepancies in the data extracted by the two reviewers wi be resoved through discussion, with invovement of a third reviewer (SJE) if necessary. Data from intention-to-treat (ITT) anayses wi be extracted (per protoco (PP) data wi aso be extracted for use in a sensitivity anaysis). Shoud a tria not report ITT data, we wi treat missing data as treatment faiures to aow our anaysis to conform to an ITT anaysis. For the purpose of this review, ITT wi be defined as patients being anaysed in the treatment group they were aocated to at randomisation regardess of whether they received the wrong intervention, withdrew or were ost to foow-up. Study authors wi be contacted to suppy any additiona information not incuded in pubished sources (incuding pre-crossover resuts in those trias reporting ony post-crossover resuts) and there wi be a pre-specified deadine by which we woud require a response. The deadine wi be 1 caendar month from the date of sending the request by which time they must have contacted us with at east an initia response acknowedging their intent to suppy some of the information required. We may aow additiona time for them to suppy the data requested depending on where we are in the review process, however any information received after the deadine wi be incuded in ony the discussion section of the review. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 123

APPENDIX 1 Quaity assessment strategy Outcomes from the studies that meet the incusion criteria wi be assessed using the updated risk of bias too deveoped by the Cochrane Coaboration (March 2011). 7 These criteria assess the foowing areas: 1. Random sequence generation 2. Aocation conceament 3. Binding of participants and personne 4. Binding of outcomes assessment 5. Incompete outcome data 6. Seective reporting 7. Other bias Based on these criteria, an assessment for each outcome reported in the tria wi be aocated based on the identified risk of bias. The three bias assessment categories used wi be: ow risk, high risk and uncear risk. Uncear risk is ikey to be assigned owing to poor reporting of how the tria was conducted rather than a poory conducted tria. 8 Trias that are deemed to be at ow or uncear risk of bias wi be incuded in the main anaysis; however, the trias rated high risk wi be incuded in a sensitivity anaysis. Two reviewers (VH and SB) wi independenty rate the tria outcomes for incusion and any differences in opinion wi be arbitrated by a third reviewer (SJE). An outcome from an RCT wi be considered appropriate for incusion uness the tria demonstrates some feature that necessitates the excusion of that outcome. Methods of anaysis/synthesis Data wi be tabuated and, where appropriate, meta-anaysis wi be empoyed to estimate a summary measure of effect on reevant outcomes based on ITT anayses (with a sensitivity anaysis based on per protoco data). We wi not be assuming there is a cass effect for any of the drugs incuded and so each individua drug wi be considered separatey in the review, i.e. each SSRI and atypica antipsychotic or ithium combination wi form separate anayses. Standard pairwise meta-anaysis wi be conducted when more than one tria is identified for incusion for any pair of treatments under investigation. This wi be carried out using a fixed effects mode with the Mante-Haensze method. 9 Sensitivity anaysis wi be conducted using a random effects mode with the DerSimonian & Laird method. 10 It is anticipated that a mixed treatment comparison (MTC; aso caed a mutipe treatment meta-anaysis and network meta-anaysis) wi need to be conducted to estimate the effects of the different treatments incuded in the research. A MTC can be seen as an extension of traditiona pairwise meta-anaysis. 11 13 It has advantages over standard pairwise meta-anaysis as it is based on a network of connected trias where a new tria may enter the network if it is in a cinicay comparabe patient popuation, has a simiar design to other trias incorporated in the network and contains at east one treatment that aready exists within the network. It has been argued that this underying assumption of exchangeabiity of data is no different from the practice within standard pairwise meta-anaysis of combining simiar trias. The MTC wi be conducted based on a fixed effects and a random effects mode with the most appropriate mode identified as the one with the owest deviance information criterion (DIC). 14 DIC measures the fit of the mode whie penaising for the number of effective parameters. 12,15 For the chosen mode, consistency of the evidence wi be assessed using the posterior mean residua deviance, which shoud approximate the number of unconstrained data points in a good-fitting mode. 124 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 For dichotomous outcomes we wi use odds ratio as the summary statistic, and for continuous outcomes we wi use the weighted mean difference as the summary statistic. Primary anaysis wi be: Disease severity (measured by a reduction of at east 50% on Hamiton Depression Rating Scae (HDRS) 16 or Montgomery-Åsberg Depression Rating Scae (MADRS). 17 Where a study reports both we wi use ony the HDRS data). Secondary anayses wi be: Quaity of ife (QoL) as reported using a vaidated QoL rating scae18, e.g. EQ-5D, SF-36, HUI. Adverse effects (data wi be coected on those adverse effects most burdensome to patients such as agitation, akathisia, anxiety, cognitive duing, constipation, diarrhoea, dry mouth, dyspepsia, extrapyramida symptoms, kidney and thyroid dysfunction, ipid disturbance, gastrointestina beeding, (orthostatic) headache, hypergycaemia, hypotension, nausea, poyuria, restessness, sedation, sexua dysfunction, seep disturbance, thirst, tremor, visua probems, and weight gain). Withdrawas (a cause) as a surrogate outcome for adherence to medication Reapse rate Mortaity (a cause) 8-week outcome data wi be coected where reported. If 8-week data are not avaiabe, we wi use outcome data reported from the nearest avaiabe time point Subgroup anayses wi be performed in the foowing popuations on ony the primary outcome (disease severity), subject to the avaiabiity of data: Different durations of depression (i.e. time since first onset of current episode of depression, short term < 6 months, ong term > 6 months) Casses of previous antidepressants (e.g. SSRI or tricycic antidepressant) Sex (i.e. maes and femaes) Age (i.e. those 75 years and those < 75 years od) Peope with different severity's of depression, i.e. based on tria entry HDRS rating using the foowing categories: 16 0 7 = Norma 8 13 = Mid Depression 14 18 = Moderate Depression 19 22 = Severe Depression 23 = Very Severe Depression In the absence of suitabe data to perform a meta-anaysis, the avaiabe data wi be tabuated where possibe and discussed in a narrative review. Heterogeneity In addition to the existing pre-specified subgroups, other potentia sources of cinica heterogeneity coud be a resut of combining different preparations of drugs. For pairwise meta-anaysis, heterogeneity wi be expored through consideration of the study popuations, methods and interventions, by visuaisation of resuts and, in statistica terms, by the χ 2 test for homogeneity and the I 2 statistic. Statisticay significant heterogeneity wi be defined as p < 0.10. Leves of inconsistency wi be assessed using I 2 and wi be defined as foows: I 2 of: 0% 25% = ow eve of inconsistency; 26% 50% = moderate eve of inconsistency; and > 50% = high eve of inconsistency. 19 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 125

APPENDIX 1 If statisticay significant heterogeneity is detected in any of the primary or secondary anayses, hypothesis-generating subgroup anaysis wi be conducted, but the resuts from such anayses wi be treated with caution. Meta-regression wi be attempted if significant statistica heterogeneity is identified among trias anaysed and there are 10 or more trias in the review. For the MTC, where a random effects mode is deemed the best fit, the degree of heterogeneity wi be investigated by evauating the posterior mean tau-squared. Where possibe, any cosed oops formed by the network of trias wi be assessed separatey to determine if the resuts from the direct evidence is coherent with the indirect evidence when the wider network is introduced. Any incoherence identified wi be investigated. Sensitivity anaysis Sensitivity anayses are panned on the primary anaysis and consist of: different number of prior antidepressants for the current episode of depression; assuming a cass effect with SSRIs and atypica antipsychotics; changing the quaity assessment to incude the tria outcomes excuded on grounds of methodoogica quaity; i.e. those categorised as of high risk of bias; changing the anaysis from using ITT (intention to treat) data to per protoco data. Pubication bias For each of the primary pairwise meta-anayses, a funne pot wi be used to assess pubication bias. A regression of normaised effect compared with precision wi aso be cacuated as a test for sma study effects (using a p < 0.10 as an indicator of a significant resut). 20 6. Report methods for synthesising evidence of cost-effectiveness Identifying and systematicay reviewing pubished cost-effectiveness studies The foowing databases wi be used to identify studies of the cost-effectiveness. MEDLINE, EMBASE, PsycINFO, CINAHL NHS Economic Evauation Database, Heath Technoogy Assessment Database and Office of Heath Economics Heath Economic evauation database. We wi appy a cost search fiter to the comprehensive cinica search strategy described in Section 5. In order to express cinica outcomes in the form of QALYs, utiity weights for heath states reating to treatment resistant depression are required. Utiity weights represent the heath reated quaity of ife (HRQOL) associated with specific heath states; they are estimated based on peope's preferences and perceptions of quaity of ife characterising the heath states under consideration. We wi undertake a systematic quaity of ife search where heath economics and quaity-of-ife search fiters wi be used in MEDLINE, EMBASE, PsycINFO and CINAHL to identify reevant studies. The incusion and excusion criteria for economic evauations wi be the same as those for the systematic review of cinica effectiveness and in addition the heath economic evauation wi aso incude: non-randomised studies wi be incuded (e.g. decision-mode based anaysis or anaysis of person-eve cost and effectiveness data aongside observationa studies) fu cost-effectiveness anayses, cost utiity anayses, cost benefit anayses and cost consequence anayses wi be incuded stand-aone UK cost anaysis wi aso be sought and appraised 126 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Tites and abstracts returned by the search strategy wi be assessed independenty by two heath economists (LN and NT) and screened for possibe incusion. Any disagreements wi be resoved by a third heath economist (SJE). Evauation of costs and cost-effectiveness (may incude deveopment of a de novo economic mode) The methodoogica quaity of economic evauations wi be assessed according to internationay accepted criteria such as the Consensus on Heath Economic Criteria ist questions deveoped by Evers et a. (2005). 21 Any studies based on decision modes wi be assessed using the checkist deveoped by Phiips et a. (2004). 18 In addition, a new economic evauation wi be carried out from the perspective of the UK NHS using a probabiistic decision-anaytic (Markov) modeing approach to estimate the costs and QALYs of SSRI with an atypica antipsychotic compared with SSRI with ithium in the management of treatment resistant unipoar depression. An annua discount rate of 3.5% wi be used for both costs and QALYs in accordance with NICE guidance. 22 Mode structure, data inputs and modeing assumptions wi be determined in consutation with cinica experts to ensure they refect the best current cinica practice and evidence. Uncertainty in the data used to popuate the mode wi be characterised using appropriate methods, such as probabiistic sensitivity anaysis. The time horizon of our anaysis wi preferaby be a patient's ifetime in order to refect the chronic nature of the disease. However time horizon may be dictated by the avaiabiity of data in which case shorter time horizons wi be modeed. Ideay, evidence on the impact of these therapies on HRQoL wi be avaiabe directy from the trias incuded within the review. In the absence of such evidence, the mathematica mode may use indirect evidence on quaity of ife from aternative sources, such as reated technoogy appraisas or cinica guideines. Quaity of ife data wi be reviewed and used to generate the quaity adjustment weights required for the mode. We wi aso adjust utiity for age using data from the Heath Survey of Engand. 23 Resuts wi be presented as incrementa cost-effectiveness ratios (ideay cost per quaity adjusted ife year) and cost-effectiveness acceptabiity curves, which quantify the degree of uncertainty. 7. Expertise in this TAR team TAR Centre The BMJ Evidence Centre comprises over 30 speciaists with a weath of experience in diverse heath-reated areas and incudes cinicians, pharmacists, information speciaists, heath informatics speciaists, project managers, systematic reviewers, cinica guideine deveopers and heath economists. The BMJ-TAG core team consists of 5 members. Together, we have an array of experience amongst us in producing focussed reports in a short timescae for poicy customers such as NICE. Pease see beow for further detais of each team member's experience. Dr Steven J Edwards DPhi MSc BSc (Hons), Head of Heath Technoogy Assessment: Over the past 12 years, Steve has conducted over 40 systematic reviews and heath economic evauations in a range of therapeutic areas incuding cardiovascuar, CNS, gastroenteroogy, infection, oncoogy and respiratory medicine. His interests are in the use of the best avaiabe evidence for decision making with an emphasis on the design and conduct of cinica trias, systematic reviews, meta-anayses, adjusted indirect comparisons and their subsequent use in economic evauations. His postgraduate research in this area at the University of Oxford resuted in him being awarded the first doctorate of evidence based heath care. In addition, Steve is an honorary senior ecturer in heath economics at the London Schoo of Hygiene & Tropica Medicine, a member of the Cochrane Statistica Methods Group, the Campbe & Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 127

APPENDIX 1 Cochrane Economics Methods Group, and an Editoria Board member of the Internationa Journa of Cinica Practice. Dr Samantha Barton PhD BSc (Hons), Heath Technoogy Assessment Anayst: Sam has extensive experience in the critica appraisa of studies. During the past 4 years, she has contributed to the pubication of over 50 systematic reviews on prevention and treatment of various cinica conditions. She has worked on reviews in the areas of menta heath, sexua heath, infectious diseases, cardiovascuar disorders, respiratory disorders and oncoogy. Dr Victoria Hamiton MBChB, Heath Technoogy Assessment Anayst: Vicky has a cinica background with reevant experience in the fieds of genera surgery, genera medicine, genera practice, paediatrics and orthopaedic surgery. Vicky aso has experience in the critica appraisa of cinica studies and over the ast year has contributed to the pubication of systematic reviews in a variety of cinica areas. She aso has experience in the process and use of cinica audit to review current cinica practice within both primary and secondary care settings. Mr Leo Nherera MSc BSc (Hons), Heath Economist: Over the past 6 years, Leo has been working for the NICE cinica guideine programme and has successfuy worked in eight pubished cinica guideines and one Pubic Heath guideine. His work invoved appraising economic evauations as we as doing origina economic anaysis for various guideine questions to assist in guideine recommendations. Leo was invoved in organising and teaching the Heath Economics modue at Queen Mary University of London. He has aso peer-reviewed papers for the Internationa Journa of Cinica Practice. His interests are in the use of the best avaiabe evidence for decision making with an emphasis on systematic reviews and meta-anayses and their subsequent use in economic evauations. Ms Nicoa Trevor MSc BSc (Hons), Heath Economist: Nicoa has a strong mathematica background, with a Masters in anaytica, numerica and statistica modeing techniques, which over the past 2 years she has appied in the fied of heath economics, conducting economic evauations and statistica anaysis for systematic review in disease areas such as mutipe scerosis, cardiovascuar disease, Gaucher's disease and oncoogy. Her interests are in the use of the best avaiabe techniques for decision making with an emphasis on surviva anaysis, meta-anaysis, modeing approaches and the use of Bayesian methods in economic evauations. Recent pubications from the team members incude: Nherera L, Marks D, Minhas R, et a. Probabiistic cost-effectiveness anaysis of cascade screening for Famiia Hyperchoesteroaemia using aternative diagnostic and identification strategies. Heart 2011;97:1175 81. Nationa Coaborating Centre for Women's and Chidren Heath. Mutipe pregnancy: The management of twin and tripet pregnancy in the antenata period London, Roya Coege of Obstetricians and Gynaecoogists, 2011; (in press). Edwards SJ, Wordsworth S, Carke MJ. Treating pneumonia in critica care in the UK foowing faiure of initia antibiotic: a cost-effectiveness anaysis comparing meropenem with piperaciin/tazobactam. European Journa of Heath Economics 2011;12: (avaiabe onine first at: www.springerink.com/content/ q044j5t32601vt4/). Hapin DMG, Gray J, Edwards SJ, et a. Budesonide/formotero versus sametero/futicasone in COPD: a systematic review and adjusted indirect comparison of pneumonia in randomized controed trias. Internationa Journa of Cinica Practice 2011;65:764 74. Edwards SJ, Borri J. Network meta-anaysis: importance of appropriate tria seection. Vaue in Heath 2010; 13:681 2. Edwards SJ, von Matzahn R, Naya IP, et a. Budesonide/formotero for maintenance and reiever therapy: a meta-anaysis of randomised controed trias. Internationa Journa of Cinica Practice 2010; 64:619 27. 128 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Gray J, Edwards SJ, Lip GYH. Comparison of sequentia rosuvastatin doses in hyperchoesteroaemia: a meta-anaysis of randomized controed trias. Current Medica Research and Opinion 2010;26:537 47. Trevor NC, Anwick K. How can the use of predictive biomarkers ead to positive HTA recommendations? Vaue in Heath 2010;13:A423 4. Trevor NC, Tang M, Samues ER. Investigating the impact of R&D investment and poicy on innovative performance in Europe. Vaue in Heath 2010;13:A414. Edwards SJ, Gray J. Budesonide/formotero pus tiotropium (BUD/FORM+TIO) vs sametero/futicasone pus tiotropium (SALM/FLU+TIO): a systematic review and adjusted indirect comparison between two aternative tripe treatments in chronic obstructive pumonary disease (COPD). Vaue in Heath 2010;13:A319. Edwards SJ, Weton NJ, Borri J. Gefitinib compared with doubet chemotherapy for first-ine treatment non-sma-ce ung cancer (NSCLC) a systematic review and adjusted indirect comparison. Vaue in Heath 2010;13:A252 3. Edwards SJ, Weton NJ, Borri J. Toerabiity of first-ine treatments of ocay advanced or metastatic non-sma-ce ung cancer (NSCLC) a systematic review and adjusted indirect comparison. Vaue in Heath 2010;13:A250. Nherera L, Cavert NW, DeMott K, et a. Cost effectiveness anaysis of the use of a high intensity statin compared to a ow intensity statin in the management of patients with famiia hyperchoesteroaemia. Current Medica Research and Opinion 2010;26:529 36. Visintin C, Muggestone MA, Amerie MQ, et a. on behaf of the Guideine Deveopment Group. Management of hypertensive disorders during pregnancy: summary of NICE guidance. British Medica Journa 2010;341:c2207. Nationa Coaborating Centre for Women's and Chidren Heath. Hypertension in pregnancy: The management of hypertensive disorders during pregnancy. London, Roya Coege of Obstetricians and Gynaecoogists 2010. Externa Cinica Expert Advisors Professor Phiip J. Cowen MRC Cinica Scientist and Professor of Psychopharmacoogy; Speciaist in Psychopharmacoogy of Mood Disorders Neurosciences Buiding, Warneford Hospita, Oxford OX3 7JX, United Kingdom phi.cowen@psych.ox.ac.uk Recent pubications incude: McCabe C, Mishor Z, Cowen PJ, et a. Diminished neura processing of aversive and rewarding stimui during seective serotonin reuptake inhibitor treatment. Bioogica Psychiatry 2010;67:439 45. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 129

APPENDIX 1 Harmer CJ, O'Suivan U, Favaron E, et a. Effect of acute antidepressant administration on negative affective bias in depressed patients. American Journa of Psychiatry 2009;166:1178 84. Harmer CJ, Goodwin GM, Cowen PJ. Why do antidepressants take so ong to work? A cognitive neuropsychoogica mode of antidepressant drug action. British Journa of Psychiatry 2008;195:102 8. Geder M, Cowen P, Harrison P. Shorter Oxford Textbook of Psychiatry. Oxford University Press, Oxford, 1995, 2001, 2006. Dr Luiz Dratcu Consutant Psychiatrist and Speciaist in Psychopharmacoogy, Treatment Resistant Menta Iness, Schizophrenia and Affective Disorders Maudsey Hospita South London & Maudsey NHS Foundation Trust Denmark Hi, London SE5 8AZ, United Kingdom uiz.dratcu@sam.nhs.uk Recent pubications incude: Dratcu L. The quest for the pharmacoogica treatment of schizophrenia: from conventiona neuroeptics to atypica anti-psychotics and beyond. Vertex 2010;21:385 93. Dratcu L. The future of depression: a compex neuroendocrine, infammatory and neurodegenerative systemic iness. Vertex 2009:20:329 41. Dratcu L, Grandison A, McKay G, et a. Cozapine-resistant psychosis, smoking, and caffeine: managing the negected effects of substances that our patients consume every day. American Journa of Therapeutics 2007;14:314 8. Dratcu L, Oowu P, Hawramy M, et a. Aripiprazoe in the acute treatment of mae patients with schizophrenia: effectiveness, acceptabiity, and risks in the inner-city hospita setting. Neuropsychiatric Disease and Treatment 2006;2:191 7. 8. Competing interests of authors Steve Edwards has previousy been an empoyee of AstraZeneca, which hods the marketing authorisation for Seroque (quetiapine). He has no ongoing financia connection nor owns significant shares with AstraZeneca. Professor Phiip J. Cowen has received consutancy fees from Servier, Lundbeck and Ei Liy, and fees for speaking from AstraZeneca, Servier and Lundbeck. He has aso provided advice to ega representatives of GSK. Dr Luiz Dratcu has received consutancy fees, fees for speaking and hospitaity from BMS/Otsuka and Merck. He has aso received hospitaity from Liy. 130 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 9. Timetabe/miestones Finaise protoco June 2011 Send progress report to NETSCC, HTA February 2012 Submit assessment report to NETSCC, HTA March 2012 The timetabe is based on a 6-month working time-frame, commencing in mid-juy assuming that the fina approva of the protoco has been received by this time. Timeines may be subject to change in the event of any additiona urgent work commitments such as STA work for NICE; however we wi endeavour to inform NETSCC of any commitments which may deay the competion of this project at the eariest possibe date. 10. Appendices 10.1.1. Draft MEDLINE search strategy (Cinica) 10.1.2. Draft MEDLINE search strategy (Heath Economics and Quaity of ife) 10.2. Data extraction form 10.3 Team members' contributions 10.4 References Appendix 10.1.1 Draft MEDLINE search strategy < 1948 to June Week 1 2011 > Search Strategy: 1. Randomized Controed Trias as Topic / (73,451) 2. randomized controed tria/ (308,386) 3. Random Aocation/ (71,692) 4. Doube Bind Method/ (110,600) 5. Singe Bind Method/ (15,044) 6. cinica tria/ (463,236) 7. cinica tria, phase i.pt. (11,244) 8. cinica tria, phase ii.pt. (17,834) 9. cinica tria, phase iii.pt. (6176) 10. cinica tria, phase iv.pt. (614) 11. controed cinica tria.pt. (82,578) 12. randomized controed tria.pt. (308,386) 13. muticenter study.pt. (131,287) 14. cinica tria.pt. (463,236) 15. exp Cinica Trias as topic/ (242,013) 16. or/1-15 (859,148) 17. (cinica adj tria$).tw. (155,138) 18. ((sing$ or doub$ or treb$ or trip$) adj (bind$3 or mask$3)).tw. (107,934) 19. PLACEBOS/ (29,733) 20. pacebo$.tw. (129,547) Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 131

APPENDIX 1 21. randomy aocated.tw. (12,594) 22. (aocated adj2 random$).tw. (14,831) 23. or/17-22 (326,697) 24. 16 or 23 (954,423) 25. case report.tw. (158,367) 26. etter/ (716,157) 27. historica artice/ (275,084) 28. or/25-27 (1,139,766) 29. 24 not 28 (928,335) 30. exp Depression/ or exp Depressive Disorder/ (127,115) 31. (depress* or adjustment disorder* or mood disorder* or affective disorder* or affective symptom* or dysthymi* or dysphori*).mp. (344,159) 32. 30 or 31 (344,159) 33. 29 and 32 (39,592) 34. Serotonin Uptake Inhibitors/ or Antidepressive Agents, Second-Generation/ or ssri*.mp. or exp Serotonin 35. Antagonists/ (59,672) 36. citaopram.mp. or exp Citaopram/ (4058) 37. escitaopram.mp. (779) 38. fuoxetine.mp. or exp Fuoxetine/ (9218) 39. fuvoxamine.mp. or exp Fuvoxamine/ (2286) 40. paroxetine.mp. or exp Paroxetine/ (4469) 41. sertraine.mp. or exp Sertraine/ (2966) 42. 36 or 37 or 38 or 39 or 40 (16,313) 43. 35 and 41 (1722) 44. 35 or 37 or 38 or 39 or 40 (18,583) 45. 36 and 43 (713) 46. 35 or 36 or 38 or 39 or 40 (11,526) 47. 37 and 45 (2095) 48. 35 or 36 or 37 or 39 or 40 (17,294) 49. 38 and 47 (931) 50. 35 or 36 or 37 or 38 or 40 (15,845) 51. 39 and 49 (1665) 52. 35 or 36 or 37 or 38 or 39 (16,994) 53. 40 and 51 (1311) 54. 42 or 44 or 46 or 48 or 50 or 52 (3310) 55. 33 and 53 (799) 56. ithium.mp. or exp Lithium Carbonate/ or exp Lithium/ or exp Lithium Compounds/ or exp Lithium Choride/ (29,746) 57. (antipsychotic* or anti?psychotic* or anti-psychotic*).mp. (41,739) 58. amisupride.mp. (571) 59. aripiprazoe.mp. (1454) 60. cozapine.mp. (8530) 61. oanzapine.mp. (5275) 62. paiperidone.mp. (153) 63. quetiapine.mp. (2428) 64. risperidone.mp. (5910) 65. or/34-40 (66,927) 66. or/55-63 (72,543) 67. 33 and 64 and 65 (713) 68. 54 or 66 (1455) 132 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 10.1.2 Drat MEDLINE search strategy (Heath Economics and Quaity of ife) Economics search terms 1. exp economics/ (438,053) 2. exp Costs and Cost Anaysis/ (38,816) 3. Cost Benefit Anaysis/ (51,007) 4. vaue of ife/ (5162) 5. exp modes economic/ (7945) 6. exp fees/and charges/ (7703) 7. exp budgets/ (10,939) 8. (economic adj2 burden).tw. (2622) 9. (expenditure* not energy).tw. (14,210) 10. budget*.tw. (14,415) 11. (economic* or price* or pricing or financ*or fee* or pharmacoeconomic* or pharmaeconomic* or pharmaco-economic*).tw. (128,436) 12. (decision adj1 (tree* or anays* or mode*)).tw. (6411) 13. Resource Aocation/ (6522) 14. (unit cost or unit-cost or unit-costs or unit costs or drug cost or drug costs or hospita costs or heath-care costs or heath care cost or medica cost or medica costs).tw. (16,355) 15. ((vaue or vaues or vauation) adj2 (money or monetary or ife or ives or costs or cost)).tw. (3225) 16. Markov Chains/ (7220) 17. exp Decision Support Techniques/ (48,239) 18. (resource adj2 (use* or utii* or aocat*)).tw. (10,801) 19. (cost adj2 (uti* or effective* or efficac* or benefit* or consequence* or anays* or minimi* or aocation* or contro* or iness* or affordabe* or fee* or charge* or charges)).tw. (71,017) 20. or/1 19 (627,358) Combining condition, intervention, comparator and cost terms gets a tota of 36 studies potentia cost-effectiveness abstracts. Quaity of ife search terms 1. exp quaity of ife/ (90,943) 2. quaity of ife.tw (100,676) 3. ife quaity.tw (2525) 4. (sf 36 or sf36 or sf thirty six or sf thirty six or short form 36 or short form thirty six or short form thirtysix or shortform 36).tw (11,072) 5. (euroqo or eq5d or eq 5d).tw (2147) 6. quaity adjusted ife$.tw (3963) 7. (QALY$ or ifeyear$ or ife year$ or ((quait$3 or vaue) adj3 (ife or surviva))).tw. (108,136) 8. ((burden adj3 (disease or iness)) or (resource adj3 (aocation$ or utiit$)) or (vaue adj5 money)).tw. (12,216) 9. (budget$ or cost$ or econom$ or expenditure$ or financ$ or fisca$ or funding or pharmacoeconomic$ or price or prices or pricing).tw. (441,366) 10. (Hamiton depression rating scae$).ab. (2004) 11. (Montgomery-Åsberg depression rating scae$).ab. (1004) 12. or/1-11 (575,570) Combining condition, intervention, comparator and quaity of ife terms gets a tota of 136 potentia quaity of ife abstracts. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 133

APPENDIX 1 Appendix 10.2 Data extraction form PART ONE: REVIEW, REVIEWER AND STUDY INFORMATION Study ID: Reviewer name: Date of competion of this form: Tite of paper/abstract: Source (journa, year, voume, pages): Authors: Language of pubication: Type of paper (e.g. fu paper/abstract/poster): PART TWO: VERIFICATION OF STUDY ELIGIBILITY Type of cinica tria 1) Is the study randomised? YES UNCLEAR NO Popuation in the cinica tria 2) Is the popuation aduts 18 years od? YES UNCLEAR NO 3) Did the RCT incude peope with unipoar depression? YES UNCLEAR NO 4) Did the RCT incude peope with treatment resistant depression (defined as faiure to respond to 2 antidepressants)? YES UNCLEAR NO Interventions in the cinica tria 5) Does the tria compare SSRI + atypica antipsychotic or ithium or no treatment with SSRI + ithium or atypica antipsychotic or pacebo or no treatment? 6) Did both groups experience the same care except for the two interventions under investigation? YES UNCLEAR NO YES UNCLEAR NO Outcomes of the cinica tria 7) Does the study report on outcomes during the treatment of the acute phase of depression? 8) Did the cinica tria investigate at east one of the foowing: disease severity, quaity of ife, adverse events, withdrawas (a cause), reapse rate, mortaity (a cause)? 9) Are the outcomes measured after 4 weeks treatment with study medication? YES UNCLEAR NO YES UNCLEAR NO YES UNCLEAR NO If you answered NO to any of the above questions do not proceed to Part 3. 134 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 PART THREE: INFORMATION ABOUT THE STUDY Characteristics of the tria Country(ies) where the cinica tria was conducted: Sponsors of the cinica tria: Any conficts of interest reported for any of the researchers? Date the cinica tria was conducted: Type of cinica tria design (e.g. parae, crossover, or custer tria): If the tria was of crossover design, are there pre-crossover resuts reported? Was the tria muticentre? If so, how many centres were there? Characteristics of the patients Incusion criteria: how and where were patients enroed, were any patient risk factors used? What detais of the antidepressant(s) patients had faied to respond to are provided? Excusion criteria: were specific groups of peope excuded? Tota number of peope randomised: Information on the age of the patients: Information on the sex of the patients (m/f): Information on the ethnicity of the patients: Information on patients' medica history (i.e. previous depression): Type of intervention Intervention 1: SSRI + XX (where XX = atypica antipsychotic or ithium or no treatment) SSRI name and brand: SSRI dose and regimen used (e.g. 80 mg OD): Deivery of SSRI (e.g. PO tabet/dissovabe/enteric coated): Number of doses of SSRI given per day (with SD/SE if given): Duration of SSRI treatment in days (with SD/SE if given): What was XX (name and brand)? XX dose and regimen used (e.g. 80 mg OD): Deivery of XX (e.g. PO tabet/dissovabe/enteric coated): Number of doses of XX given per day (with SD/SE if given): Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 135

APPENDIX 1 Duration of XX treatment in days (with SD/SE if given): Number of patients randomised: Intervention 2: SSRI + YY (where YY = ithium or atypica antipsychotic or pacebo or no treatment) SSRI name and brand: SSRI dose and regimen used (e.g. 80 mg OD): Deivery of SSRI (e.g. PO tabet/dissovabe/enteric coated): Number of doses of SSRI given per day (with SD/SE if given): Duration of SSRI treatment in days (with SD/SE if given): What was YY (name and brand)? YY dose and regimen used (e.g. 80 mg OD): Deivery of YY (e.g. PO tabet/dissovabe/enteric coated): Number of doses of YY given per day (with SD/SE if given): Duration of YY treatment in days (with SD/SE if given): Number of patients randomised: Was the formuation and appearance of YY (e.g. ithium) matched to that of XX (e.g. atypica antipsychotic)? Were any additiona interventions given to either or both groups? Types of outcome Which of the foowing outcomes have been assessed in the cinica tria? YES UNCLEAR NO Disease severity? YES UNCLEAR NO How was disease severity defined in the cinica tria? YES UNCLEAR NO Quaity of ife? YES UNCLEAR NO How was quaity of ife defined in the cinica tria? YES UNCLEAR NO Adverse events? YES UNCLEAR NO How were adverse events defined in the cinica tria? (e.g. investigator attributed?) YES UNCLEAR NO Withdrawa (a cause)? YES UNCLEAR NO How was withdrawa defined in the cinica tria? YES UNCLEAR NO Reapse rate? YES UNCLEAR NO How was reapse rate defined in the cinica tria? YES UNCLEAR NO 136 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 A-cause mortaity? YES UNCLEAR NO How was a-cause mortaity defined in the cinica tria? YES UNCLEAR NO Any other outcomes reported in tria (pease ist)? YES UNCLEAR NO ITT data coection tabe: Outcomes Disease severity Quaity of ife Withdrawas (a cause) Reapse rate A-cause mortaity Adverse events (pease specify) 50% reduction in HDRS 50% reduction in MADRS Tria scae: Timeframe (weeks) SSRI + XX SSRI + YY n N n N n = number of patients with the outcome; N = number of patients assessed Per protoco data coection tabe: Outcomes Disease severity Quaity of ife Withdrawas (a cause) Reapse rate A-cause mortaity Adverse events (pease specify) 50% reduction in HDRS 50% reduction in MADRS Tria scae: Timeframe (weeks) SSRI + XX SSRI + YY n N n N n = number of patients with the outcome; N = number of patients assessed Did the RCT carry out any subgroup anayses of interest? (i.e. Different durations of depression, different casses of previous antidepressants, different genders, age, different severity's of depression or different number of prior antidepressants) If yes, pease give detais here. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 137

APPENDIX 1 PART FOUR: CLINICAL TRIAL QUALITY Pease describe the method of randomisation and aocation conceament used in the cinica tria: Pease describe the method of binding and who was binded in the cinica tria: Pease describe the number of patients ost to foow up (the overa number and number by treatment group, give reasons for oss to foow up): How woud you describe the trias design to minimise bias for (pease tick): Outcome Risk of bias Low risk Uncear risk High risk Comments Disease severity Quaity of ife Withdrawas (a cause) Reapse rate A-cause mortaity Random sequence generation Aocation conceament Binding (participants & personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Random sequence generation Aocation conceament Binding (participants & personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Random sequence generation Aocation conceament Binding (participants & personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Random sequence generation Aocation conceament Binding (participants & personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Random sequence generation Aocation conceament Binding (participants & personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias 138 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Outcome Risk of bias Low risk Uncear risk High risk Comments Adverse events Random sequence generation Aocation conceament Binding (participants & personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias How woud you rate the trias overa risk of bias? Low risk Uncear High risk Do you have any additiona comments you woud ike to make about this cinica tria? Ideay, woud you ike further information about the cinica tria from the authors (If so, pease give detais)? Appendix 10.3 Team members' contributions Steve Edwards, Head of HTA, wi deveop the protoco, act as the third reviewer for assessment of trias and cost-effectiveness studies, vaidate data extraction and any data anaysis required, vaidate the economic mode, contribute to writing/editing of the report, be overa director of the project and act as guarantor of the report. Sam Barton, HTA Anayst, wi act as co-reviewer for assessing trias for incusion and data extraction, and contribute to the writing/editing of the report. Vicky Hamiton, HTA Anayst, wi provide overa project management, deveop the protoco, write and run the search strategy, act as co-reviewer for assessing trias for incusion and data extraction (and perform data anaysis as required), and contribute to the writing/editing of the report. Leo Nherera, Heath Economist, wi deveop the protoco, act as co-reviewer of the cost-effectiveness studies, deveop the economic mode, and contribute to the writing/editing of the report. Nicoa Trevor, Heath Economist, wi act as co-reviewer of the cost-effectiveness studies, vaidate the economic mode, and contribute to the writing/editing of the report. Professor Cowen and Dr Dratcu, Cinica Expert Advisors, wi provide cinica advice as required through out the protoco deveopment and review processes. Appendix 10.4 References 1. http://www.who.int/menta_heath/management/depression/definition/en/ 2. Nationa Institute for Heath and Cinica Exceence (2009) Depression: the treatment and management of depression in aduts (update). CG90. London: Nationa Institute for Heath and Cinica Exceence. 3. DSM-IV-TR Diagnostic and Statistica Manua of Menta Disorders Fourth Edition, Text Revision. DOI: 10.1176/appi.books.9780890423349 4. http://www.who.int/cassifications/icd/en/ 5. Neson JC, Papakostas GI. Atypica Antipsychotic Augmentation in Major Depressive Disorder: A Meta-Anaysis of Pacebo-Controed Randomized Trias. Am J Psychiatry. 2009;166:980 991. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 139

APPENDIX 1 6. Schuz KF, Atman DG, Moher D, for the CONSORT Group. CONSORT 2010 Statement: updated guideines for reporting parae group randomised trias. Ann Int Med. 2010;152. Epub 24 March. 7. Higgins JPT, Green S (editors). Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March 2011]. The Cochrane Coaboration, 2011. Avaiabe from www.cochrane-handbook.org 8. Soares HP, Danies S, Kumar A, Carke M, Scott C, Swann S, Djubegovic B. Radiation Therapy Oncoogy Group. Bad reporting does not mean bad methods for randomised trias: observationa study of randomised controed trias performed by the Radiation Therapy Oncoogy Group. BMJ. 2004 Jan 3;328(7430):22 25. 9. Mante N, Haensze W. Statistica aspects of the anaysis of data from retrospective studies of disease. J Nat Cancer Inst. 1959;22:719 748. 10. DerSimonian R, Laird N. Meta-anaysis in cinica trias. Contro Cin Trias. 1986 Sep;7(3):177 188. 11. Ades AE. A chain of evidence with mixed comparisons: modes for mutiparameter synthesis and consistency of evidence. Stat Med. 2003 Oct 15;22(19):2995 3016. 12. Lu G, Ades AE. Combination of direct and indirect evidence in mixed treatment comparisons. Stat Med. 2004 Oct 30;23(20):3105 3124. 13. Cadwe DM, Ades AE, Higgins JPT. Simutaneous comparison of mutipe treatments: combining direct and indirect evidence. BMJ. 2005 Oct 15;331(7521):897 900. 14. Spiegehater DJ, Best NG, Carin BP, Van Der Linde, A. Bayesian measures of mode compexity and fit. J Roy Statist Soc B. 2002;64(3):583 639. 15. Dempster AP. The direct use of ikeihood for significance testing. Stat Comp. 1997;7:247 252. 16. Hamiton M. A rating scae for depression. J Neuro Neurosurgery Psychiatry. 1960;23:56 62. 17. Montgomery SA, Åsberg M. A new depression scae designed to be sensitive to change. Br J Psychiatry. 1979 Apr;134:382 389. 18. Phiips Z, Ginney L, Scupher M, Caxton K, Goder S, Riemsma R, et a. Review of guideines for good practice in decision-anaytic modeing in heath technoogy assessment. Heath Techno Assess. 2004 Sep;8(36):iii-iv, ix-xi, 1 158. 19. Higgins JP, Thompson SG, Deeks JJ, Atman DG. Measuring inconsistency in meta-anayses. BMJ. 2003 Sep 6;327(7414):557 560. 20. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-anaysis detected by a simpe, graphica test. BMJ. 1997 Sep 13;315(7109):629 634. 21. Evers S, Goossens M, de Vet H, van Tuder M, Ament A. Criteria ist for assessment of methodoogica quaity of economic evauations: Consensus on Heath Economic Criteria. Int J Techno Assess Heath Care. 2005 Spring;21(2):240 245. 22. Nationa Institute for Heath and Cinica Exceence. Guide to the methods of technoogy appraisa June 2008 (reference N1618). http://www.nice.org.uk/media/b52/a7/ TAMethodsGuideUpdatedJune2008.pdf 23. Age adjusted utiity/adjusting utiity for age Heath Survey of Engand 1996. http://www.archive. officia-documents.co.uk/document/doh/survey96/tab5-29.htm 140 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 2 Literature search strategies for the cinica review Database: Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations and Ovid MEDLINE(R) 1946 to present Search strategy 1. Depression/ or Depressive Disorder/ or Depressive Disorder, Major/ (126,194) 2. depress$.tw. (267,648) 3. 1 or 2 (300,405) 4. Serotonin Uptake Inhibitors/ or Antidepressive Agents, Second-Generation/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. or Serotonin Antagonists/ (35,140) 5. (citaopram or citaopramum or ceexa or ciprami).tw. or Citaopram/ (4259) 6. (escitaopram or S-citaopram or exapro or cipraex).tw. (939) 7. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or Fuoxetine/ (9738) 8. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or Fuvoxamine/ (2372) 9. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or Paroxetine/ (4698) 10. (sertrain$ or ustra or zooft).tw. or Sertraine/ (3149) 11. 4 or 5 or 6 or 7 or 8 or 9 or 10 (42,763) 12. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. (1,393,254) 13. treatment faiure/ (22,171) 14. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. (36,866) 15. 12 or 13 or 14 (1,427,769) 16. randomized controed tria.pt. (314,314) 17. (random$ or rct or pacebo$).tw. (623,289) 18. ((sing$ or doub$ or trip$ or treb$) adj (bind$3 or mask$3 or sham$ or dummy)).tw. (112,536) 19. (case reports or comment or editoria or in vitro or etter).pt. (2,881,916) 20. 19 not (randomized controed tria or controed cinica tria).pt. (2,874,149) 21. (16 or 17 or 18) not 20 (693,670) 22. exp animas/ not humans.sh. (3,650,161) 23. 21 not 22 (621,736) 24. (anti-depress$ or antidepress$).tw. (41,423) 25. Antidepressive Agents/ (28,927) 26. 24 or 25 (54,177) 27. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. (2,822,627) 28. Drug Therapy, Combination/ (125,552) 29. 27 or 28 (2,888,000) 30. Lithium/ or exp Lithium Compounds/ (23,585) 31. antipsychotic agents/ or cozapine/ or risperidone/ (38,767) 32. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. (25,357) 33. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. (19,696) Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 141

APPENDIX 2 34. or/30-33 (74,184) 35. (26 and 29 and 34) or 11 (44,571) 36. 3 and 15 and 23 and 35 (938) Database: PsycINFO 1806 to August week 2 2011 Search strategy 1. exp Major Depression/ (75,749) 2. depress$.tw. (180,819) 3. 1 or 2 (182,464) 4. Serotonin Antagonists/ or serotonin reuptake inhibitors/ or (ssri$ or (serotonin adj2 inhibitor$)).tw. (9804) 5. (citaopram or citaopramum or ceexa or ciprami).tw. or Citaopram/ (1781) 6. (escitaopram or S-citaopram or exapro or cipraex).tw. (615) 7. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or Fuoxetine/ (5066) 8. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or Fuvoxamine/ (1376) 9. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or Paroxetine/ (2593) 10. (sertrain$ or ustra or zooft).tw. or Sertraine/ (1930) 11. 4 or 5 or 6 or 7 or 8 or 9 or 10 (16,639) 12. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. (167,518) 13. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. (8907) 14. 12 or 13 (174,891) 15. treatment outcome cinica tria.md. (19,559) 16. treatment effectiveness evauation/ (12,499) 17. (random$ or rct or pacebo$).tw. (117,674) 18. ((sing$ or doub$ or trip$ or treb$) adj (bind$3 or mask$3 or sham$ or dummy)).tw. (16,252) 19. 15 or 16 or 17 or 18 (136,659) 20. (comment or editoria or etter).dt. (31,151) 21. 20 not treatment outcome cinica tria.md. (30,861) 22. 19 not 21 (135,679) 23. antidepressant drugs/ (13,273) 24. (antidepress$ or anti-depress$).tw. (26,031) 25. 23 or 24 (28,058) 26. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. (527,381) 27. poypharmacy/ (536) 28. Drug Augmentation/ (740) 29. 26 or 27 or 28 (527,592) 30. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. (8478) 31. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. (14,792) 32. exp Lithium Carbonate/ or Lithium/ (5297) 33. neuroeptic drugs/ or aripiprazoe/ or cozapine/ or oanzapine/ or quetiapine/ or risperidone/ (20,778) 142 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 34. 30 or 31 or 32 or 33 (30,510) 35. 25 and 29 and 34 (1550) 36. 11 or 35 (17,771) 37. 3 and 14 and 22 and 36 (720) Database: EMBASE 1974 to 2011 week 32 Search strategy 1. *depression/ or *major depression/ (107,356) 2. depress$.tw. (302,295) 3. 1 or 2 (321,132) 4. serotonin antagonist/ or serotonin uptake inhibitor/ or fuoxetine pus oanzapine/ or fuvoxamine maeate/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. (46,279) 5. (citaopram or citaopramum or ceexa or ciprami).tw. or citaopram/ (14,277) 6. (escitaopram or S-citaopram or exapro or cipraex).tw. or escitaopram/ (4407) 7. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or fuoxetine/ or fuoxetine pus oanzapine/ (32,341) 8. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or fuvoxamine/ or fuvoxamine maeate/ (10,747) 9. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or paroxetine/ (19,872) 10. (sertrain$ or ustra or zooft).tw. or sertraine/ (16,373) 11. 4 or 5 or 6 or 7 or 8 or 9 or 10 (79,964) 12. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. (1,520,915) 13. drug treatment faiure/ (10,532) 14. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. (40,713) 15. 12 or 13 or 14 (1,555,120) 16. randomized controed tria/ (284,436) 17. (random$ or rct or pacebo$).tw. (707,279) 18. ((sing$ or doub$ or trip$ or treb$) adj (bind$3 or mask$3 or sham$ or dummy)).tw. (129,146) 19. 16 or 17 or 18 (787,496) 20. (editoria or etter).pt. (1,110,740) 21. 20 not (16 or cinica tria/) (1,070,423) 22. 19 not 21 (785,365) 23. exp animas/ not exp humans/ (1,237,993) 24. 22 not 23 (761,970) 25. (anti-depress$ or antidepress$).tw. (52,230) 26. antidepressant agent/ (59,124) 27. 25 or 26 (85,174) 28. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. (3,077,561) 29. drug combination/ (53,343) 30. 28 or 29 (3,112,487) 31. exp ithium/ or exp ithium derivative/ or ithium carbonate/ or ithium choride/ or ithium citrate/ (46,132) 32. atypica antipsychotic agent/ or ziprasidone/ or risperidone/ or quetiapine/ or paiperidone/ or oanzapine/ or cozapine/ or cozapine derivative/ or cozapine n oxide/ or aripiprazoe/ or amisupride/ (46,689) 33. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. (29,256) Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 143

APPENDIX 2 34. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. (27,771) 35. or/31-34 (96,393) 36. (27 and 30 and 35) or 11 (81,812) 37. 3 and 15 and 24 and 36 (1303) Database: Cochrane Centra Register of Controed Trias Search strategy #1 MeSH descriptor Depressive Disorder, this term ony (4140) #2 MeSH descriptor Depression, this term ony (4118) #3 depress*:ti,ab in Cinica Trias (25,500) #4 MeSH descriptor Depressive Disorder, Major, this term ony (1781) #5 MeSH descriptor Antidepressive Agents, Second-Generation, this term ony (1065) #6 MeSH descriptor Serotonin Antagonists, this term ony (878) #7 MeSH descriptor Serotonin Uptake Inhibitors, this term ony (2018) #8 MeSH descriptor Citaopram, this term ony (606) #9 MeSH descriptor Fuoxetine, this term ony (1086) #10 MeSH descriptor Fuvoxamine, this term ony (347) #11 MeSH descriptor Paroxetine, this term ony (697) #12 MeSH descriptor Sertraine, this term ony (547) #13 ssri* or (serotonin near/2 inhibitor*) or citaopram or citaopramum or ceexa or ciprami or escitaopram or S-citaopram or exapro or cipraex or fuoxetin* or fuokset* or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax or fuvoxamin* or fuvoksami* or faverin or uvox or paroxetin* or parokset* or seroxat or paxi or pexeva or sertrain* or ustra or zooft in Cinica Trias (6667) #14 (#5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13) (7660) #15 refract* or resistan* or nonrespon* or (no* near/2 respon*) or unrespons* or fai* or (incompet* near/1 respon*) or (inadequat* near/1 respons*) or (sub* near/2 respon*) or (poor* near/1 respon*): ti,ab in Cinica Trias (75,271) #16 MeSH descriptor Treatment Faiure, this term ony (2311) #17 (#15 OR #16) (75,271) #18 antidepress* or anti-depress*:ti,ab in Cinica Trias (7707) #19 MeSH descriptor Antidepressive Agents, this term ony (2441) #20 augment* or adjunct* or combin* or add* or potentiation:ti,ab in Cinica Trias (365,207) #21 MeSH descriptor Drug Therapy, Combination, this term ony (22,330) 144 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 #22 Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs in Cinica Trias (1579) #23 (atypica near/1 antipsychotic*) or (atypica near/1 anti-psychotic*) or Amisuprid* or Soian or Aripiprazo* or Abiify or Cozapin* or Cozari or Kotsapiini or Kozapin* or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin* or Oantsapiini or Zyprexa or Zypadhera or Paiperidon* or Invega or Xepion or Quetiapin* or Seroque or Risperidon* or Riszperidon or Rysperydon or Risperda or Ziprasidon* or Geodon in Cinica Trias (4083) #24 MeSH descriptor Lithium, this term ony (626) #25 MeSH descriptor Lithium Compounds expode a trees (349) #26 MeSH descriptor Lithium Carbonate, this term ony (239) #27 MeSH descriptor Lithium Choride, this term ony (20) #28 MeSH descriptor Antipsychotic Agents, this term ony (3068) #29 MeSH descriptor Cozapine, this term ony (378) #30 MeSH descriptor Risperidone, this term ony (733) #31 (#22 OR #23 OR #24 OR #25 OR #26 OR #27 OR #28 OR #29 OR #30) (6535) #32 (( #18 OR #19 ) AND ( #20 OR #21 ) AND #31) (403) #33 (( #1 OR #2 OR #3 OR #4 ) AND ( #14 OR #32 ) AND #17) (756) Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 145

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 3 Copy of data extraction form TREATMENT-RESISTANT DEPRESSION DATA EXTRACTION FORM PART ONE: REVIEW, REVIEWER AND STUDY INFORMATION Study ID: Reviewer name: Date of competion of this form: Tite of paper/abstract: Source (journa, year, voume, pages): Authors: Language of pubication: Type of paper (e.g. fu paper/abstract/poster): PART TWO: VERIFICATION OF STUDY ELIGIBILITY Type of cinica tria 1) Is the study randomised? YES UNCLEAR NO Popuation in the cinica tria 2) Is the popuation aduts 18 years od? YES UNCLEAR NO 3) Did the RCT incude peope with unipoar depression? YES UNCLEAR NO 4) Did the RCT incude peope with treatment resistant depression (defined as faiure to respond to 2 antidepressants)? YES UNCLEAR NO Interventions in the cinica tria 5) Does the tria compare SSRI + atypica antipsychotic or ithium or no treatment with SSRI + ithium or atypica antipsychotic or pacebo or no treatment? 6) Did both groups experience the same care except for the two interventions under investigation? YES UNCLEAR NO YES UNCLEAR NO Outcomes of the cinica tria 7) Does the study report on outcomes during the treatment of the acute phase of depression? 8) Did the cinica tria investigate at east one of the foowing: disease severity, quaity of ife, adverse events, withdrawas (a cause), reapse rate, mortaity (a cause)? 9) Are the outcomes measured after 4 weeks treatment with study medication? YES UNCLEAR NO YES UNCLEAR NO YES UNCLEAR NO If you answered NO to any of the above questions do not proceed to Part 3. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 147

APPENDIX 3 PART THREE: INFORMATION ABOUT THE STUDY Characteristics of the tria Country(ies) where the cinica tria was conducted: Sponsors of the cinica tria: Any conficts of interest reported for any of the researchers? Date the cinica tria was conducted: Type of cinica tria design (e.g. parae, crossover, or custer tria): If the tria was of crossover design, are there pre-crossover resuts reported? Was the tria muticentre? If so, how many centres were there? Characteristics of the patients Incusion criteria How and where were patients enroed, were any patient risk factors used? What detais of the antidepressant(s) patients had faied to respond to are provided? How was treatment resistance defined? Excusion criteria Were specific groups of peope excuded? Tota number of peope randomised: Information on the age of the patients: Information on the sex of the patients (m/f): Information on the ethnicity of the patients: Information on patients' medica history (i.e. previous depression): Type of intervention: Intervention 1 SSRI + XX (where XX = atypica antipsychotic or ithium or no treatment) SSRI name and brand: SSRI dose and regimen used [e.g. 80 mg once daiy (OD)]: Deivery of SSRI [e.g. per os (p.o.) tabet/dissovabe/enteric coated]: Number of doses of SSRI given per day (with SD/SE if given): Duration of SSRI treatment in days (with SD/SE if given): 148 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 What was XX (name and brand)? XX dose and regimen used (e.g. 80 mg OD): Deivery of XX (e.g. p.o. tabet/dissovabe/enteric coated): Number of doses of XX given per day (with SD/SE if given): Duration of XX treatment in days (with SD/SE if given): Number of patients randomised: Intervention 2 SSRI + YY (where YY = ithium or atypica antipsychotic or pacebo or no treatment) SSRI name and brand: SSRI dose and regimen used (e.g. 80 mg OD): Deivery of SSRI (e.g. p.o. tabet/dissovabe/enteric coated): Number of doses of SSRI given per day (with SD/SE if given): Duration of SSRI treatment in days (with SD/SE if given): What was YY (name and brand)? YY dose and regimen used (e.g. 80 mg OD): Deivery of YY (e.g. p.o. tabet/dissovabe/enteric coated): Number of doses of YY given per day (with SD/SE if given): Duration of YY treatment in days (with SD/SE if given): Number of patients randomised: Was the formuation and appearance of YY (e.g. ithium) matched to that of XX (e.g. atypica antipsychotic)? Were any additiona interventions given to either or both groups? Types of outcome Which of the foowing outcomes have been assessed in the cinica tria? Response? How was response defined in the cinica tria? YES UNCLEAR NO QoL? How was QoL defined in the cinica tria? YES UNCLEAR NO Adverse events? How were adverse events defined in the cinica tria? (e.g. investigator attributed?) YES UNCLEAR NO continued Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 149

APPENDIX 3 Withdrawa (a cause)? How was withdrawa defined in the cinica tria? YES UNCLEAR NO Reapse rate? How was reapse rate defined in the cinica tria? YES UNCLEAR NO A-cause mortaity? How was a-cause mortaity defined in the cinica tria? YES UNCLEAR NO Remission rate? How was remission rate defined in the cinica tria? YES UNCLEAR NO Other outcomes Any other outcomes reported in tria (pease ist)? YES UNCLEAR NO ITT data coection tabe Outcomes Response QoL Withdrawas (a cause) Reapse rate Remission rate A-cause mortaity Adverse events (pease specify) 50% reduction in MADRS 50% reduction in HAMD Improvement from baseine on MADRS Improvement from baseine on CGI Improvement from baseine on HAMA Improvement from baseine on BPRS Tria scae Weight gain Time frame (weeks) SSRI + oanzapine SSRI + pacebo n N n N BPRS, Brief Psychiatric Rating Scae; HAMA, Hamiton Anxiety Scae; n, number of patients with the outcome; N, number of patients assessed. 150 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Per-protoco data coection tabe Outcomes Response QoL Withdrawas (a cause) Reapse rate Remission rate A-cause mortaity Adverse events (pease specify) 50% reduction in MADRS 50% reduction in HAMD Improvement from baseine on MADRS Improvement from baseine on CGI Improvement from baseine on HAMA Improvement from baseine on BPRS Tria scae Weight gain Time frame (weeks) SSRI + oanzapine SSRI + pacebo n N n N BPRS, Brief Psychiatric Rating Scae; HAMA, Hamiton Anxiety Scae; n, number of patients with the outcome; N, number of patients assessed. Did the RCT carry out any subgroup anayses of interest? (i.e. Different durations of depression, different casses of previous antidepressants, different genders, age, different severities of depression or different number of prior antidepressants.) If yes, pease give detais here: Did the RCT provide any detais of maintenance therapy +/ outcomes? Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 151

APPENDIX 3 PART FOUR: CLINICAL TRIAL QUALITY Pease describe the method of randomisation and aocation conceament used in the cinica tria: Pease describe the method of binding and who was binded in the cinica tria: How woud you describe the tria's design to minimise bias for (pease tick): Outcome Risk of bias Low risk Uncear risk High risk Comments Response 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias QoL 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias Withdrawas (a cause) 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias Reapse 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias Remission 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias 152 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Outcome Risk of bias Low risk Uncear risk High risk Comments A-cause mortaity 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias Adverse events 1) Random sequence generation 2) Aocation conceament 3) Binding (participants and personne) 4) Binding of outcomes assessment 5) Incompete outcome data 6) Seective reporting 7) Other bias How woud you rate the tria's overa risk of bias? Low risk Uncear High risk Do you have any additiona comments you woud ike to make about this cinica tria? Ideay, woud you ike further information about the cinica tria from the authors (if so, pease give detais)? Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 153

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 4 Tabe of excuded cinica effectiveness studies with rationae Reference detais Austin MP, Souza FG, Goodwin GM. Lithium augmentation in antidepressant-resistant patients. A quantitative anaysis. Br J Psychiatry 1991;159:510 14 Bauer M, Dopfmer S. Lithium augmentation in treatment-resistant depression: meta-anaysis of pacebo-controed studies. J Cin Psychopharmaco 1999;19:427 34 Bauer M, Adi M, Baethge C, Berghofer A, Sasse J, Heinz A, et a. Lithium augmentation therapy in refractory depression: cinica evidence and neurobioogica mechanisms. Can J Psychiatr Rev Canad Psychiatr 2003;48:440 8 Bauer M, Forsthoff A, Baethge C, Adi M, Berghofer A, Dopfmer S, et a. Lithium augmentation therapy in refractory depression-update 2002. Eur Arch Psychiatr Cin Neurosci 2003;253:132 9 Bauer M, Pretorius HW, Constant EL, Earey WR, Szamosi J, Brecher M, et a. Extended-reease quetiapine as adjunct to an antidepressant in patients with major depressive disorder: resuts of a randomized, pacebo-controed, doube-bind study. J Cin Psychiatr 2009;70:540 9 Bauer M, Adi M, Bschor T, Pihatsch M, Pfennig A, Sasse J, et a. Lithium's emerging roe in the treatment of refractory major depressive episodes: augmentation of antidepressants. Neuropsychobioogy 2010;62:36 42 Baumann P, Ni R, Souche A, Montadi S, Baettig D, Lambert S, et a. A doube-bind, pacebo-controed study of citaopram with and without ithium in the treatment of therapy-resistant depressive patients: a cinica, pharmacokinetic, and pharmacogenetic investigation. J Cin Psychopharmaco 1996;16:307 14 Bin W, Mei H, Po L. Fuoxetine and oanzapine in the treatment of refractory depression. Shandong Arch Psychiatry 2006;19:87 9 Bobo WV, Sheton RC. Fuoxetine and oanzapine combination therapy in treatment-resistant major depression: review of efficacy and safety data. Exp Opin Pharmacother 2009;10:2145 59 Bouton DW, Bach AH, Royzman K, Pate CG, Berman RM, Maikaarjun S, et a. The pharmacokinetics of standard antidepressants with aripiprazoe as adjunctive therapy: studies in heathy subjects and in patients with major depressive disorder. J Psychopharmaco 2010;24:537 46 Browne M, Lapierre YD, Hrdina PD, Horn E. Lithium as an adjunct in the treatment of major depression. Int Cin Psychopharmaco 1990;5:103 10 Bschor T, Bauer M. Efficacy and mechanisms of action of ithium augmentation in refractory major depression. Curr Pharm Des 2006;12:2985 92 Cooper C, Katona C, Lyketsos K, Bazer D, Brodaty H, Rabins P, et a. A systematic review of treatments for refractory depression in oder peope. Am J Psychiatr 2011;168:681 8 Crossey NA, Bauer M. Acceeration and augmentation of antidepressants with ithium for depressive disorders: two meta-anayses of randomized, pacebo-controed trias. J Cin Psychiatr 2007;68:935 40 Reason for excusion Systematic review Systematic review Systematic review Systematic review Did not meet popuation or intervention incusion criteria Systematic review Did not meet popuation or treatment duration incusion criteria Unabe to obtain a print copy Systematic review Did not report any outcomes of interest Did not meet popuation or intervention incusion criteria Systematic review Systematic review Systematic review Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 155

APPENDIX 4 Reference detais Denko TCF. Augmentation strategies in STAR*D: a review. Prim Psychiatr 2007;14:46 50 Dorée JP, Des Rosiers J, Lew V, Gendronc A, Eied R, Stipe E, et a. Quetiapine augmentation of treatment-resistant depression: a comparison with ithium. Curr Med Res Opin 2007;23:333 41 Dube S. Meta-anaysis of oanzapine-fuoxetine in treatment-resistant depression. 155th Annua Meeting of the American Psychiatric Association;18 23 May 2002, Phiadephia, PA, USA Dube S, Dube S, Andersen SW, Corya SA, Sanger TM, Toefson GD. Oanzapine-fuoxetine for treatment-resistant depression. XII Word Congress of Psychiatry, 24 29 August 2002, Yokohama, Japan Dunner DL, Amsterdam JD, Sheton RC, Loebe A, Romano SJ. Efficacy and toerabiity of adjunctive ziprasidone in treatment-resistant depression: a randomized, open-abe, piot study. J Cin Psychiatr 2007;68:1071 7 E-Khaii N, Joyce M, Atkinson S, Buynak RJ, Datto C, Lindgren P, et a. Extended-reease quetiapine fumarate (quetiapine XR) as adjunctive therapy in major depressive disorder (MDD) in patients with an inadequate response to ongoing antidepressant treatment: a muticentre, randomized, doube-bind, pacebo-controed study. Int J Neuropsychopharmaco 2010;13:917 32 Fava M, Rosenbaum JF, McGrath PJ, Stewart JW, Amsterdam JD, Quitkin FM. Lithium and tricycic augmentation of fuoxetine treatment for resistant major depression: a doube-bind, controed study. Am J Psychiatr 1994;151:1372 4 Fava M, Apert J, Nierenberg A, Lagomasino I, Sonawaa S, Tedow J, et a. Doube-bind study of high-dose fuoxetine vs ithium or desipramine augmentation of fuoxetine in partia responders and nonresponders to fuoxetine. J Cin Psychopharmaco 2002;22:379 87 Feurence R, Wiiamson R, Jing Y, Kim E, Tran QV, Pikaov AS, et a. A systematic review of augmentation strategies for patients with major depressive disorder. Psychopharmaco Bu 2009;42:57 90 Kanto D, McNevin S, Leichner P, Harper D, Krenn M. The benefit of ithium carbonate adjunct in refractory depression fact or fiction? Can J Psychiatr Rev Canad Psychiatr 1986;31:416 18 Keitner GI, Garow SJ, Ryan CE, Ninan PT, Soomon DA, Nemeroff CB, et a. A randomized, pacebo-controed tria of risperidone augmentation for patients with difficut-to-treat unipoar, non-psychotic major depression. J Psychiatr Res 2009;43:205 14 Kennedy SH, Lam RW, Cohen NL, Ravindran AV. Cinica guideines for the treatment of depressive disorders. IV. Medications and other bioogica treatments. Can J Psychiatr Rev Canad Psychiatr 2001;46(Supp. 1):38 58 Li H, Zhang Y, Zhang Y. Doube-bind study of fuoxetine augmented with oanzapine in the treatment of treatment-resistant depression. Shandong Arch Psychiatry 2006;19:85 6 Mahmoud RA, Pandina GJ, Turkoz I, Kosik-Gonzaez C, Canuso CM, Kujawa MJ, et a. Risperidone for treatment-refractory major depressive disorder: a randomized tria. Ann Int Med 2007;147:593 602 Neson JC, Papakostas GI. Atypica antipsychotic augmentation in major depressive disorder: a meta-anaysis of pacebo-controed randomized trias. Am J Psychiatr 2009;166:980 91 Papakostas GI, Sheton RC, Smith J, Fava M. Augmentation of antidepressants with atypica antipsychotic medications for treatmentresistant major depressive disorder: a meta-anaysis. J Cin Psychiatr 2007;68:826 31 Reason for excusion Systematic review Did not meet popuation or intervention incusion criteria Unabe to obtain and a systematic review Unabe to obtain a print copy Decision to excude based on high doses of ziprasidone and not used in the UK Did not meet popuation or intervention incusion criteria Did not meet popuation or intervention incusion criteria Did not meet popuation or intervention incusion criteria Unabe to obtain a print copy Did not meet popuation or intervention incusion criteria Did not meet popuation or intervention incusion criteria Systematic review Unabe to obtain a print copy Did not meet popuation or intervention incusion criteria Systematic review Systematic review 156 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Reference detais Patten SB, Lupin DA, Boucher SA, Lamarre CJ. Pharmacoogic management of refractory depression. CMAJ 1992;146:483 7 Phiip NS, Carpenter LL, Tyrka AR, Price LH. Augmentation of antidepressants with atypica antipsychotics: a review of the current iterature. J Psychiatr Pract 2008;14:34 44 Price LHC. Lithium augmentation for refractory depression: a critica reappraisa. Prim Psychiatr 2008;15:35 42 Rapaport MH, Gharabawi GM, Canuso CM, Mahmoud RA, Keer MB, Bossie CA, et a. Effects of risperidone augmentation in patients with treatment-resistant depression: Resuts of open-abe treatment foowed by doube-bind continuation. Neuropsychopharmacoogy 2006;31:2505 13 Rouion F, Gorwood P. The use of ithium to augment antidepressant medication. J Cin Psychiatr 1998;59:32 41 Santos MAH. Treatment-resistant depression: review of pharmacoogic antidepressant strategies. J Bras Psiquiatr 2006;55:232 42 Savas HAK. Aripiprazoe in the treatment of depression: a review. Kinik Psikofarmakooji Buteni 2010;20:S26 30 Schmauss M, Erfurth A. [Combination therapies in antidepressive drug refractory depression: an overview.] Fortschr Neuroogie Psych 1996;64:390 402 Schmauss M, Messer T. [Augmentation strategies for therapy resistant depression a review.] Psychiat Prax 2007;34:165 74 Schweitzer I, Tuckwe V, Johnson G. A review of the use of augmentation therapy for the treatment of resistant depression: impications for the cinician. Aust NZ J Psychiat 1997;31:340 52 Seis MA, Peeters FP. [Augmentation with atypica antipsychotics for the treatment of patients with a therapy-resistant depression: a review.] Tijdschr Psychiatr 2008;50:213 22 Souche A, Montadi S, Uehinger C, Kasas A, Reymond MJ, Reymond P, et a. [Treatment of resistant depression with the citaopram ithium combination. Methodoogy of a doube-bind muticenter study and preiminary resuts.] Encephae 1991;17:213 19 Stefanescu C, Mavros M, Chirita V. Resistant depression: a comparison between antidepressants and the efficacy of oanzapine augmentation. Psychiatriki 2005;16:252 Stimpson N, Agrawa N, Lewis G. Randomised controed trias investigating pharmacoogica and psychoogica interventions for treatment-refractory depression. Systematic review. Br J Psychiatry 2002;181:284 94 Toefson G, Gannon K, Jacobs T, Sheton R, Tohen M, Stah S. The study of oanzapine pus fuoxetine in treatment-resistant major depressive disorder without psychotic features. XI Word Congress of Psychiatry, 6 11 August 1999, Hamburg, Germany, Abstracts Voume II:133 Trivedi MH, Thase ME, Osuntokun O, Heney DB, Case M, Watson SB, et a. An integrated anaysis of oanzapine/fuoxetine combination in cinica trias of treatment-resistant depression. J Cin Psychiatr 2009;70:387 96 Tyrer P, Marsden CA, Casey P, Seivewright N. Cinica efficacy of paroxetine in resistant depression. J Psychopharmaco 1987;1:251 7 Reason for excusion Systematic review Systematic review Systematic review Did not meet incusion criteria of acute phase treatment Systematic review Systematic review Systematic review Systematic review Systematic review Systematic review Systematic review Did not meet treatment duration incusion criteria Unabe to obtain a print copy Systematic review Unabe to obtain a print copy Systematic review Unabe to obtain a print copy Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 157

APPENDIX 4 Reference detais Wang X-H, Guo X. Effectiveness and safety of oanzapine combined with fuoxetine for refractory depression: a systematic review. Chin J Evid-Based Med 2010;10:1102 9 Zhornitsky S, Potvin S, Moteshafi H, Dubreucq S, Rompre PP, Stip E. Dose-response and comparative efficacy and toerabiity of quetiapine across psychiatric disorders: a systematic review of the pacebo-controed monotherapy and add-on trias. Int Cin Psychopharmaco 2011; 26:183 92 Zusky PM, Biederman J, Rosenbaum JF, et a. Adjunct ow dose ithium carbonate in treatment-resistant depression: a pacebo-controed study. J Cin Psychopharmaco 1988;8:120 4 Reason for excusion Systematic review Systematic review Did not meet popuation or intervention incusion criteria 158 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 5 Quaity assessment of cinica trias Pease note that the risk-of-bias anayses reported beow are reported ony for the outcomes for which the tria has been incuded in an anaysis for that outcome. Berman et a. 2007 47 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Withdrawas (a cause) Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 159

APPENDIX 5 Berman et a. 2009 48 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Withdrawas (a cause) Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk 160 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Corya et a. 2006 49 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on tria Aocation conceament methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on tria Aocation conceament methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Remission Random sequence generation Limited detais reported on tria Aocation conceament methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Withdrawas (a cause) Random sequence generation Limited detais reported on tria Aocation conceament methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Adverse events Random sequence generation Limited detais reported on tria Aocation conceament methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 161

APPENDIX 5 Feng et a. 2008 43 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported Aocation conceament on tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Adverse events Random sequence generation Limited detais reported Aocation conceament on tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Remission rate Random sequence generation Limited detais reported Aocation conceament on tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk 162 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Franco et a. 2010 30 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Katona et a. 1995 59 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on tria Aocation conceament methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on tria Aocation conceament methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 163

APPENDIX 5 Outcome Withdrawas (a cause) Risk of bias Low risk Uncear risk High risk Comments Random sequence generation Limited detais reported on tria Aocation conceament methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Adverse events Random sequence generation Limited detais reported on tria Aocation conceament methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Mattingy et a. 2006 46 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk 164 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Marcus et a. 2008 50 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Withdrawas (a cause) Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Sheton et a. 2001 51 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on tria methodoogy Aocation conceament Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 165

APPENDIX 5 Outcome Risk of bias Low risk Uncear risk High risk Comments Withdrawas (a cause) Random sequence generation Limited detais reported on tria methodoogy Aocation conceament Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Adverse events Random sequence generation Limited detais reported on tria methodoogy Aocation conceament Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Sheton et a. 2005 52 Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Aocation conceament Limited detais reported on tria methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Mean change in MADRS Random sequence generation Aocation conceament Limited detais reported on tria methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias 166 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Outcome Risk of bias Low risk Uncear risk High risk Comments Remission Random sequence generation Aocation conceament Limited detais reported on tria methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Withdrawas (a cause) Random sequence generation Aocation conceament Limited detais reported on tria methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Adverse events Other bias Random sequence generation Aocation conceament Limited detais reported on tria methodoogy. Potentia issue of non-binding of investigators at Binding (participants and personne) randomisation: patients might not be as treatment resistant as in Binding of outcomes assessment other trias Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk Thase et a. 2007 53 (studies a and b ) Outcome Risk of bias Low risk Uncear risk High risk Comments Response Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 167

APPENDIX 5 Outcome Mean change in MADRS Risk of bias Low risk Uncear risk High risk Comments Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Remission Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Withdrawas (a cause) Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Adverse events Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias QoL Random sequence generation Limited detais reported on Aocation conceament tria methodoogy Binding (participants and personne) Binding of outcomes assessment Incompete outcome data Seective reporting Other bias Overa tria risk-of-bias rating: Low risk Uncear High risk 168 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 6 Summary of trias incuded for each outcome in the network meta-anaysis Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 169

APPENDIX 6 Outcome Study Response (using ithium a priori data) a Response (using ithium post hoc data) a,b Mean change in MADRS score from baseine a,c Withdrawas (a cause) d Tota adverse Dry SA SA SA SA SA events d Somnoence a mouth a Headache a 1 a 2 a 3 a 4 a 5 d Berman et a. (2007) 47 Berman et a. (2009) 48 Corya et a. (2006) 49 Feng et a. (2008) 43 Franco et a. (2010) 30 Katona et a. (1995) 59 Marcus et a. (2008) 50 Mattingy et a. (2006) 46 Sheton et a. (2001) 51 Sheton et a. (2005) 52 Thase et a. (2007a) 53 e e e e Thase et a. (2007b) 53 e e e e a Random-effects mode. b Primary anaysis for this review. c (Weighted) MD. d Fixed-effects mode. e Thase et a. 53 (a+b) pooed study data used for this outcome. Notes SA 1 cass-based sensitivity anaysis for response. SA 2 sensitivity anaysis for response imiting to trias reporting faiure to two or more antidepressants in the current episode of depression. SA 3 sensitivity anaysis for response imiting to trias reporting response based on the MADRS scae. SA 4 cass-based sensitivity anaysis for mean change in MADRS. SA 5 cass-based sensitivity anaysis for discontinuations. 170 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 7 Heath economics iterature search strategies Economics search strategy Database: EMBASE Search strategy 1. *depression/ or *major depression/ 2. depress$.tw. 3. or/1-2 4. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. 5. drug treatment faiure/ 6. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. 7. or/4-6 8. (anti-depress$ or antidepress$).tw. 9. antidepressant agent/ 10. 8 or 9 11. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. 12. drug combination/ 13. 11 or 12 14. exp ithium/ or exp ithium derivative/ or ithium carbonate/ or ithium choride/ or ithium citrate/ 15. atypica antipsychotic agent/ or ziprasidone/ or risperidone/ or quetiapine/ or paiperidone/ or oanzapine/ or cozapine/ or cozapine derivative/ or cozapine n oxide/ or aripiprazoe/ or amisupride/ 16. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. 17. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. 18. or/14-17 19. serotonin antagonist/ or serotonin uptake inhibitor/ or fuoxetine pus oanzapine/ or fuvoxamine maeate/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. 20. (citaopram or citaopramum or ceexa or ciprami).tw. or citaopram/ 21. (escitaopram or S-citaopram or exapro or cipraex).tw. or escitaopram/ 22. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or fuoxetine/ or fuoxetine pus oanzapine/ 23. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or fuvoxamine/ or fuvoxamine maeate/ 24. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or paroxetine/ 25. (sertrain$ or ustra or zooft).tw. or sertraine/ 26. or/19-25 27. 3 and 7 28. 10 and 13 and 18 29. 26 or 28 30. 27 and 29 31. heath economics/ Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 171

APPENDIX 7 32. exp economic evauation/ 33. exp pharmacoeconomics/ 34. exp heath care cost/ 35. or/31-34 36. (price or prices or pricing).tw. 37. vaue for money.tw. 38. (economic$ or pharmaeconomic$ or pharmacoeconomic$ or pharmaco-economic$).tw. 39. (cost or costs or costy or costing or costed).tw. 40. or/36-39 41. 35 or 40 42. etter.pt. 43. editoria.pt. 44. note.pt. 45. or/42-44 46. 41 not 45 47. 30 and 46 Database: MEDLINE Search strategy 1. Depression/ or Depressive Disorder/ or Depressive Disorder, Major/ 2. depress$.tw. 3. or/1-2 4. treatment faiure/ 5. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. 6. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. 7. or/4-6 8. 3 and 7 9. (anti-depress$ or antidepress$).tw. 10. Antidepressive Agents/ 11. 9 or 10 12. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. 13. Drug Therapy, Combination/ 14. 12 or 13 15. Serotonin Uptake Inhibitors/ or Antidepressive Agents, Second-Generation/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. or Serotonin Antagonists/ 16. (citaopram or citaopramum or ceexa or ciprami).tw. or Citaopram/ 17. (escitaopram or S-citaopram or exapro or cipraex).tw. 18. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or Fuoxetine/ 19. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or Fuvoxamine/ 20. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or Paroxetine/ 21. (sertrain$ or ustra or zooft).tw. or Sertraine/ 22. or/15-21 23. Lithium/ or exp Lithium Compounds/ 24. antipsychotic agents/ or cozapine/ or risperidone/ 25. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. 172 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 26. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. 27. or/23-26 28. 11 and 14 and 27 29. 22 or 28 30. 8 and 29 31. economics/ 32. exp costs/ and cost anaysis/ 33. exp economics, hospita/ 34. economics, medica/ 35. economics, pharmaceutica/ 36. (economic$ or pharmaeconomic$ or pharmacoeconomic$ or pharmaco-economic$).tw. 37. (cost or costs or costy or costing or costed).tw. 38. (price or prices or pricing).tw. 39. vaue for money.tw. 40. or/31-39 41. etter.pt. 42. editoria.pt. 43. comment.pt. 44. or/41-43 45. 40 not 44 46. 30 and 45 Database: PsycINFO Search strategy 1. costs.mp. and cost anaysis/ [mp = tite, abstract, heading word, tabe of contents, key concepts, origina tite, tests & measures] 2. cost containment/ 3. pharmacoeconomics/ 4. heath care economics/ 5. or/1-4 6. (economic adj2 evauation$).tw. 7. (economic adj2 anay$).tw. 8. (economic adj2 (study or studies)).tw. 9. (cost adj2 evauation$).tw. 10. (cost adj2 anay$).tw. 11. (cost adj2 (study or studies)).tw. 12. (cost adj2 effective$).tw. 13. (cost adj2 benefit$).tw. 14. (cost adj2 utii$).tw. 15. (cost adj2 minimi$).tw. 16. (cost adj2 consequence$).tw. 17. (cost adj2 comparison$).tw. 18. (cost adj2 identificat$).tw. 19. (pharmaeconomic$ or pharmacoeconomic$ or pharmaco-economic$).tw. 20. or/6-19 21. 5 or 20 22. editoria.dt. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 173

APPENDIX 7 23. etter.dt. 24. or/22-23 25. 21 not 24 26. exp Major Depression/ 27. depress$.tw. 28. or/26-27 29. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. 30. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. 31. or/29-30 32. antidepressant drugs/ 33. (antidepress$ or anti-depress$).tw. 34. or/32-33 35. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. 36. poypharmacy/ 37. Drug Augmentation/ 38. or/35-37 39. Serotonin Antagonists/ or serotonin reuptake inhibitors/ or (ssri$ or (serotonin adj2 inhibitor$)).tw. 40. (citaopram or citaopramum or ceexa or ciprami).tw. or Citaopram/ 41. (escitaopram or S-citaopram or exapro or cipraex).tw. 42. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. 43. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or Fuvoxamine/ 44. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or Paroxetine/ 45. or/39-44 46. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. 47. exp Lithium Carbonate/ or Lithium/ 48. neuroeptic drugs/ or aripiprazoe/ or cozapine/ or oanzapine/ or quetiapine/ or risperidone/ 49. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. 50. or/46-49 51. 28 and 31 52. 34 and 38 and 50 53. 45 or 52 54. 51 and 53 55. 25 and 54 Quaity-of-ife search strategies Database: EMBASE Quaity-of-ife search strategy 1. exp *quaity of ife/ 2. quaity of ife.tw. 3. (quaity adjusted ife year$ or quaity-adjusted ife year$).tw. 4. qay$.tw. 5. qo.tw. 174 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 6. hrqo.tw. 7. euroqo.tw. 8. short form 36.tw. 9. (disabiity adjusted ife year$ or disabiity-adjusted ife year$).tw. 10. ((vaue or vauation or quaity) adj2 (ife or ives or surviva)).tw. 11. or/1-10 12. etter.pt. 13. editoria.pt. 14. note.pt. 15. or/12-14 16. 11 not 15 17. *depression/ or *major depression/ 18. depress$.tw. 19. or/17-18 20. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. 21. drug treatment faiure/ 22. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. 23. or/20-22 24. (anti-depress$ or antidepress$).tw. 25. antidepressant agent/ 26. 24 or 25 27. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. 28. drug combination/ 29. 27 or 28 30. exp ithium/ or exp ithium derivative/ or ithium carbonate/ or ithium choride/ or ithium citrate/ 31. atypica antipsychotic agent/ or ziprasidone/ or risperidone/ or quetiapine/ or paiperidone/ or oanzapine/ or cozapine/ or cozapine derivative/ or cozapine n oxide/ or aripiprazoe/ or amisupride/ 32. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. 33. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. 34. or/30-33 35. serotonin antagonist/ or serotonin uptake inhibitor/ or fuoxetine pus oanzapine/ or fuvoxamine maeate/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. 36. (citaopram or citaopramum or ceexa or ciprami).tw. or citaopram/ 37. (escitaopram or S-citaopram or exapro or cipraex).tw. or escitaopram/ 38. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or fuoxetine/ or fuoxetine pus oanzapine/ 39. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or fuvoxamine/ or fuvoxamine maeate/ 40. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or paroxetine/ 41. (sertrain$ or ustra or zooft).tw. or sertraine/ 42. or/35-41 43. 19 and 23 44. 26 and 29 and 34 45. 42 or 44 46. 43 and 45 47. 16 and 46 Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 175

APPENDIX 7 Database: MEDLINE Quaity-of-ife search strategy 1. quaity of ife/ 2. vaue of ife/ 3. quaity-adjusted ife years/ 4. cost of iness/ 5. quaity of ife.tw. 6. (quaity adjusted ife year$ or quaity-adjusted ife year$).tw. 7. qay$.tw. 8. qo.tw. 9. hrqo.tw. 10. euroqo.tw. 11. short form 36.tw. 12. (disabiity adjusted ife year$ or disabiity-adjusted ife year$).tw. 13. ((vaue or vauation or quaity) adj2 (ife or ives or surviva)).tw. 14. or/1-4 15. or/5-13 16. 14 or 15 17. etter.pt. 18. editoria.pt. 19. comment.pt. 20. or/17-19 21. 16 not 20 22. Depression/ or Depressive Disorder/ or Depressive Disorder, Major/ 23. depress$.tw. 24. or/22-23 25. treatment faiure/ 26. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. 27. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. 28. or/25-27 29. 24 and 28 30. (anti-depress$ or antidepress$).tw. 31. Antidepressive Agents/ 32. 30 or 31 33. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. 34. Drug Therapy, Combination/ 35. 33 or 34 36. Serotonin Uptake Inhibitors/ or Antidepressive Agents, Second-Generation/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. or Serotonin Antagonists/ 37. (citaopram or citaopramum or ceexa or ciprami).tw. or Citaopram/ 38. (escitaopram or S-citaopram or exapro or cipraex).tw. 39. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. or Fuoxetine/ 40. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or Fuvoxamine/ 41. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or Paroxetine/ 42. (sertrain$ or ustra or zooft).tw. or Sertraine/ 43. or/36-42 44. Lithium/ or exp Lithium Compounds/ 45. antipsychotic agents/ or cozapine/ or risperidone/ 176 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 46. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. 47. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. 48. or/44-47 49. 32 and 35 and 48 50. 43 or 49 51. 29 and 50 52. 21 and 51 Database: PsycINFO Quaity-of-ife search strategy 1. quaity of ife/ 2. quaity of ife.tw. 3. (quaity adjusted ife year$ or quaity-adjusted ife year$).tw. 4. qay$.tw. 5. qo.tw. 6. hrqo.tw. 7. euroqo.tw. 8. short form 36.tw. 9. (disabiity adjusted ife year$ or disabiity-adjusted ife year$).tw. 10. ((vaue or vauation or quaity) adj2 (ife or ives or surviva)).tw. 11. or/1-10 12. editoria.dt. 13. etter.dt. 14. or/12-13 15. 11 not 14 16. exp Major Depression/ 17. depress$.tw. 18. or/16-17 19. (refract$ or resistan$ or nonrespon$ or unrespon$ or fai$ or (incompet$ adj respon$) or (no$ adj2 respon$)).tw. 20. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. 21. or/19-20 22. antidepressant drugs/ 23. (antidepress$ or anti-depress$).tw. 24. or/22-23 25. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. 26. poypharmacy/ 27. Drug Augmentation/ 28. or/25-27 29. Serotonin Antagonists/ or serotonin reuptake inhibitors/ or (ssri$ or (serotonin adj2 inhibitor$)).tw. 30. (citaopram or citaopramum or ceexa or ciprami).tw. or Citaopram/ 31. (escitaopram or S-citaopram or exapro or cipraex).tw. 32. (fuoxetin$ or fuokset$ or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax).tw. 33. (fuvoxamin$ or fuvoksami$ or faverin or uvox).tw. or Fuvoxamine/ 34. (paroxetin$ or parokset$ or seroxat or paxi or pexeva).tw. or Paroxetine/ Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 177

APPENDIX 7 35. or/29-34 36. (Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs).tw. 37. exp Lithium Carbonate/ or Lithium/ 38. neuroeptic drugs/ or aripiprazoe/ or cozapine/ or oanzapine/ or quetiapine/ or risperidone/ 39. ((atypica adj antipsychotic?) or (atypica adj anti-psychotic?) or Amisuprid$ or Soian or Aripiprazo$ or Abiify or Cozapin$ or Cozari or Kotsapiini or Kozapin$ or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin$ or Oantsapiini or Zyprexa or Zypadhera or Paiperidon$ or Invega or Xepion or Quetiapin$ or Seroque or Risperidon$ or Riszperidon or Rysperydon or Risperda or Ziprasidon$ or Geodon).tw. 40. or/36-39 41. 18 and 21 42. 24 and 28 and 40 43. 35 or 42 44. 41 and 43 45. 15 and 44 Database: NHS Economic Evauation Database, Heath Technoogy Assessment Database Search strategy #1 Depression #2 Depressive Disorder #3 Mood Disorders #4 (depress* or dysphori* or dysthym* or seasona affective disorder* or meancho*): ti or (depress* or dysphori* or dysthym* or seasona affective disorder* or meancho*):ab #5 (#1 OR #2 OR #3 OR #4) #6 ithium #7 atypica antipsychotics #8 serotonin reuptake inhibitors OR SSRI #9 (#6 OR #7 OR #8) #10 (#5 AND #9) Database: The Cochrane Library Search strategy #1 Depressive Disorder in Economic Evauations #2 Depression in Economic Evauations #3 depress*:ti,ab in Economic Evauations 178 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 #4 Major Depressive Disorder in Economic Evauations #5 (#1 OR #2 OR #3 OR #4) #6 Antidepressive Agents OR Second-Generation in Economic Evauations #7 Serotonin Antagonists in Economic Evauations #8 Serotonin Uptake Inhibitors in Economic Evauations #9 Citaopram in Economic Evauations #10 Fuoxetine in Economic Evauations #11 Fuvoxamine in Economic Evauations #12 Paroxetine in Economic Evauations #13 Sertraine in Economic Evauations #14 ssri* or (serotonin near/2 inhibitor*) or citaopram or citaopramum or ceexa or ciprami or escitaopram or S-citaopram or exapro or cipraex or fuoxetin* or fuokset* or feicium or prozac or prozep or prozit or sarafem or sefemra or symbyax or fuvoxamin* or fuvoksami* or faverin or uvox or paroxetin* or parokset* or seroxat or paxi or pexeva or sertrain* or ustra or zooft in Economic Evauations #15 (#6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14) #16 refract* or resistan* or nonrespon* or (no* near/2 respon*) or unrespons* or fai* or (incompet* near/1 respon*) or (inadequat* near/1 respons*) or (sub* near/2 respon*) or (poor* near/1 respon*): ti,ab in Economic Evauations #17 Treatment Faiure in Economic Evauations #18 (#16 OR #17) #19 antidepress* or anti-depress*:ti,ab in Economic Evauations #20 Antidepressive Agents in Economic Evauations #21 (#19 OR #20) #22 augment* or adjunct* or combin* or add* or potentiation:ti,ab in Economic Evauations #23 Drug Therapy in Economic Evauations #24 (#22 OR #23) #25 Lithium or Camcoit or Liskonum or Priade or Li-Liquid or Litarex or Lithonate or Phasa or Lithny or Licio or Lithii or Litio or Litium or Lityum or diithium or itu or Cibaith-S or Eskaith or Lithane or Lithobid or Lithonate or Lithotabs in Economic Evauations #26 (atypica near/1 antipsychotic*) or (atypica near/1 anti-psychotic*) or Amisuprid* or Soian or Aripiprazo* or Abiify or Cozapin* or Cozari or Kotsapiini or Kozapin* or FazaCo or Fazaco or Denzapine or Zaponex or Oanzapin* or Oantsapiini or Zyprexa or Zypadhera or Paiperidon* or Invega or Xepion or Quetiapin* or Seroque or Risperidon* or Riszperidon or Rysperydon or Risperda or Ziprasidon* or Geodon in Economic Evauations #27 Lithium in Economic Evauations #28 Lithium Compounds in Economic Evauations #29 Antipsychotic Agents in Economic Evauations continued Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 179

APPENDIX 7 #30 Cozapine in Economic Evauations #31 Risperidone in Economic Evauations #32 (#25 OR #26 OR #27 OR #28 OR #29 OR #30 OR #31) #33 (#5 AND 15 AND #18 AND #21 AND #24 AND #32) 180 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 8 Quaity assessment of incuded studies (cost-effectiveness studies) Leeahanaj 2010 88 Dimension of quaity Structure S1: Statement of decision probem/objective S2: Statement of scope/ perspective Comments Ceary stated The mode scope and perspective (Thai) are ceary stated and the mode outcomes are consistent with the scope and overa objective of the mode S3: Rationae for structure Athough no justification was given for adopting a decision tree, the author's approach is considered appropriate given that the mode focused on acute treatment of MDD. The mode was we constructed and data sources were we described S4: Structura assumptions The mode is considered to be we constructed and assumptions ceary stated. However, it is important to note that authors made some very optimistic assumptions that may not be credibe. For instance adverse events were not modeed and there was no justification as to why they did not incude them. Aso the assumption that a patients who are given ECT wi enter remission may not necessariy be true S5: Strategies/comparators The mode compared adjunctive therapy with aripiprazoe or pacebo S6: Mode type Correct, cost utiity anaysis S7: Time horizon Six weeks matching the duration of tria data S8: Disease states/pathways The pathways/heath states modeed are reevant. However, authors coud have aso incuded response rates as a separate heath state as this is one of the most common reported outcomes in MDD S9: Cyce ength This was a decision tree that captured acute treatment outcomes Data D1: Data identification Data were systematicay sourced, ceary described and justified by the authors D2: Pre-mode data anaysis The authors did not report how they synthesised the tria data they used in the mode. Data derived from the trias were directy impemented in the mode D2a: Baseine data D2b: Treatment effects D2c: Costs D2d: Quaity-of-ife weights (utiities) Events reported in either pacebo or aripiprazoe were used to popuate the decision tree Events reported in either pacebo or aripiprazoe were used to popuate the decision tree. RRs and 95% CrIs were cacuated for aripiprazoe Appropriate costs were incuded and source of cost data ceary described Derived from iterature and ceary referenced D3: Data incorporation The authors ceary described how data were used in the mode D4: Assessment of uncertainty The assessment of sensitivity was thorough and robust D4a: Methodoogica D4b: Structura D4c: Heterogeneity D4d: Parameter The authors used appropriate anaytica methods for the decision probem The authors described deterministic sensitivity anaysis and scenario anaysis undertaken for the mode Heterogeneity was not addressed, no subgroup anaysis was undertaken OWSA and scenario anaysis were undertaken. PSA was not carried out Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 181

APPENDIX 8 Dimension of quaity Comments Consistency C1: Interna consistency The mode seems to be mathematicay sound with no obvious inconsistencies C2: Externa consistency The resuts of the mode are appicabe to patients with MDD. Efficacy data were taken from two US studies that may not be directy appicabe to the UK popuation. Costs data were taken from oca Thai hospita, which has a different heath-care system to the UK. Overa the study findings and concusions cannot be generaised to the UK setting Xie 2009 87 Dimension of quaity Structure S1: Statement of decision probem/objective S2: Statement of scope/ perspective Comments Ceary stated The mode scope and perspective are ceary stated and the mode outcomes are consistent with the scope and overa objective of the mode. The anaysis was conducted from a Singaporean societa perspective S3: Rationae for structure Athough no justification was given for adopting a decision tree, the authors' approach is considered appropriate and is in ine with other pubished economic modes in the area of MDD. The mode was we constructed and data sources were we described S4: Structura assumptions The mode is considered to be we constructed and assumptions ceary stated. However, it is important to note that authors made some very optimistic assumptions that may not be credibe. Owing to ack of head-to-head tria of escitaopram vs. fuvoxamine, authors used efficacy data for citaopram as a proxy for the efficacy of fuvoxamine. Aso the assumption that reapse rates foowing augmentation or combination therapy are the same foowing first-ine treatment contradicts externa evidence S5: Strategies/comparators The mode compared escitaopram vs. venafaxine and vs. fuvoxamine S6: Mode type Cost-effectiveness anaysis S7: Time horizon Six months foowing the initiation of therapy, in accordance with the Singaporean depression cinica guideine S8: Disease states/pathways The pathways/heath states modeed are reevant. However, authors coud have aso incuded response rates as a separate heath state as this is one of the most common reported outcomes in MDD S9: Cyce ength This was a decision tree which captured 6 months' treatment outcomes Data D1: Data identification Data were systematicay sourced, ceary described and justified by the authors D2: Pre-mode data anaysis The authors did not report any pre-mode data anaysis D2a: Baseine data D2b: Treatment effects D2c: Costs D2d: Quaity-of-ife weights (utiities) Events reported in the incuded studies were used in the mode. Remission rates were taken from a referenced meta-anaysis and as stated earier data for citaopram were used as a proxy for the efficacy of fuvoxamine Events reported in either escitaopram vs. venafaxine or escitaopram vs. fuvoxamine and 95% CrIs were reported Costs appropriate to the perspective of the anaysis were used, sources were ceary stated Not reported, this was not a cost utiity anaysis D3: Data incorporation The authors ceary described how data were used in the mode D4: Assessment of uncertainty The assessment of sensitivity was thorough and robust as both deterministic and PSA were reported 182 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Dimension of quaity D4a: Methodoogica D4b: Structura D4c: Heterogeneity D4d: Parameter Comments The authors used appropriate anaytica methods for the decision probem; however, they coud have undertaken a cost utiity study The authors described deterministic sensitivity anaysis and PSA was undertaken for the mode Heterogeneity was partiay addressed as the mode considered primary-care patients and secondary care in separate anaysis. However, there was no further subgroup anaysis reported Both deterministic and PSA were reported Consistency C1: Interna consistency The mode seems to be mathematicay sound with no obvious inconsistencies C2: Externa consistency The resuts of the mode are appicabe to patients with MDD. Efficacy data were taken from studies conducted in the US, which may not be directy appicabe to the UK popuation. Costs data were estimated from a survey of oca practitioners in Singapore, which has a different heath-care system to the UK. Overa, the study findings and concusions cannot be generaised to the UK setting Benedict 2010 86 Dimension of quaity Structure S1: Statement of decision probem/objective S2: Statement of scope/ perspective Comments Ceary stated The mode scope and perspective are ceary stated and the mode outcomes are consistent with the scope and overa objective of the mode. The mode was conducted from a Scottish NHS perspective S3: Rationae for structure Athough no justification was given for adopting a Markov, the authors' approach is considered appropriate given the cycica nature of the disease S4: Structura assumptions The mode is considered to be we constructed and assumptions ceary stated S5: Strategies/comparators The mode compared duoxetine vs. venafaxine ER or mirtazapine S6: Mode type Correct, cost utiity anaysis S7: Time horizon Forty-eight-week time horizon chosen to capture treatment duration as recommended by NICE [6 months foowing remission (NICE 2007)] and reapses within 1 year S8: Disease states/pathways The pathways/heath states modeed are reevant S9: Cyce ength Eight-week cyce ength was chosen, which is the typica treatment duration of most trias in MDD Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 183

APPENDIX 8 Dimension of quaity Comments Data D1: Data identification Data were systematicay sourced, ceary described and justified by the authors D2: Pre-mode data anaysis The authors did not report any pre-mode anaysis. Data derived from different meta-anaysis for different parameters was directy impemented in the mode D2a: Baseine data D2b: Treatment effects D2c: Costs D2d: Quaity-of-ife weights (utiities) Events reported in different meta-anaysis studies were used to popuate the mode Treatment effect reported in different meta-anaysis studies were used to popuate the mode Appropriate costs incuded in base case and sensitivity anaysis, sources are transparent Derived from iterature and ceary referenced D3: Data incorporation The authors ceary described how data were used in the mode D4: Assessment of uncertainty The assessment of sensitivity was thorough and robust as both deterministic and PSA were undertaken D4a: Methodoogica D4b: Structura D4c: Heterogeneity D4d: Parameter The authors used appropriate anaytica methods for the decision probem The authors described deterministic sensitivity anaysis and PSA was aso undertaken Heterogeneity was partiay assessed by reporting resuts separatey for primary and secondary-care patients; however, no further subgroup anaysis was undertaken OWSA and PSA were undertaken Consistency C1: Interna consistency The mode seems to be mathematicay sound with no obvious inconsistencies C2: Externa consistency The resuts of the mode are appicabe to patients with MDD. Resource-use data were provided by a pane of practising practitioners in Scotand and refected the Scottish practice. Overa, the study findings and concusions cannot be directy generaised to the UK setting but are nonetheess informative Simpson 2009 89 Dimension of quaity Structure S1: Statement of decision probem/objective S2: Statement of scope/ perspective Comments Ceary stated The mode scope and perspective are ceary stated and the mode outcomes are consistent with the scope and overa objective of the mode. The mode was conducted from a US payer and societa perspective S3: Rationae for structure No justification was given for adopting a hybrid mode structure, i.e. a combination of decision tree for the acute phase of the mode and then a Markov for the maintenance phase of the mode. However, this approach is considered appropriate S4: Structura assumptions The mode is considered to be we constructed and assumptions ceary stated S5: Strategies/comparators The mode compared TMS with sham TMS and pharmacotherapy as usua S6: Mode type Cost utiity anaysis S7: Time horizon One-year time horizon chosen to capture disease progression. Acute and taper phase outcomes were captured at 6 weeks using decision tree and the Markov was extended for 1 year S8: Disease states/pathways The pathways/heath states modeed are reevant S9: Cyce ength Three-monthy cyces and no justification was given for the chosen cyce ength 184 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Dimension of quaity Comments Data D1: Data identification Data were systematicay sourced, ceary described and justified by the authors D2: Pre-mode data anaysis The authors did not report any pre-mode anaysis. Data derived from different meta-anaysis for different parameters was directy impemented in the mode D2a: Baseine data D2b: Treatment effects D2c: Costs D2d: Quaity-of-ife weights (utiities) Events reported in three different studies were used to popuate the mode Treatment effect reported in three different studies (referenced) incuding the STAR*D were used to popuate the mode Adverse event costs were excuded, otherwise incuded costs were appropriate and resources ceary stated Derived from iterature and ceary referenced D3: Data incorporation The authors ceary described how data were used in the mode D4: Assessment of uncertainty The assessment of sensitivity was thorough and robust as both deterministic and PSA were undertaken D4a: Methodoogica D4b: Structura D4c: Heterogeneity D4d: Parameter The authors used appropriate anaytica methods for the decision probem The authors described deterministic sensitivity anaysis and PSA was aso undertaken Heterogeneity was assessed by reporting resuts separatey by depression severity, i.e. mid, moderate and severey depressed OWSA and PSA were undertaken Consistency C1: Interna consistency The mode seems to be mathematicay sound, with no obvious inconsistencies C2: Externa consistency The resuts of the mode are appicabe to patients with MDD. Resource-use data were obtained from individua patient surveys. Overa, the study findings and concusions cannot be directy generaised to the UK setting but nonetheess the methodoogy was informative to our economic mode Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 185

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Appendix 9 Tabe of excuded heath economics studies with rationae Cost-effectiveness studies Bibiographic information Bosmans J, de Bruijne M, van Hout H, van Marwijk H, Beekman A, Bouter L, et a. Cost-effectiveness of a disease management program for major depression in edery primary care patients. J Gen Int Med 2006;21:1020 6 Wes KB, Schoenbaum M, Duan N, Miranda J, Tang L, Sherbourne C. Cost-effectiveness of quaity improvement programs for patients with subthreshod depression or depressive disorder. Psychiatr Serv 2007;58:1269 78 Peris RHP. When is pharmacogenetic testing for antidepressant response ready for the cinic? A costeffectiveness anaysis based on data from the STAR*D study. Neuropsychopharmacoogy 2009;34:2227 36 Armstrong EP, Skrepnek GH, Erder MH. Cost-utiity comparison of escitaopram and sertraine in the treatment of major depressive disorder. Curr Med Res Opin 2007; 23:251 8 Sicras-Mainar A, Navarro-Artieda R, Banca-Tamayo M, Gimeno-deaFuente V, Savatea-Pasant J. Comparison of escitaopram vs. citaopram and venafaxine in the treatment of major depression in Spain: cinica and economic consequences. Curr Med Res Opin 2010;26:2757 64 Einarson TR, Addis A, Iskedjian M. Pharmacoeconomic anaysis of venafaxine in the treatment of major depressive disorder. PharmacoEconomics 1997;12:286 96 van Baardewijk M, Vis PM, Einarson TR. Cost effectiveness of duoxetine compared with venafaxine-xr in the treatment of major depressive disorder. Curr Med Res Opin 2005;21:1271 9 Casciano J, Arikian S, Tarride J, Doye JJ, Casciano R. A pharmacoeconomic evauation of major depressive disorder. Epidemio Psichiatr Soc 1999;8:220 31 Trivedi MH, Wan GJ, Maick R, Chen JL, Casciano R, Geisser EC, et a. Cost and effectiveness of venafaxine extendedreease and seective serotonin reuptake inhibitors in the acute phase of outpatient treatment for major depressive disorder. J Cin Psychopharmaco 2004;24:497 506 Wade AG, Fernandez JL, Francois C, Hansen K, Danchenko N, Despiege N. Escitaopram and duoxetine in major depressive disorder: a pharmacoeconomic comparison using UK cost data. PharmacoEconomics 2008;26:969 81 Freeman H, Arikian S, Lenox-Smith A. Pharmacoeconomic anaysis of antidepressants for major depressive disorder in the United Kingdom. PharmacoEconomics 2000;18:143 8 Reason for excusion Patient popuation: First ine of therapy. Aso the intervention/comparators, i.e. GP trained to screen and treat vs usua care Patient popuation: Subthreshod depression and dysthymic disorder, interventions are quaity improvements programmes Patient popuation: First-ine MDD, the study assessed the cost-effectiveness of pharmacogenetic testing Patient popuation: First ine of therapy (athough does incude data on the probabiity of augmentation) Patient popuation: First ine of therapy Patient popuation: First ine of therapy Patient popuation: Not treatment resistant were excuded, aso interventions Patient popuation: First ine of therapy Patient popuation: Not treatment resistant. There is no augmentation Patient popuation: First ine of therapy Patient popuation: Not treatment resistant; aso first ine then titration and/or augmentation upon faiure of first ine Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 187

APPENDIX 9 Bibiographic information Revicki DA, Siddique J, Frank L, Chung JY, Green BL, Krupnick J, et a. Cost-effectiveness of evidence-based pharmacotherapy or cognitive behavior therapy compared with community referra for major depression in predominanty ow-income minority women. Arch Gen Psychiatry 2005;62:868 75 Lenox-Smith A, Conway P, Knight C. Cost effectiveness of representatives of three casses of antidepressants used in major depression in the UK. Pharmacoeconomics 2004; 22:311 19 Lenox-Smith A, Greenstreet L, Bursem K, Knight C. Cost effectiveness of venafaxine compared with generic fuoxetine or generic amitriptyine in major depressive disorder in the UK. Cin Drug Invest 2009;29:173 84 Armstrong EP, Maone DC, Erder MH. A Markov cost-utiity anaysis of escitaopram and duoxetine for the treatment of major depressive disorder. Curr Med Res Opin 2008;24:1115 21 Revicki DA, Brown RE, Keer MB, Gonzaes J, Cupepper L, Haes RE, et a. Cost-effectiveness of newer antidepressants compared with tricycic antidepressants in managed care settings. J Cin Psychiatry 1997;58:47 58 Machado M, Lopera MM, az-rojas J, Jaramio LE, Einarson TR, Universidad Naciona de Coombia Pharmacoeconomics Group. Pharmacoeconomics of antidepressants in moderateto-severe depressive disorder in Coombia. Pan American Journa of Pubic Heath 2008;24:233 9 Lave JR, Frank RG, Schuberg HC, Kamet MS. Costeffectiveness of treatments for major depression in primary care practice. Arch Gen Psychiatry 1998;55:645 51 Gibson TBJ. Cost burden of treatment resistance in patients with depression. Cost-utiity comparison of escitaopram and sertraine in the treatment of major depressive disorder. Am J Manag Care 2010;16:370 7 Reason for excusion Patient popuation: First ine of therapy Patient popuation: First ine of therapy Patient popuation: First ine of therapy Patient popuation: First-ine MDD Patient popuation: Not treatment resistant at baseine. Ony 12% are assumed to have TRD in the mode Patient popuation: First episode of moderate to severe depression aso first ine then titration and/or augmentation upon faiure of first ine Patient popuation: First ine of therapy Not a cost-effectiveness study (a modeing study showing that patients with TRD have higher medica costs than non-trd patients) 188 NIHR Journas Library www.journasibrary.nihr.ac.uk

DOI: 10.3310/hta17540 HEALTH TECHNOLOGY ASSESSMENT 2013 VOL. 17 NO. 54 Excuded quaity-of-ife studies Bibiographic information Cipriani A, Smith K, Burgess S, Carney S, Goodwin G, Geddes J, et a. Lithium vs. antidepressants in the ong-term treatment of unipoar affective disorder. Cochrane Database Syst Rev 2006;4:CD003492 Armstrong EPM. A Markov cost-utiity anaysis of escitaopram and duoxetine for the treatment of major depressive disorder. Curr Med Res Opin 2008;24:1115 21 Benedict A, Areano J, De Cock E, Baird JE-MA, Benedict AAB. Economic evauation of duoxetine vs. serotonin seective reuptake inhibitors and venafaxine XR in treating major depressive disorder in Scotand. J Affect Disord 2010;120:1 3 Revicki AD, Hanon J, Martin S, Gyuai L, Ghaemi NS, Lynch F, et a. Patient-based utiities for bipoar disorder-reated heath states. J Affective Dis 1998;87:203 10 Leeahanaj T, Leeahanaj T. The cost-effectiveness of aripiprazoe as adjunctive therapy in major depressive disorder: Thai economic mode. J Med Assoc Thai 2010;93(Supp. 6):43 50 Ng FD. Combination pharmacotherapy in unipoar depression. Expert Rev 2006;6:1049 60 Gismondi RL. Effect of adjunctive aripiprazoe on quaity of ife in patients with major depressive disorder: pooed data from three cinica trias. European Psychiatry 2010, conference (various pages) Keitner GIG. A randomized, pacebo-controed tria of risperidone augmentation for patients with difficut-to-treat unipoar, non-psychotic major depression. J Psychiatr Res 2009;43:205 14 Franco Martn M, Figueira ML. Quetiapine XR monotherapy and quetiapine XR+ongoing antidepressant vs. ithium+ongoing AD for Stage II treatment-resistant major depressive disorder. Eur Neuropsychopharmaco 2010;20:347 King M, Sibbad B, Ward E, Bower P, Loyd M, Gabbay M, et a. Randomised controed tria of nondirective counseing, cognitive-behaviour therapy and usua genera practitioner care in the management of depression as we as mixed anxiety and depression in primary care. Heath Techno Assess 2000;4(19) Peveer R, Kendrick T, Buxton M, Longworth L, Badwin D, Moore M, et a. A randomised controed tria to compare the cost-effectiveness of tricycic antidepressants, seective serotonin reuptake inhibitors and ofepramine. Heath Techno Assess 2005;9(16) Lenert LA. The reiabiity and interna consistency of an internet capabe program for measuring utiities. Qua Life Res 2000;9:811 17 Reason for excusion Review CEA study, which used QoL data from Revicki which was incuded Incuded in the CEA review Conducted in bipoar patients, a simiar one in unipoar was incuded Incuded in the CEA review A review on pharmacotherapy and has no QoL data Conference abstract Conference abstract Conference abstract CEA study, which reports treatment-specific utiities CEA study, which reports treatment-specific utiities A feasibiity study testing the reiabiity of a software CEA, cost-effectiveness anaysis. Queen s Printer and Controer of HMSO 2013. This work was produced by Edwards et a. under the terms of a commissioning contract issued by the Secretary of State for Heath. This issue may be freey reproduced for the purposes of private research and study and extracts (or indeed, the fu report) may be incuded in professiona journas provided that suitabe acknowedgement is made and the reproduction is not associated with any form of advertising. Appications for commercia reproduction shoud be addressed to: NIHR Journas Library, Nationa Institute for Heath Research, Evauation, Trias and Studies Coordinating Centre, Apha House, University of Southampton Science Park, Southampton SO16 7NS, UK. 189

QUERY TO THE AUTHOR

EME HS&DR HTA PGfAR PHR Part of the NIHR Journas Library www.journasibrary.nihr.ac.uk This report presents independent research funded by the Nationa Institute for Heath Research (NIHR). The views expressed are those of the author(s) and not necessariy those of the NHS, the NIHR or the Department of Heath Pubished by the NIHR Journas Library