Evidence Review Group s Report Ustekinumab for treating active and progressive psoriatic arthritis

Transcription

1 Evidence Review Group s Report Ustekinumab for treating active and progressive psoriatic arthritis Produced by Centre for Reviews and Dissemination (CRD), University of York Authors Dawn Craig, Research Fellow, CRD Joanne O Connor, Research Fellow, CRD Mark Rodgers, Research Fellow, CRD Rocio Rodriguez-Lopez, Information Specialist, CRD Alison Smith, Research Fellow, CRD Nerys Woolacott, Senior Research Fellow, CRD Correspondence to Xx xxxxx xxxxxxxx Xxxxx xxx xxxxxxxx xxx xxxxxxxxxxxx Xxxxxxxxx xx xxxx xxxxxxxxxx xxxx xxxx xxx Date completed Date completed (30/10/2013) Source of funding This report was commissioned by the NIHR HTA Programme as project number HTA 12/58. Declared competing interests of the authors None Acknowledgements We would like to thank our clinical advisor, Professor Ian Bruce (Professor of Rheumatology and NIHR Senior Investigator, Arthritis Research UK Epidemiology Unit, Centre for Musculoskeletal Research, Institute of Inflammation and Repair). 1

2 We would also like to thank Professor Stephen Palmer (Professor of Health Economics, CHE) for advice and for providing comments on the ERG report. Rider on responsibility for report The views expressed in this report are those of the authors and not necessarily those of the NIHR HTA Programme. Any errors are the responsibility of the authors. This report should be referenced as follows: Craig D, O Connor J, Rodgers M, Smith A, Woolacott N. Ustekinumab for treating active and progressive psoriatic arthritis: A Single Technology Appraisal. Centre for Reviews and Dissemination, Contributions of authors Mark Rodgers, Dawn Craig and NerysWoolacott wrote the clinical effectiveness sections of the report; Alison Smith, Joanne O Connor and Dawn Craig wrote the cost effectiveness sections of the report and performed the economic modelling. Rocio Rodriguez-Lopez wrote the sections of the report dealing with search strategies and provided information support. 2

3 Table of Contents List of abbreviations 12 Glossary 15 1 Summary Critique of the decision problem in the manufacturer s submission Summary of clinical effectiveness evidence submitted by the manufacturer Ustekinumab versus conventional management Ustekinumab versus TNFα inhibitors Safety of ustekinumab Summary of the ERG s critique of clinical effectiveness evidence submitted Ustekinumab versus conventional management Summary of cost effectiveness submitted evidence by the manufacturer Summary of the ERG s critique of cost effectiveness evidence submitted ERG commentary on the robustness of evidence submitted by the manufacturer Strengths Weaknesses and areas of uncertainty Summary of exploratory and sensitivity analyses undertaken by the ERG Anti-TNFα naïve population Anti-TNFα experienced population Sequencing analysis 28 2 Background Critique of manufacturer s description of underlying health problem Critique of manufacturer s overview of current service provision 30 3 Critique of manufacturer s definition of decision problem Population Intervention Comparators Outcomes Other relevant factors 33 4 Clinical Effectiveness Critique of the methods of review(s) Searches Clinical effectiveness search Inclusion criteria Critique of data extraction Quality assessment 37 3

4 4.1.5 Evidence synthesis Critique of trials of the technology of interest, their analysis and interpretation (and any standard meta-analyses of these) Description of PSUMMIT trial methods Outcomes collected in the PSUMMIT trials Internal validity of PSUMMIT trials External validity of PSUMMIT trials Analysis of PSUMMIT trial data Summary of PSUMMIT trials results ERG commentary on PSUMMIT trials Description and evaluation of adverse event studies Critique of the indirect comparison and/or multiple treatment comparison and the trials identified and included Conclusions of the clinical effectiveness section 67 5 Cost Effectiveness Overview ERG comment on manufacturer s review of cost-effectiveness evidence ERG s summary and critique of manufacturer s submitted economic evaluation Population Interventions and comparators Anti-TNFα naïve population Anti-TNFα experienced population Conventional management therapies Model structure Perspective, time horizon and discounting Treatment effectiveness and extrapolation PASI Score PASI Score changes HAQ Score PsARC response rates HAQ score change from baseline HAQ rebound and natural disease progression Long term withdrawal Mortality Adverse events Health related quality of life Resources and costs 92 4

5 Drug acquisition costs Administration costs Medical resource utilization Health state costs (HAQ- and PASI-related costs) Adverse event costs Cost effectiveness results Cost-effectiveness results for the anti-tnfα naïve patients Cost-effectiveness results for the anti-tnfα experienced patients Model validation and face validity check Exploratory and sensitivity analyses undertaken by the ERG Conclusions of the cost effectiveness section Impact on the ICER of additional clinical and economic analyses undertaken by the ERG Overview ERG corrections and adjustments to the manufacturer s base case model Additional ERG analyses Weight based NMA vs. ITT NMA (naïve population) Alternative HAQ score rebound assumptions (naïve population) HAQ score changes from ERG NMA (naïve population) Time horizon (naïve population) Week 24 vs. 12 as the PsARC response assessment time point (naïve population) Inclusion of the cost of phototherapy in the PASI-related cost estimate (naïve population) The ERG s preferred base case for anti-tnf naïve population Anti-TNFα experienced population Sequencing of treatment ERG sequencing analysis Results of ERG sequencing analysis Budget impact model Conclusions from ERG analyses End of life Overall conclusions Anti-TNFα naïve population Anti-TNFα experienced population Implications for research References Appendices PSUMMIT outcome data for dactylitis, DLQI and SF

6 10.2 NMA trial baseline comparison as provided by the manufacturer ERG NMA compared with MS NMA PsARC response HAQ conditional on PsARC ERG re-calculated baseline HAQ and PASI score ERG additional analyses Efficacy results derived from ITT trial data where possible (anti-tnfα experienced population) HAQ rebound to natural progression (anti-tnfα experienced population) Alternative time horizons Week 12 vs. 24 PsARC assessment time point Including the cost of phototherapy Natural history of patients with psoriatic arthritis eligible for biologic therapy Introduction Methods Results Discussion Conclusion Sequencing analysis Formulas used by Rodgers et al ERG estimates based on the formulas from Rodgers et al Full results of the Sequencing analysis 169 6

7 Table of Tables Table 1: Drug survival rates at 3 years (NOR-DMARDstudy) Table 2: Drug survival rates from observational studies of Anti-TNFαs for PsA. (Taken directly from DANBIO Registry (Glintborg et al. 2013)) Table 3: Proportion of patients achieving a response in terms of ACR 20/50/70, a Good EULAR response and DAS28 remission. (Constructed from data reported from DANBIO Registry (Glintborg et al. 2013)) Table 4: Baseline characteristics of PSUMMIT trials according to prior biologic exposure Table 5 Number of patients receiving 1, 2, 3 or more previous anti-tnfs (PSUMMIT 2) Table 6: ACR, PsARC and PASI75 response rates in PSUMMIT trials Table 7: Ustekinumab 45mg vs. 90mg. Random effects meta-analyses (relative risk, 95% confidence interval) Table 8: Discontinuation of study agent due to adverse events (PSUMMITtrials) Table 9: Quality assessment of included RCTs (from Rodgers et al) Table 10: (1 of 2) Summary of trial population baseline characteristics (placebo arm only) Table 11: (2 of 2) Summary of trial population baseline characteristics (treatment arm only) Table 12 Baseline DMARD use among RCTs included in the MTC Table 13: Summary of the trials used* to conduct the MTC (Manufacturer s submission, table 1, pg 73.) Table 14 Number of patients included in weight-based NMA Table 15: PASI 75 - Proportion of responders for the anti-tnfα naïve population, using the weightbased dosing criteria compared with ITT Table 16. PASI 90 - Proportion of responders for the anti-tnfα naïve population, using the weightbased dosing criteria compared with ITT Table 17. PsARC - Proportion of responders for the anti-tnfα naive population, using the weight based dosing criteria compared with ITT Table 18: Summary of the manufacturer s economic evaluation (and signposts to MS) Table 19: NICE reference checklist Table 20: Treatment dosing regimens (adapted from Table 5, Section 4.1.1, pg. 20 in the MS) Table 21. Manufacturer and ERG baseline PASI scores Table 22. Sources used to derive PASI response rates in the MS Table 23. PASI score change for PASI 75 responders and non-responders for the anti-tnfα naïve population (taken from the manufacturer s model, Inp_Efficacy!C247:D254 ) Table 24. PASI score change for PASI 75 responders and non-responders for the anti-tnfα experienced subgroup analysis (taken from manufacturer s model, Inp_Efficacy! F247:D248 ) Table 25. Manufacturer and ERG baseline HAQ estimates Table 26. Proportion of Ustekinumab and conventional management PsARC responders for both populations

8 Table 27. HAQ score change from baseline for PsARC responders and non-responders (naïve population) Table 28. HAQ score change from baseline for PsARC responders and non-responders (experienced population) Table 29. Unit costs used in the manufacturer's model (adapted from Table 55, pg. 139 of MS) Table 30: Base-case deterministic results, anti-tnfα naïve patients (based on Table 64, Section 7.7.6, pg. 165 in the MS) Table 31: Amended base-case deterministic results, TNF naïve patients (pg. 50 in the manufacturer s response to clarification) Table 32: PSA results; mean of 10,000 simulations, anti-tnfα naïve patients (pg 50 in the manufacturer s response to clarification) Table 33. Summary of scenario analyses and results (based MS. Tables 68-34, pg174-7) Table 35: Base-case results, anti-tnfα experienced patients (based on Table 65, Section 7.7.6, pg. 165 in the MS) Table 36: Base-case results, anti-tnfα experienced patients (pg 54 in the manufacturer s response to clarification) Table 37: PSA results; mean of 10,000 simulations, anti-tnfα experienced patients (Table 67, Section 7.7.8, pg. 172 in the MS) Table 38: PSA results; mean of 10,000 simulations, anti-tnfα experienced patients (pg 54 of manufacturer s response to clarification) Table 39. Summary of scenario analyses and results (MS. Table 40, pg178) Table 41: Results comparison: TA199; TA220 and Ustekinumab submission Table 42: Comparisons with previous submissions Table 43: Corrected model probabilistic results (MS corrections and ERG baseline HAQ and PASI estimates (anti-tnfα naive population) Table 44: Corrected model probabilistic results (MS corrections and ERG baseline HAQ and PASI estimates (anti-tnfα experienced population) Table 45. Probabilistic results for model using full ITT NMA results (anti-tnfα naive population)114 Table 46. Probabilistic results for model with ITT NMA results used, excluding ustekinumab 90mg (anti-tnfα naive population) Table 47. Probabilistic results for model assuming HAQ rebounds to natural progression ("worst case scenario"; anti-tnfα naive population) Table 48. Probabilistic results for model using HAQ score change results from updated Yang NMA using 12-week PSUMMIT trial data (anti-tnfα naive population) Table 49. Probabilistic results for model using HAQ score change results from updated Yang NMA using 24-week PSUMMIT trial data (anti-tnfα naive population) Table 50. Alternative time horizons, ICER results (anti-tnfα naive population) Table 51. Alternative time horizons, ICER vs. conventional management strategy results (anti-tnf naive population) Table 52. Probabilistic results for model using week 24 weeks as the PsARC assessment time point for all treatments (anti-tnfα naïve population)

9 Table 53. Probabilistic results for model with cost of phototherapy included in the PASI-related cost estimate (anti-tnfα naive population) Table 54. ERG's preferred base case (anti-tnf naive population) Table 55: Probabilistic results for scenario analyses on the anti-tnfα experienced population Table 56: Alternative time horizons (anti-tnfα experienced population) Table 57: Cost-effective treatments for various WTP thresholds for scenario Table 58: Cost-effective treatments for various WTP thresholds for scenario Table 59: Cost-effective treatments for various WTP thresholds for scenario Table 60: Estimated budget impact (Table 86, pg 186 of MS) Table 61: Re-estimated budget impact, assuming ustekinumab has the projected market share of golimumab Table 62: Comparison of baseline characteristics NMA RCTs Table 63: MTC comparison manufacturer s data compared with ERG reanalysis for PsARC response Table 64: ERG HAQ PsARC response compared with MS estimate Table 65 ERG baseline PASI estimate for anti-tnf naive population Table 66. Standard error and standard deviations for ERG baseline PASI estimates (anti-tnfα naive population) Table 67. ERG baseline PASI estimate for the anti-tnfα experienced population Table 68. Standard deviation and standard error for ERG baseline PASI estimate (anti-tnfα experienced population) Table 694: ERG baseline HAQ estimate for the anti-tnfα naive population Table 70. Standard deviation and standard error for ERG baseline HAQ estimates (anti-tnfα naive population) Table 71. ERG baseline HAQ score for anti-tnf experienced population Table 72. Standard deviation and standard error for ERG baseline HAQ estimate (anti-tnfα experienced population) Table 73. Probabilistic results for model using all available ITT trial results (anti-tnfα experienced population) Table 74. Probabilistic results for model using all ITT trial results for ustekinumab 45mg only (anti- TNFα experienced population) Table 75. Probabilistic results for model assuming HAQ rebounds to natural progression ("worst case scenario"; anti-tnfα experienced population) Table 76. Probabilistic results for model using 5 year time horizon (anti-tnfα naive population) Table 77. Probabilistic results for model using 5 year time horizon (anti-tnfα experienced population) Table 78. Probabilistic results for model using 10 year time horizon (anti-tnfα naive population) 159 Table 79. Probabilistic results for model using 10 year time horizon (anti-tnfα experienced population) Table 80. Probabilistic results for model using 20 year time horizon (anti-tnfα naïve population) 159 9

10 Table 81. Probabilistic results for model using 20 year time horizon (anti-tnfα experienced population) Table 82. Probabilistic results for model using 40 year time horizon (anti-tnfα naive population) 160 Table 83. Probabilistic results for model using a 40 year time horizon (anti-tnfα experienced population) Table 84. Probabilistic results for model using week 12 weeks as the PsARC assessment time point for all treatments (anti-tnfα naïve population) Table 85. Probabilistic results for model using week 12 weeks as the PsARC assessment time point for all treatments (anti-tnfα experienced population) Table 86. Probabilistic results for model using week 24 weeks as the PsARC assessment time point for all treatments (anti-tnfα experienced population) Table 87. Probabilistic results for model including the cost of phototherapy in the PASI-related cost (anti-tnfα experienced population) Table 88: Initial PsARC response given patient discontinued first course because of a lack of efficacy (manufacturer s data derived from model: Cells Inp_Efficacy!C22:C39 ) Table 89: Withdrawal rates after three months, given patient discontinued first course of treatment because of lack of efficacy( manufacturer s estimates derived from model: Cells Inp_Outomes!C16:D20 ) Table 90: Withdrawal rates after three months, given patient discontinued first course of treatment because of an adverse event (manufacturer s estimates derived from model: Cells Inp_Outomes!C16:D20 ) Table 91: Costs and QALYs of biologics used as second-line therapy if first biologic fails due to inefficacy (the manufacturer s data was used from PSUMMIT2 for ustekinumab and the ERG estimates were used for the other comparators) Table 92: Costs and QALYs of biologics used as second-line therapy if first biologic fails due to adverse events Table 93: Costs and QALYs of biologics used as second-line therapy if first biologic fails due to inefficacy (ERG estimates were used for all comparators)

11 Table of Figures Figure 1: ACR20 response through week 52 (PSUMMIT1) Figure 2: ACR20 response through week 52 (PSUMMIT2) Figure 3: PsARC response through week 52 (PSUMMIT1) Figure 4: PsARC response through week 52 (PSUMMIT2) Figure 5: PASI 75 response through week 52; randomised subjects with 3% BSA psoriasis skin involvement at baseline (PSUMMIT2) Figure 6: PASI 75 response through week 52; randomised subjects with 3% BSA psoriasis skin involvement at baseline (PSUMMIT1) Figure 7: Decision tree structure (Figure 20, Section 7.2.2, pg. 97 in the MS) Figure 8: Markov model structure (Figure 31, Section 7.2.2, pg. 98 in the MS)

12 List of abbreviations ACR American College of Rheumatology response criteria ADEPT AE BAD BNF BSA BSR CEAC CI CrI DLQI DMARD EMA ERG EQ-5D HAQ HRQoL ICER IMPACT ITT Adalimumab Effectiveness in Psoriatic Arthritis Trial Adverse event British Association of Dermatologists British National Formulary Body surface area British Society of Rheumatologists Cost-effectiveness acceptability curve Confidence interval Credible Interval Dermatology Life Quality Index Disease modifying anti-rheumatic drug European Medicines Agency Evidence Review Group EuroQol-5D Health Assessment Questionnaire Health-related quality of life Incremental cost-effectiveness ratio Infliximab Multinational Psoriatic Arthritis Controlled Trial Intention-to-treat 12

13 MS MTC MTX N/A NHS NICE NMA NOAR NSAID OLS OR PASI PsA PSA PsARC QALYs RCT RR SD SF-36 SPC STA Manufacturer s submission Mixed treatment comparison Methotrexate Not applicable National Health Service National Institute for Health and Clinical Excellence Network Meta-Analysis Norfolk Arthritis Register Nonsteroidal anti-inflammatory drug Ordinary Least Square Odds Ratio Psoriasis Area and Severity Index Psoriatic arthritis Probabilistic sensitivity analysis Psoriatic Arthritis Response Criteria Quality adjusted life years Randomised Controlled Trial Relative risk Standard deviation Short Form Questionnaire 36 Items Summary of product characteristics Single Technology Appraisal 13

14 TNF Tumour necrosis factor 14

15 Glossary Adverse effect An abnormal or harmful effect caused by and attributable to exposure to a chemical (e.g. a drug), which is indicated by some result such as death, a physical symptom or visible illness. An effect may be classed as adverse if it causes functional or anatomical damage, causes irreversible change in the homeostasis of the organism, or increases the susceptibility of the organism to other chemical or biological stress. Ankylosing spondylitis A rheumatic disease that affects the spine and may lead to some degree of stiffness in the back. As the inflammation goes and healing takes place, bone grows out from both sides of the vertebrae and may join the two together; this stiffening is called ankylosis. Arthritis A term meaning inflammation of the joint(s), but which is often used to include all joint disorders. Sometimes joints are damaged through the disease process of arthritis. Articular Of or relating to the joints. Biologic therapies Medical preparations derived from living organisms. Includes anti-tnfα drug and other new drugs which target the pathologically active T cells involved in psoriasis, and psoriatic arthritis. Confidence interval (CI) The typical ( Classical or Frequentist ) definition is the range within which the "true" value (e.g. size of effect of an intervention) would be expected to lie if sampling could be repeated a large number of times (e.g. 95% or 99%). Corticosteroid 15

16 A synthetic hormone similar to that produced naturally by the adrenal glands that is available in pill, topical, and injectable forms. Cost-benefit analysis An economic analysis that converts the effects or consequences of interventions into the same monetary terms as the costs and compares them using a measure of net benefit or a cost-benefit ratio Cost-effectiveness analysis An economic analysis that expresses the effects or consequences of interventions on a single dimension. This would normally be expressed in natural units (e.g. cases cured, life-years gained, additional strokes prevented). The difference between interventions in terms of costs and effects is typically expressed as an incremental cost-effectiveness ratio (e.g. the incremental cost per life-year gained). Cost-utility analysis The same as a cost-effectiveness analysis but the effects or consequences of interventions are expressed in generic units of health gain, usually quality-adjusted life years (QALYs). Crohn's disease An inflammatory condition of the digestive tract; rheumatic diseases are often associated with it and ulcerative colitis is related to it. C-reactive protein (CRP) Concentrations of this protein in the blood can be measured as a test of inflammation or disease activity. Ciclosporin A medication originally developed to prevent the immune system from rejecting transplanted organs, which has also proved helpful in treating psoriasis. Disease-modifying anti-rheumatic drugs (DMARDs) DMARDs are drugs capable of modifying the progression of rheumatic disease. The term is, however, applied to what are now considered to be traditional disease modifying drugs, in particular sulphasalazine, methotrexate and ciclosporin, as well as azathioprine, cyclophosphamide, 16

17 antimalarials, penicillamine and gold. The newer agent leflunomide may be included as a DMARD. The biologics such as etanercept and infliximab are not generally referred to as DMARDS. Effect size A generic term for the estimate of effect for a study. Emollient An agent that holds moisture in the skin, and by doing so softens or soothes it. Erythrocyte sedimentation rate (ESR) One of the tests designed to measure the degree of inflammation. Fixed effect model A statistical model that stipulates that the units under analysis (e.g. people in a trial or study in a metaanalysis) are the ones of interest, and thus constitute the entire population of units. Only within-study variation is taken to influence the uncertainty of results (as reflected in the confidence interval) of a meta-analysis using a fixed effect model. Heterogeneity In systematic reviews heterogeneity refers to variability or differences between studies in the estimates of effects. A distinction is sometimes made between "statistical heterogeneity" (differences in the reported effects), "methodological heterogeneity" (differences in study design) and "clinical heterogeneity" (differences between studies in key characteristics of the participants, interventions or outcome measures). Immunomodulator A substance that alters the body s immune response. Intention-to-treat An intention-to-treat analysis is one in which all the participants in a trial are analysed according to the intervention to which they were allocated, whether they received it or not. Methotrexate 17

18 One of the oldest chemotherapy drugs used to treat cancer; used in the treatment of psoriasis. Mixed treatment comparison Mixed treatment comparison (also known as network meta analysis) is a form of meta-analysis used to strengthen inference concerning the relative efficacy of two treatments. It uses data based on direct comparisons (A vs. B and B vs. C trials) and indirect comparisons (A vs C trials) also, it facilitates simultaneous inference regarding all treatments in order to select the best treatments. Monoclonal antibody An antibody produced in a laboratory from a single clone that recognizes only one antigen. Non-steroidal anti-inflammatory drugs (NSAIDs) Consists of a large range of drugs of the aspirin family, prescribed for different kinds of arthritis which reduce inflammation and control pain, swelling and stiffness. Network meta analysis See Mixed treatment comparison PASI score Psoriasis Area Severity Index score, a number representing the size, redness, thickness, and scaliness of a person s psoriasis. Placebo An inactive substance or procedure administered to a patient, usually to compare its effects with those of a real drug or other intervention, but sometimes for the psychological benefit to the patient through a belief that s/he is receiving treatment. Plaque psoriasis The most common form of psoriasis, also known as psoriasis vulgaris, recognized by red, raised lesions covered by silvery scales. About 80% of psoriasis patients have this type. Psoriasis 18

19 A chronic skin disease characterized by inflammation and scaling. Scaling occurs when cells in the outer layer of skin reproduce faster than normal and pile up on the skin s surface. It is understood to be a disorder of the immune system. Psoriatic arthritis This disease is characterized by stiffness, pain, and swelling in the joints especially of the hands and feet. It affects about 23% of people with psoriasis. Early diagnosis and treatment can help inhibit the progression of joint deterioration. Quality Adjusted Life Year (QALY) An index of health gain where survival duration is weighted or adjusted by the patient s quality of life during the survival period. QALYs have the advantage of incorporating changes in both quantity (mortality) and quality (morbidity) of life. Quality of Life A concept incorporating all the factors that might impact on an individual s life, including factors such as the absence of disease or infirmity as well as other factors which might affect their physical, mental and social well-being. Random effects model A statistical model sometimes used in meta-analysis in which both within-study sampling error (variance) and between-studies variation are included in the assessment of the uncertainty (confidence interval) of the results of a meta-analysis. Randomised controlled trial (RCT) (Synonym: randomised clinical trial) An experiment in which investigators randomly allocate eligible people into intervention groups to receive or not to receive one or more interventions that are being compared. Relative Risk (RR) (synonym: risk ratio) The ratio of risk in the intervention group to the risk in the control group. The risk (proportion, probability or rate) is the ratio of people with an event in a group to the total in the group. A relative risk of one indicates no difference between comparison groups. For undesirable outcomes an RR that is less than one indicates that the intervention was effective in reducing the risk of that outcome. 19

20 Remission A lessening or abatement of the symptoms of a disease. Rheumatoid arthritis A chronic autoimmune disease characterized by pain, stiffness, inflammation, swelling, and, sometimes, destruction of joints. Sensitivity analysis An analysis used to determine how sensitive the results of a study or systematic review are to changes in how it was done. Sensitivity analyses are used to assess how robust the results are to uncertain decisions or assumptions about the data and the methods that were used. Statistical significance An estimate of the probability of an association (effect) as large or larger than what is observed in a study occurring by chance, usually expressed as a P-value. T cell A type of white blood cell that is part of the immune system that normally helps protect the body against infection and disease. Topical agent A treatment such as a cream, salve, or ointment that is applied to the surface of the skin. Tumor necrosis factor (TNF) One of the cytokines, or messengers, known to be fundamental to the disease process that underlies psoriasis. It often plays a key role in the onset and the continuation of skin inflammation. Variance A measure of the variation shown by a set of observations, defined by the sum of the squares of deviations from the mean, divided by the number of degrees of freedom in the set of observations. 20

21 1 Summary 1.1 Critique of the decision problem in the manufacturer s submission The decision problem addressed in the manufacturer s submission was largely appropriate and matched the scope. It should be noted that the manufacturer focused on the effectiveness of ustekinumab in two subgroups of patients with PsA: (1) anti-tnfα naïve patients and (2) anti-tnfα experienced patients (i.e. those who had previously received one or more TNFα inhibitors). The submission considered a different comparison for each of these subgroups: (1) ustekinumab versus licensed TNFα inhibitors for anti-tnfα naïve patients; and (2) ustekinumab versus conventional management (excluding the use of anti-tnfα inhibitors) for anti-tnfα experienced patients. Neither of these scenarios fully reflect UK clinical practice where a number of anti-tnfαs can be tried in (no fixed) sequence. The ERG noted that ustekinumab versus a second biologic might be a more appropriate comparison for the substantial proportion of experienced patients who had received only a single prior TNFα inhibitor. Comparison with only conventional management implies that a longer sequence of anti-tnf treatment has failed. 1.2 Summary of clinical effectiveness evidence submitted by the manufacturer Ustekinumab versus conventional management The majority of evidence presented on the efficacy of ustekinumab for PsA in the MS was derived from two methodologically similar RCTs (PSUMMIT 1 and PSUMMIT 2) comparing 45mg and 90mg doses of ustekinumab against placebo (plus conventional non-biologic management). Data from these trials suggest that ustekinumab is a more effective treatment than conventional management over weeks in terms of both joint (ACR 20/50/70, PsARC) and skin (PASI 75/90) response, and that these benefits are likely to persist for at least 52 weeks. No data is available on longer term radiographic progression. On the basis of the available subgroup data from the PSUMMIT 2 trial, there is no convincing evidence of a substantial difference in the efficacy of ustekinumab between anti-tnfα experienced patients and patients who had not received any prior anti-tnfα treatment. Treatment effects were not significantly different for 45mg and 90mg doses of ustekinumab Ustekinumab versus TNFα inhibitors In the absence of head-to-head comparisons between alternative anti-tnfα agents, the manufacturer conducted a network meta-analysis (NMA) incorporating nine placebo controlled RCTs to estimate the relative efficacy between the five biologic agents of interest: adalimumab (2 RCTs), etanercept (2 21

22 RCTs), infliximab (2 RCTs), golimumab (1 RCT) and ustekinumab (2 RCTs). These trials appear to represent the complete relevant body of evidence for the included comparators. Trials of all five agents had a common comparator, placebo, which allowed a network to be established. An NMA was undertaken for the anti-tnfα naïve population only, which reflected the population included in all of the trials, with the exception of the treatment experienced subgroup in PSUMMIT2. Separate networks were constructed for three outcomes: PASI75, PASI90 and PsARC. These three outcomes were considered for two time points weeks and 24 weeks. Overall the results of the NMA found that *********** **** had the lowest or one of the lowest response rates for skin (PASI 75 and 90) and joints (PsARC). For PASI 75 response, ********** was the most effective treatment, and for PASI 90 response ********** was the most effective treatment. For all analyses, ********* achieved the greatest proportion of PsARC responders, followed by ******************** and **********. The manufacturer made no attempt to compare the effectiveness of ustekinumab with TNFα inhibitors among anti-tnfα experienced patients Safety of ustekinumab Four data sources were used to investigate adverse events associated with ustekinumab: (1) the PSUMMIT trials; (2) five-year extensions of four RCTs of ustekinumab for psoriasis; (3) the observational Psoriasis Longitudinal Assessment and Registry (PSOLAR) study; and (4) an observational study from the British Society for Rheumatology Biologics Register (BSRBR) for anti- TNFα therapies in PsA. With the exception of a slightly higher proportion of injection site reactions for 90mg observed in the PSUMMIT trials (1% 90mg vs. 0.6% 45mg vs. 0.4% placebo), the included trial data did not suggest any obvious excess in adverse events, serious adverse events or treatment discontinuation for ustekinumab treated patients. 1.3 Summary of the ERG s critique of clinical effectiveness evidence submitted Ustekinumab versus conventional management The PSUMMIT trials comparing ustekinumab against conventional management were of adequate methodological quality, though interruption of the placebo-controlled phase at week 16 (where participants were allowed to cross-over to an alternative treatment arm) and the termination of the 22

23 controlled phase altogether at 24 weeks provided only a very brief comparison for a chronic condition such as PsA. However, the results obtained over this short period are likely to be valid. The three-arm design of the PSUMMIT trials meant that a substantial proportion of participants were randomised to an unlicensed ustekinumab dose (i.e. 90mg in patients 100kg). However, efficacy estimates for the 45mg and 90mg ustekinumab arms were similar, and there did not appear to be an excess of adverse events associated with the higher dose. Analyses of the anti-tnfα experienced population were based on a subgroup of just 180 or fewer patients from the PSUMMIT 2 trial. The number of prior TNFα inhibitors and duration of prior treatment varied among patients within this subgroup, so data for patients who could be considered truly anti-tnfα refractory is scarce. The additional adverse event studies appear to be appropriate sources for establishing longer term safety information on ustekinumab (albeit in a psoriasis population) and anti-tnfα treatment in PsA. Other than the PSUMMIT trials, the manufacturer did not assess the quality of these adverse event studies, and the possibility of observational data being confounded by differing selection criteria across treatments cannot be excluded. RCTs included in the NMA were of a consistent and adequate quality. There was some heterogeneity; in particular prior DMARD exposure was variable across trials, though where reported the mean number of prior DMARDs was typically fewer than two. On this basis it would appear that, while baseline disease severity is broadly comparable across trials, the NMA population may be less severe than that routinely considered for biologic treatment in practice. The manufacturer assumed that in practice, ustekinumab patients will be treated according to weightbased dosing criteria ( i.e. 45mg for patients with body weight 100kg; and 90mg for those with body weight >100kg), and so excluded PSUMMIT trial patients not meeting these criteria from their NMA. However, it is not clear to the ERG that this is entirely in line with the licence (which permits but does not mandate 90mg in patients >100 kg) or the manner in which the trials were conducted. Furthermore, this resulted in a significant amount of data being lost from the analyses. Nevertheless, despite some limitations in the NMA presented by the manufacturer, the degree of clinical heterogeneity between the included trials in terms of joint and skin disease severity and functional status was reasonable and the assumption of exchangeability between the trials for the purposes of the analyses was considered acceptable. Overall, the ERG believes that it was appropriate to undertake the network meta-analysis and the results obtained by the manufacturer were robust when compared with the results of the ERG analysis. 23

24 1.4 Summary of cost effectiveness submitted evidence by the manufacturer The manufacturer presented a decision model, which was based on previously published models. It comprised two elements: a short-term decision tree, which evaluated patients initial response to treatment; followed by a longer-term Markov model which captured the longer term costs and consequences of treatments. They evaluated treatments over a lifetime horizon, from a NHS/PSSRU perspective. Treatment response, in the base case, was determined by PsARC response. For all comparators response was determined at 12 weeks, but for ustekinumab it was determined at 24 weeks, based on dosing requirements; this was tested in a sensitivity analysis. Patients who did not achieve a response at the appropriate time point were considered non-responders and moved to conventional management. After the initial response point a constant withdrawal rate was applied to account for withdrawals due to lack of efficacy and adverse events in the long run. Health states within the Markov model captured both the joint (PsARC) and skin (PASI) components of PsA, as both of these are related to quality-of-life and costs. PASI score changes were dependent on achieving a PASI 75 response; HAQ (functional status measure) scores were dependent on achieving a PsARC response. Both HAQ and PASI scores were incorporated into health-related quality-of-life through the use of published mapping equations. Resource use data were taken from a previous submission. Drug prices were taken from the BNF (2013). Health state costs were estimated as a function of HAQ and PASI through the use of previously published work. Deterministic and probabilistic results were presented. In addition, numerous scenario analyses were undertaken. The manufacturers presented analyses for two populations: anti-tnfα naïve patients and anti-tnfα experienced patients (i.e. those who had previously received one or more TNFα inhibitors). For anti- TNFα naïve patients ustekinumab was compared with conventional management, adalimumab, etanercept, golimumab and infliximab. This analysis utilised effectiveness evidence from the manufacturers NMA, a previous ERG NMA, and PSUMMIT data (naïve population only). In addition, they also utilised data on resource use, HAQ rebound, HAQ natural history and the relationship between HAQ, PASI and EQ-5D from the previous Rodgers et al. submission. For anti-tnfα naïve patients when a full probabilistic incremental analysis was undertaken both adalimumab and ustekinumab were found to be dominated. That is, both of these treatments were more costly and less effective when compared with the next most costly alternative. Golimumab achieved an ICER of 12,235; the ICERs for etanercept and infliximab both exceeded 100,

25 Probabilistic results did not differ significantly from the deterministic results. The probabilistic ICERs for each of the pairwise comparisons with conventional management strategies ranged from 13,134 to 40,068 with the ICER for ustekinumab compared with conventional management strategies being 23,048. Golimumab, etanercept and adalimumab all had lower pairwise ICERs. For anti-tnfα experienced patients, ustekinumab was compared with conventional management only. This analysis utilised evidence from a sub-group of the PSUMMIT 2 trial. In addition, they also utilised data on resource use, HAQ rebound, HAQ natural history and the relationship between HAQ, PASI and EQ-5D from the previous Rodgers et al. Submission to NICE for an appraisal of anti-tnfs for psoriatic arthritis. For anti-tnfα experienced patients, ustekinumab achieved a probabilistic ICER of 29,451 when compared with conventional management. Sensitivity analysis undertaken for both the naïve and experienced evaluations did not significantly change the results obtained. 1.5 Summary of the ERG s critique of cost effectiveness evidence submitted The model met the NICE reference standard and is in line with the NICE scope. The structure of the model was based on previous submissions and the ERG felt that it was appropriate. Both the manufacturer and the ERG undertook a number of sensitivity analyses in an attempt to explore some of the uncertainty in data inputs. In general these analyses did not have a significant impact on the results obtained but the ERG feels that some of these analyses represented a more realistic base case and have presented this as a preferred analysis for the modelling of the anti-tnf naive population. Due to a lack of data to evaluate the anti-tnfα experienced population and the inappropriate single comparator no preferred analysis is presented, rather the ERG have elected to explore the issue of sequencing (see Section 1.7). 1.6 ERG commentary on the robustness of evidence submitted by the manufacturer Strengths The short-term estimates of efficacy for all agents against placebo were derived from RCTs of adequate methodological quality and are likely to be reliable. The manufacturer s NMA was robust, with little variation in estimates from the ITT, weight-based, or combined-dose analyses. The model structure was appropriate and the evaluation met the NICE reference case Weaknesses and areas of uncertainty 25

26 Ustekinumab was not compared with other licensed biologics for the anti-tnfα experienced population. The manufacturer suggested that that ustekinumab provides a novel treatment option when anti-tnfα treatment has failed, and so the relevant comparator in anti-tnfα experienced patients would be conventional treatment without any biologic component. However the model used evidence derived from a population who had experienced anti-tnfs but not failed them as a class. Whilst observational evidence suggests diminishing response rates and treatment persistence associated with the sequential use of anti-tnfα agents, no robust evidence for the relative efficacy of ustekinumab and anti-tnfs when used as second or later-line therapy is available. *** *** ******* ******* **** **** ********* ****** *** **** *********** **** *********** ** ******** **** ***** *********** *** ** ******** ** ********* ** ******* ******* *** ********** ********* ** ******** ** * ** * * ** * ** * ** **** ********* ****** ***** **** ***** *** ***** ** ******** ******** *** *********** ** ********* *********** ********. The ERG tried to address this issue in an additional cost-effectiveness analysis incorporating sequential use of anti- TNFs, but this was based on a number of strong assumptions and suboptimal data. Given the lack of comparative data for second-line biologic treatment, a head-to-head trial of ustekinumab versus TNFα inhibitor therapy in this population may be warranted. Any such trial should clearly distinguish between prior anti-tnfα experience and anti-tnfα class failure in its selection of participants. The ERG believes that some uncertainty still remains due to a lack of robust evidence on some key model parameters. These include: The natural history of HAQ score change. This parameter was taken from a previous publication (Bojke et al; Rodgers et al.) and is based on a secondary analysis of data from the NOAR database. The analysis was undertaken in 2009 and a number of limitations were highlighted by the ERG at that time. In 2009 it represented the best available evidence; it is not clear to the ERG that it remains the best available evidence. Rebound to HAQ on withdrawal from treatment. An alternative scenario has been presented in this submission, but no further evidence has been presented to support which is most likely to reflect clinical reality. HAQ and PASI related costs. These data were also taken from a previous submission. Whilst this allows for consistent comparison the limitations of these data were fully discussed in the previous submission and no attempt has been made to address the limitations. A relatively large though statistically non-significant PASI75 benefit was observed in the anti-tnfα experienced subgroup in PSUMMIT 2. This may be an artefact of the small number of patients in the analysis and could be confirmed or refuted in a properly powered trial in this patient population. 26

27 1.7 Summary of exploratory and sensitivity analyses undertaken by the ERG The ERG found that the baseline HAQ and baseline PASI were incorrectly calculated. Therefore, a corrected model was presented. This corrected model estimated that for the anti-tnfα naïve population, the probabilistic ICER for ustekinumab vs. conventional management was 23,508 and that ustekinumab was dominated in the incremental analysis. For the anti-tnfα experienced population the probabilistic ICER for ustekinumab vs. conventional management was estimated as 29,451. All further analyses undertaken by the ERG used this corrected version of the model. Several sensitivity analyses were then undertaken for both the anti-tnfα naïve and the anti-tnfα experienced populations as well as an exploratory sequencing analysis Anti-TNFα naïve population The analyses undertaken for the naïve population are listed below, with probabilistic ICER results for ustekinumab for vs. conventional management and for the incremental analysis presented for each: Full ITT NMA results for all treatments used: Ustekinumab vs. conventional management ICER: 23,480 Incremental analysis ICER: Dominated Full ITT NMA results, except for ustekinumab 90mg used: Ustekinumab vs. conventional management ICER: 23,542 Incremental analysis ICER: Dominated Assumption that HAQ rebounds to natural progression used: Ustekinumab vs. conventional management ICER: 41,500 Incremental analysis ICER: Dominated HAQ score change estimates from updated Yang NMA using 12-week PSUMMIT trial data used: Ustekinumab vs. conventional management ICER: 23,140 Incremental analysis ICER: Dominated HAQ score change estimates from updated Yang NMA using 24-week PSUMMIT trial data used Ustekinumab vs. conventional management ICER: 22,987 Incremental analysis ICER: Dominated 27

28 24 week PsARC assessment point for all treatments was used: Ustekinumab vs. conventional management ICER: 30,495 Incremental analysis ICER: Dominated Phototherapy cost was included: Ustekinumab vs. conventional management ICER: 22,455 Incremental analysis ICER: Dominated In addition, 5, 10, 20 and 40 year time horizons were estimated. Ustekinumab was dominated for each time horizon, when compared with the next best non-dominated alternative. When compared with conventional management, the ICER for ustekinumab varied from 55,029 (5 year time horizon) to 23,509 (52 year time horizon). In Section 6, a preferred base case for the anti-tnfα naïve population was also presented. When compared with conventional management, the ICER for ustekinumab was 23,246. When compared with the next best non-dominated alternative, ustekinumab was dominated Anti-TNFα experienced population The analyses undertaken for the experienced population are listed below, with probabilistic ICER results presented for each: Available ITT trial results for all treatments used: ICER estimated as 29,797 ITT trial results used for ustekinumab 45mg arm only: ICER estimated as 29, week PsARC assessment point for all treatments: ICER estimated as 39, week PsARC assessment point for all treatments: ICER estimated as 32,162 Phototherapy cost was included: ICER estimated as 28,670 Assumption that HAQ rebounds to natural progression: ICER estimated as 52,408 5 year time horizon: ICER estimated as 69, year time horizon: ICER estimated as 46, year time horizon: ICER estimated as 32, year time horizon: ICER estimated as 29, Sequencing analysis In addition to the sensitivity analysis undertaken, an exploratory analysis into the sequencing of biologic treatments was presented. There are a number of caveats to this analysis (See Section 6.3.x) The ICERs for second-line treatment, when the first-line treatment had failed, either due to lack of 28

29 efficacy or due to adverse events were calculated. Within this analysis, ustekinumab was never estimated to be cost-effective at a willingness to pay threshold of 30,000, although these results were subject to significant uncertainty. 29

30 2 Background 2.1 Critique of manufacturer s description of underlying health problem. The manufacturer provided a brief but accurate overview of psoriatic arthritis (PsA). They state that both indications for ustekinumab (psoriasis and PsA) cover approximately 27,000 patients in England and Wales, though it should be noted that only 7,000 of these are patients with PsA. 2.2 Critique of manufacturer s overview of current service provision The manufacturer s submission (MS) provided an accurate overview of the treatment pathway for active and progressive PsA patients, based on the current clinical pathway recommended by British Society of Rheumatology (BSR) and existing NICE guidance on the use of TNFα inhibitors in PsA. The manufacturer provided details of the anti-tnfα agents (etanercept, infliximab, adalimumab and golimumab) that are currently used for the treatment of active and progressive PsA in UK clinical practice. The MS briefly discussed sequential use of TNFα inhibitors, referencing observational evidence that showed lower response rates and treatment persistence after switching from one TNFα inhibitor to another. 1, 2 As this is an important issue in the prescribing of anti-tnfαs in PsA the ERG examined these studies, and they are briefly summarised below. Fagerli et al. 2013: 1 : This small (n=439) uncontrolled observational study, appeared to include all patients with inflammatory arthropathies from five centres in Norway. It was not reported whether these were specialist centres or what proportion of PsA patients in Norway were treated in these centres. It is therefore not clear whether the included patients are representative of all PsA patients in Norway that received anti-tnfαs, and by extension how typical they are of UK patients. Of note is that the diagnostic criteria for PsA were more lax than is typical such that not all patients had swollen joints. The study sample comprised all patients who started an anti-tnfα in 2001 or later. As the anti- TNFαs available has grown over the subsequent 12 years, the actual anti-tnfαs experienced by these patients may not reflect the likely experience of a typical current cohort starting their first or second anti-tnfα in All the patients appeared to be included in the analysis, however, it is not completely clear that the non-switchers continued to take their 1 st anti-tnfα, rather than merely not starting a second. The most common 1 st anti-tnfα was etanercept (50%), with adalimumab the next most common (29%). It was not reported if these proportions varied over the duration of the study. Table 1 shows drug survival rates at three years for switchers and patients who remained on their first treatment. 30

31 Table 1: Drug survival rates at 3 years (NOR-DMARDstudy) Anti-TNFαs % drug survival All patients (n=439) Those who stayed on 1 st treatment (n=344) Those who switched to 2nd treatment (n=95) 1 st 57% 71% 17% 2 nd % Glintborg et al., 2013: 2 The population in this uncontrolled observational study was apparently unselected, comprising a Danish registry started in 2000 and containing more than 90% of all adults with rheumatic disease in Denmark who were treated with a biologic (TNFα inhibitor). It was approved by the Danish Health and Medicines Authority. The registry included patients treated between 2000 and 2012 receiving a biologic marketed for PsA, and on entry to the registry patients had to be starting their first anti-tnfα. Patients using anti-tnfαs in clinical trials were not included. There was no checking of the data but the participating centres were subject to audit. The analysis of response rates did not include those who switched from first anti-tnfα after 2 years. Kaplan-Meyer curves were used to assess drug survival (i.e. staying on given anti-tnfα). The most common 1 st anti-tnfα was adalimumab (45%), with the next most common being infliximab (30%), followed by etanercept (22%) and golimumab (3%). The most frequently used 1 st anti-tnfα varied over time: from 2001 to 2006 it was infliximab; from it was adalimumab. Etanercept was the most common 2 nd and 3 rd anti-tnfα. Table 2 and Table 3 show median drug survival rates and treatment response for first, second and third TNFα inhibitor therapy. Table 2: Drug survival rates from observational studies of Anti-TNFαs for PsA. (Taken directly from DANBIO Registry (Glintborg et al. 2013)) Anti- TNFαs Withdrawn Patients still treated ( n) after Still on treatment after 2 years 0 2 yrs 4 yrs 6 yrs 8 yrs 10 yrs Median drug survival (95% CI) 1 st % 2.2 (1.9, 2.5) 2 nd % 1.3 (1.0, 1.6) 3rd % 1.1 (0.7, 1.5) 31

32 Table 3: Proportion of patients achieving a response in terms of ACR 20/50/70, a Good EULAR response and DAS28 remission. (Constructed from data reported from DANBIO Registry (Glintborg et al. 2013)) Anti-TNFαs Response criteria ACR 20 ACR 50 ACR 70 Good EULAR response DAS28 remission 1 st 47% 33% 17% 45% 43% 2 nd 22% 13% 5% 19% 34% 3rd 18% 6% 2% 17% 22% The results of these two observational studies demonstrate that for a subgroup of patients whose drug survival on a 1 st anti-tnfα is short, a substantial proportion of patients nevertheless appear to benefit from switching TNFα inhibitors. However, treatment response to a second and later anti-tnfα is lower than to a first. There is uncertainty around how generalisable these results are to current use of anti-tnfαs in UK clinical practice. 3 Critique of manufacturer s definition of decision problem 3.1 Population The manufacturer specified the relevant population as people with active and progressive PsA who have responded inadequately to previous disease-modifying anti-rheumatic drugs (DMARDs) or TNF-α inhibitors. This matches the population specified in the NICE scope, though the licence (Marketing Authorisation) only states that ustekinumab is indicated in PsA patients with an inadequate response to DMARDs; there is no mention of TNF-α inhibitors. The ERG queried whether the population included patients with predominant axial disease, but the manufacturer confirmed that such patients were not included in the population. 32

33 3.2 Intervention The scope and decision both simply define ustekinumab as the intervention of interest. Ustekinumab is a human monoclonal antibody that acts as a cytokine inhibitor through targeting interleukin-12 (IL- 12) and interleukin-23 (IL-23). The product licence states that ustekinumab alone or in combination with methotrexate (MTX) is indicated for the treatment of active psoriatic arthritis in adult patients when the response to previous DMARD therapy has been inadequate. The recommended posology is an initial dose of 45mg administered subcutaneously, followed by a 45mg dose four weeks later, and then every 12 weeks thereafter. Alternatively, 90mg may be used in patients with a body weight greater than 100 kg: the licence does not mandate strict weight-based dosing Comparators The MS reflects the NICE scope in specifying the following comparators: a) alternative TNF- inhibitors, and b) conventional management strategies for active and progressive psoriatic arthritis excluding TNF- inhibitors. In their submission, the manufacturer presents two evaluations: (1) all of the licensed treatments are compared, along with conventional management, in anti- TNF- naïve patients (i.e. all are considered as alternative first-line options); (2) ustekinumab is compared with conventional management alone in a population who have experience of prior TNF- inhibitor therapy (i.e. ustekinumab is considered the only choice for any patient with an inadequate response to a previous biologic treatment). However, this second analysis more closely resembles a scenario where ustekinumab is the only available option after all reasonable TNF- inhibitor therapy options have been exhausted (i.e. end of line). The manufacturer does not present any evidence on the effectiveness of ustekinumab relative to an alternative to second- or third-line TNF- inhibitor for patients who have failed an initial anti- TNF- agent. 3.4 Outcomes Outcomes addressed in the stated decision problem matched those in the NICE scope, namely: disease activity, effect on concomitant skin condition, functional capacity, joint damage, disease progression, adverse effects of treatment, and health related quality of life. The actual measures used (e.g. American College of Rheumatology (ACR) criteria, PsA Response Criteria (PsARC), Psoriasis Area and Severity Index (PASI) response, Disability Index of the Health Assessment Questionnaire (HAQ- DI), modified van der Heijde-Sharp Score, SF-36) are similar to those used in earlier evaluations of biologic therapy in PsA. 3.5 Other relevant factors 33

34 The MS emphasises ustekinumab s novel mechanism of action, stating that this IL-12/IL-23 inhibitor..through its unique mode of action, provides PsA patients with a new option. This is especially important for patients who have failed on previous treatment with TNFαs. This statement raises questions about the definition of failure for TNFα inhibitors as a class, and the strength of evidence to support the use of ustekinumab in patients who have failed on one or more previous TNFα inhibitors. These issues are discussed in Section 4.2 of this report. 34

35 4 Clinical Effectiveness This section contains a critique of the methods of the review(s) of clinical effectiveness data, followed by a description and critique of the trials included in the review, including a summary of their quality and results and the results of any synthesis of studies. 4.1 Critique of the methods of review(s) The MS reported a systematic review conducted to determine the clinical efficacy and tolerability of ustekinumab and the specified comparator treatments for active PsA Searches Clinical effectiveness search The MS described the search strategies used to identify relevant clinical effectiveness studies about the use of ustekinumab (Stelara ) for treating active and progressive psoriatic arthritis. Search strategies were only briefly described in the main body of the submission, however full details were provided on request (clarification A2). The electronic databases MEDLINE and MEDLINE In-Process (via Ovid), EMBASE (via Ovid), and the Cochrane Library: Cochrane database of systematic reviews, register of clinical trials (CENTRAL), NHS HEED, HTA, DARE, CINAHL, Science Citation Index (Web of Science) and Conference Proceedings Index (Web of Science) were searched to identify clinical trials and systematic reviews on the use of ustekinumab and other comparators. In addition to the formal searches, reference lists of systematic reviews and meta-analyses identified were scanned for additional trials of relevance to the research question. Unpublished data on the clinical efficacy of ustekinumab from Janssen were also included. Searches were conducted in October 2012 and subsequently updated in July Search strategies for each database were documented in Section 10.2 appendix 2. There were no date or language limits. Overall the searches were appropriate and well documented, and included the use of both subject indexing terms (MeSH and EMTREE) and free text searching. Field searching, Boolean operators and truncation were used where required. All the databases required by NICE were searched. However, medical society and regulatory body websites were not searched. The search strategies used in the manufacturer's submission were limited to clinical trials and systematic reviews by applying appropriate methodological search filters, however a search for other 35

36 study designs such as cohort or case control studies might have provided useful supplementary information about safety Inclusion criteria Clinical efficacy: RCTs (parallel, cross-over, open-label, single- or double-blinded) evaluating ustekinumab for the treatment of active PsA in adults, reporting relevant clinical, quality of life, or safety outcomes were included in the review. Eligible interventions appeared to be ustekinumab and relevant TNFα inhibitors (etanercept, infliximab, adalimumab, and golimumab). Inclusion criteria relating to comparators were not explicitly reported, though it appears that TNFα inhibitors and conventional management were eligible, as stated in the decision problem. The stated inclusion criteria were appropriate. PRISMA flow diagrams were reported for both original and update searches, though reasons for excluding four studies from the quantitative synthesis were not reported. In response to a query from the ERG, the manufacturer clarified that one of these was the RESPOND study, which was excluded because it compared infliximab+mtx versus MTX alone and so could not be connected to the network used for the MTC (see clarification A3). The three remaining studies were ongoing trials of agents not licenced for use in PsA (apremilast and brodalumab). The justifications for excluding these trials seem appropriate. Adverse events: Alongside clinical efficacy trial data, the MS presented observational data and psoriasis trial data relating to adverse events. However, no inclusion criteria were reported for this set of studies. In response to the ERG s request for clarification, the manufacturer stated sources that were recently published and relevant to the decision problem were selected and included in the original submission (see clarification A23), apparently confirming that no explicit inclusion criteria were applied to safety evidence. The ERG compared the sources for adverse effects data with the Medical Review report from the FDA (2009) and found that they included all available at that time plus all that were predicted to be available at a later date, with the exception of data from the planned Pregnancy Research Initiative or Nordic Database Initiative. 4 It is not clear whether these data were available for inclusion in the MS. However, PsA data on anti-tnfα therapy from the British Society for Rheumatology Biologics Register was included; while an appropriate source of adverse event data, populations may not have been directly comparable across these different sources (see Section 4.2). 36

37 4.1.3 Critique of data extraction Clinical efficacy: Details of data extraction methods (such as procedures used to avoid errors and bias) were not reported in the MS. Therefore ERG could not comment on the robustness of the data extraction methods in general. Adverse events: It appears that adverse event data were not extracted systematically; in response to a request from the ERG, the manufacturer provided additional data on serious infections and injectionsite reactions that were not reported in the original MS (see clarification C4) Quality assessment Clinical efficacy: The manufacturer s quality assessment was adapted from CRD guidance 5 covering: randomisation, allocation concealment, similarity of groups at outset, blinding, differential dropout, selective reporting of outcomes, and use of intention to treat analysis. This tool was appropriate. However, it appears that this was only applied to the ustekinumab trials; the quality of other RCTs included in the NMA was not assessed in the MS (see Section 4.4). The ERG queried why, despite two near identical sample size calculations, the sample size for PSUMMIT 2 is only half that of PSUMMIT 1 (clarification A6). The only apparent difference between the sample size calculations was the assumption of a slightly larger proportion of patients receiving MTX at baseline in PSUMMIT 2 (60% vs. 50% in PSUMMIT 1). It is unclear whether this would explain the large difference in sample sizes, but in their response to the clarification, the manufacturer provided tables showing that both trials were adequately powered to detect the overall observed difference between treatment groups on ACR20. Adverse events: With the exception of the PSUMMIT trials, the manufacturer did not appear to assess the quality of studies included in the adverse events section of the MS. Observational evidence can be subject to confounding and selection bias, though these issues are considered in the summary below Evidence synthesis Direct evidence of clinical efficacy: The original MS presented data on the anti-tnfα naïve subpopulation separately for each of the two PSUMMIT trials, alongside separate anti-tnfα experienced subgroup data from PSUMMIT 2. Since the two PSUMMIT trials had almost identical protocols, the ERG considered it appropriate to combine these trials in a meta-analysis, and requested that the manufacturer present pooled outcomes for two populations: (i) anti-tnfα naïve patients only and (ii) all randomised patients, regardless of prior anti-tnfα exposure (see clarification A10). A critique of the PSUMMIT trials and their synthesis is given in Section 4.4 of this report. 37

38 Network meta-analysis: The original MS presented combined data on the anti-tnfα naïve subpopulation from the two PSUMMIT trials with data from RCTs evaluating TNFα inhibitors in a network meta-analysis (NMA), details of which are presented in Section 4.4 of this report. 4.2 Critique of trials of the technology of interest, their analysis and interpretation (and any standard meta-analyses of these) Three double-blind RCTs evaluating ustekinumab were retrieved from the searches (C0743T10, PSUMMIT 1 and PSUMMIT 2). 6-8 All three trials compared ustekinumab against placebo. However, concomitant therapies (DMARDs, NSAIDs and oral corticosteroids) were permitted for all patients, and since the majority of patients received concomitant therapies ***** **** **** ********* **** ******), these placebo groups are likely to reflect conventional non-biologic management strategies for the treatment of active PsA in patients in whom DMARDs have been tried unsuccessfully. The CO743T10 RCT 6 was excluded from the analysis because the ustekinumab dosing schedule (63 or 90mg every week for four weeks) differed from the licenced posology. Exclusion on this basis seems justifiable. The remaining trials (PSUMMIT 1 and 2) both use the recommended posology Description of PSUMMIT trial methods PSUMMIT 1 7 is a multicentre, double-blind, placebo-controlled RCT in which 615 adults with active psoriatic arthritis ( 5 tender and 5 swollen joints, C-reactive protein 3 0 mg/l) were randomly assigned to 45 mg ustekinumab, 90 mg ustekinumab, or placebo at week 0, week 4, and every 12 weeks thereafter. At week 16, patients with less than 5% improvement in both tender and swollen joint counts entered blinded early-escape and were given 45 mg ustekinumab (if in the placebo group) or 90 mg ustekinumab (if in the 45 mg group). At week 24, all remaining patients in the placebo group received ustekinumab 45 mg, which they received again at week 28 and every 12 weeks thereafter. Blinding was maintained up to and including 24 weeks. The planned duration of exposure to study agent including follow-up for safety was 100 weeks. PSUMMIT 2 8 appeared to employ almost identical methodology to PSUMMIT 1, with three main differences: (i) around half the total number of patients were randomised (n=312); (ii) planned follow-up was for just 60 weeks and; (3) at least 150 but not more than 180 included subjects could have been treated with single or multiple biologic anti-tnfα therapy (at least eight weeks of etanercept, adalimumab, golimumab, certolizumab pogol, or at least 14 weeks of infliximab). Consequently, 58% of PSUMMIT 2 patients (n=180) had been exposed to at least one TNFα inhibitor prior to enrolment; these patients constitute the entire anti-tnfα experienced subpopulation discussed in the MS and in this report Outcomes collected in the PSUMMIT trials 38

39 The primary outcome of both PSUMMIT trials was the proportion of patients achieving a 20% improvement in tender or swollen joint counts using the American College of Rheumatology assessment criteria (ACR20) at 24 weeks. Secondary joint response outcomes were ACR50, ACR70, Psoriasis Area and Severity Index (PASI) 75 response at week 24 among patients with 3% body surface area (BSA) psoriasis at baseline, physical functional status measured by the Disability Index of the Health Assessment Questionnaire (HAQ-DI), and radiographic progression assessed by changes in modified van der Heijde-Sharp Score (vdh-s) of the hands and feet. Though inconsistently reported in the original MS, the manufacturer clarified that modified PsA Response Criteria (PsARC) at week 24 was also collected as secondary outcome (response to clarification C2). Though not listed as primary or secondary outcomes, the MS also provided 24 week data from the PSUMMIT trials on dactylitis, enthesitis, health related quality of life (SF-36) and adverse events. Since the clinicaltrials.gov records 9, 10 do not list all the outcomes known to have been collected in these trials and in the absence of complete protocols, it was not possible to assess whether there was selective reporting of secondary/tertiary outcomes in the MS. However, for the main joint/skin response measures, selective reporting of outcomes is not an issue Internal validity of PSUMMIT trials Baseline comparability of groups: The manufacturer s critical appraisal rated the PSUMMIT 1 and 2 trials as high quality. Insufficient data are available for the ERG to properly verify each item, but based on the limited available information in the MS, related publications, 6, 7 and response to clarifications (see clarification A4), these ratings appear to be reliable. Table 13 (p.38) in the MS suggested that patient groups were broadly comparable across the PSUMMIT trials in terms of prognostic factors at baseline. The ERG asked for similar baseline data for the anti-tnfα naïve and anti-tnfα experienced subgroups from PSUMMIT 2 (see response to clarification A5). This revealed similar baseline characteristics for the anti-tnfα naïve populations in both PSUMMIT trials. As might be expected, duration of PsA was greater in the PSUMMIT 2 anti-tnfα experienced subgroup relative to the anti-tnfα naïve groups (by around two years on average). Anti-TNF experienced patients also appeared to be slightly ******* at baseline, and a ************* ******* ********** ** **** ******** were randomised to ustekinumab in this subgroup (see Table 4 below). Early escape and crossover between treatment groups: In both PSUMMIT trials, for patients who escaped early at week 16 (due a lack of joint improvement), week 16 data were carried forward to week 24. Therefore, complete observations were only available up to 16 weeks. If it is assumed that patients on active treatment were more likely to have responded during weeks than those on placebo, the treatment benefit of ustekinumab might have been underestimated at 24 weeks. However, 39

40 the ERG requested that response data over time were provided for ACR20, PsARC and PASI75 for both PSUMMIT trials (clarification A12), and the supplied data suggested that relative effects across treatment groups were fairly consistent over weeks. The ERG asked for the proportion of patients escaping early in each treatment group (clarification A15). The manufacturer s response showed that for both PSUMMIT trials, *** ********** ** ******** ******** ***** **** ******* ****** ******** *** ******* **** **** **** ****** ********* ***** ** ********** **** *** *********** **** ************ ***** ** ****** ** *** ******* *****. However, for the majority of patients, the recommended posology of ustekinumab is an initial dose of 45 mg administered subcutaneously, followed by a 45 mg dose four weeks later, and then every 12 weeks thereafter. It is unlikely that poorly-responding patients in clinical practice would be considered treatment failures at 16 weeks after having received only the initial 45mg dose of ustekinumab. Even if this were the case, the licence would only allow the option of switching to the higher 90mg dose for patients weighing more than 100 kg. At week 24, all remaining patients in the placebo groups crossed-over to receive 45mg ustekinumab. Consequently, the PSUMMIT trials cannot provide evidence on the effects of ustekinumab relative to placebo/conventional management after 24 weeks. 40

41 Table 4: Baseline characteristics of PSUMMIT trials according to prior biologic exposure TRIAL PSUMMIT2 PSUMMIT 1 Prior anti-tnfα use anti-tnfα experienced anti-tnfα naive anti-tnfα naive Ustekinumab Ustekinumab Ustekinumab Ustekinumab Ustekinumab 45 Ustekinumab 90 Treatment arm Placebo Placebo 45 mg 90 mg 45 mg 90 mg mg mg Placebo Age, mean years (SD) **** ****** **** ****** **** ****** **** ****** **** ****** **** ****** 47.1 (12.6) 46.8 (11.8) 47.4 (12.3) Sex, % male **** **** **** **** **** **** Race, % Caucasian **** **** **** **** ***** ***** Weight, mean kg (SD) **** ****** **** ****** **** ****** **** ****** **** ****** **** ****** 87.8 (20.6) 89.8 (21.5) 87.5 (21.7) Duration of PsA, mean *** ***** *** ***** *** ***** *** ****** *** ***** *** ***** years (SD) 6.1 (6.8) 7.0 (7.6) 6.7 (7.5) Duration of PsO, mean **** ****** **** ****** **** ****** **** ****** **** ****** **** ****** years (SD) 14.9 (13.0) 15.5 (12.1) 15.9 (12.8) Treatment history, % use DAS28, mean (SD) PASI score, mean (SD) Swollen Joint Count, mean (SD) Tender Joint Count, mean (SD) ****** ***** **** ***** ****** **** ****** ***** **** ***** ****** **** ****** ***** **** ***** ****** **** ****** ***** **** ***** ****** **** ****** ***** **** ***** ****** **** ****** ***** **** ***** ****** **** DMARD: 79.5; NSAID: 89.3 DMARD: 78.4; NSAID: 90.6 DMARD: 80.6; NSAID: 87.8 ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ***** ***** ***** ***** ***** ***** 5.1 (0.9) 5.2 (0.9) 5.2 (1.1) ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ***** ***** ***** ***** ***** ***** 11.5 (11.8)* 10.6 (8.5)* 11.7 (10.3)* **** ****** **** ****** **** ****** **** ***** **** ***** **** ***** 12.5 (7.8) 12.9 (8.3) 15.0 (10.2) **** ****** **** ****** **** ****** **** ****** **** ****** **** ****** 22.2 (13.9) 23.2 (13.7) 25.1 (15.0) Enthesitis, n (%) ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ***** ***** ***** ***** ***** ***** 142 (69.3) 154 (75.5) 145 (70.4) Dactylitis, n (%) ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ** ** ******** ***** ***** ***** ***** ***** ***** 101 (49.3) 99 (48.5) 96 (46.6) CRP mg/l, mean (SD) **** ****** **** ****** **** ****** **** ****** **** ****** **** ****** 16.9 (17.8) 18.0 (18.0) 15.9 (19.1) 41

42 TRIAL PSUMMIT2 PSUMMIT 1 Prior anti-tnfα use anti-tnfα experienced anti-tnfα naive anti-tnfα naive Ustekinumab Ustekinumab Ustekinumab Ustekinumab Ustekinumab 45 Ustekinumab 90 Treatment arm Placebo Placebo 45 mg 90 mg 45 mg 90 mg mg mg **** **** **** **** **** **** ****** ***** ****** ***** ****** ***** ****** ***** ****** ***** ****** ***** (5.10; 7.60) (5.05; 7.80) Patient s assessment of pain (VAS; 0-10 cm) Patient s global assessment of disease activity (VAS; 0-10 cm) Physician s global assessment of disease activity (VAS 0-10 cm) HAQ disability index (0 3) Psoriasis on 3% body surface area **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** **** ****** ***** ** ******* ** ******* ** ******* ** ******* ** ******* ** ******* 5.70 (4.40; 7.00) 6.50 (5.00; 7.60) 1.25 (0.75; 1.75) 6.30 (4.85; 7.35) 6.80 (5.30; 7.90) 1.25 (0.75; 1.63) Placebo 6.40 (4.80; 7.60) 5.85 (4.60; 7.30) 6.40 (4.90; 7.50) 1.25 (0.75; 1.75) 145 (70.7%) 149 (73.4%) 146 (70.9%) 42

43 4.2.4 External validity of PSUMMIT trials Prior DMARD use: Current NICE guidance for TNFα inhibitor therapy recommends: The psoriatic arthritis has not responded to adequate trials of at least two standard DMARDs, administered either individually or in combination. However, though all patients included in the PSUMMIT trials were required to have active arthritis despite current or previous DMARD and/or NSAID therapy, around 20% of patients in PSUMMIT 1 and 14% of patients in PSUMMIT 2 had never received any DMARD therapy prior to enrolment (Response to clarification A8). Consequently PsA patients in the PSUMMIT studies may have less severe disease than PsA patients considered eligible for biologic treatment in practice (see Section 4.4 for further discussion of this issue). The most commonly used prior DMARDs **** ************ ****** ************* ***** *** *********** *****. The proportions were similar across treatment groups and appear to be representative of routine clinical practice. Concomitant medication use: In response to a request from the ERG, the manufacturer clarified that *** ** ******** ** *** ******* ******* ********* ** ******* **** *** ** ******** ******** *********** ****** *** *** ******** ***************. These rates were broadly similar across treatment groups and are likely to reflect rates seen in clinical practice. Prior TNFα inhibitor use: For the anti-tnfα experienced subpopulation, the MS included only data from a subgroup of patients in the placebo-controlled PSUMMIT 2 trial. There was no direct or indirect comparative evidence on the efficacy of ustekinumab relative to TNF- inhibitors in this subpopulation, so the MS only evaluates ustekinumab against conventional management strategies for the patients who have experience with TNF- inhibitors. This would be the appropriate comparator in a scenario where TNF- inhibitors are considered to have been ineffective or intolerable**** ************ ****** ************* ***** *** *********** ***** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** ***********While the existing NICE guidance does not explicitly mention sequential treatment, in clinical practice a second TNF- inhibitor (usually one of the monoclonal antibodies) is commonly used if response is not achieved with the initial agent (frequently etanercept). Therefore, the ERG believes that the proper comparison for many of the anti-tnfα experienced patients would be ustekinumab versus a second- or third line TNF- inhibitor. The *** of anti-tnfα experienced patients in PSUMMIT 2 with 3 TNF- inhibitors are more likely to represent a truly anti-tnfα refractory population in which conventional management would be an appropriate comparator (see Section 5). However, separate outcome data were not available for these patients, and the ***** ****** ** ******** may make any results unreliable. 43

44 Table 5 Number of patients receiving 1, 2, 3 or more previous anti-tnfs (PSUMMIT 2) Placebo Ustekinumab 45mg Ustekinumab 90mg Total Randomized subjects who were previously treated with biologic anti-tnfα agent(s) ** ** ** *** 1 previous anti-tnf ** ***** ******* ** ***** ** ***** 2 previous anti-tnfs ** ***** ** ***** ** ***** ** ***** 3 previous anti-tnfs ** ***** ** ***** ** ***** ** ***** Duration of treatment for a prior anti-tnfα inhibitor also varied; **** ***** **** ** ******** ***** had experience of less than one year, *** had one to three years experience, and *** had three or more years experience (see response to clarification A8) Analysis of PSUMMIT trial data Section 6.5 of the MS presented outcome data from the PSUMMIT trials in a number of tables and figures, though this presentation was somewhat inconsistent and incomplete. The main points were: The results were reported variously by trial (PSUMMIT 1 and 2) and by prior anti-tnfα experience. All results should have been reported by trial in the first instance. Results were not reported as relative effects and confidence intervals or other measures of variance were frequently not reported. Longer-term follow-up was not consistently reported; though 52 week data were presented for PASI75, and detailed follow-up data over multiple timepoints were presented for ACR20, the remaining outcomes were only reported at 24 weeks. No results were reported for the pre-planned subgroup analyses of demographics, baseline disease characteristics or concomitant or prior use of medications. Results from a meta-analysis for the main outcomes were not presented. Given the similarity of the PSUMMIT trials, meta-analyses would have been appropriate, provided greater statistical power and allowed formal exploration of heterogeneity of estimates from the anti- TNFα naïve and experienced populations. In its request for clarifications, the ERG requested that the manufacturer: report the findings of the previously described subgroup analyses (A7); provide pooled data across the PSUMMIT 1 and 2 trials for anti-tnfα naive and all randomised patient groups (A10); report relative measures of effect 44

45 with 95% confidence intervals (A11); report response-over-time data in a consistent manner across outcomes (A12); report post-24 week data consistently across outcomes (A13); provide numerous missing confidence intervals and p values (A14) and; report discontinuation rates due to adverse events for both PSUMMIT trials (A25). The manufacturer responded to each of the ERG s requests and these data are summarised in Section Summary of PSUMMIT trials results 24 week outcome data: Though the manufacturer provided relative measures of effect and 95% CIs for pooled PSUMMIT data, on closer inspection of the spreadsheet the ERG noticed that the point estimates were based on simple additive calculations and confidence intervals were typed values without any underlying formulae. It was therefore unclear whether these pooled data took the relative weight of each PSUMMIT trial into account. However, the ERG conducted separate metaanalyses that provided almost identical results to the manufacturer s estimates, so these estimates appear to be reliable. Where available, pooled relative risks from the PSUMMIT trials (provided in response to clarifications A10-A14) are discussed here, alongside anti-tnfα experienced subgroup data from the PSUMMIT 2 trial. Pooled estimates incorporating all PSUMMIT patients are considered the primary analysis for each outcome, with any variations in subgroup estimates discussed in relation to this primary analysis. Response rates and relative risks for each analysis are reported in Table 6 and response rates over time are shown in Figure 1 to Figure 5. The manufacturer confirmed that their subgroup analyses were not adjusted for multiple testing (see response to clarification A7), nor did they report any test for interaction between the anti-tnfα naïve and experienced subgroups. The ERG calculated these values and corresponding relative risk reductions (RRRs) and reports them below. Since the response-over-time data in Figure 1 to Figure 5 show separate response rates for 45mg and 90mg ustekinumab treatment groups and the product licence implies the option of choosing one of these doses for heavier patients, the ERG also conducted analyses comparing ustekinumab doses with one another using standard meta-analytic techniques (see Table 7). ACR20/50/70: For randomised patients as a whole, ustekinumab was associated with a statistically significant improvement on the primary outcome of ACR20 response relative to placebo at 24 weeks (45% vs. 22%; RR 2.06, 95% CI 1.64 to 2.59). While the relative risk of response was slightly larger for the anti-tnfα experienced subgroup than the anti-tnfα naïve subgroup (Table 6), this difference was not statistically significant (test of interaction z=-0.58, p=0.57; RRR 0.82, 95% CI 0.41 to 1.63), suggesting no good evidence for a different treatment effect in anti-tnfα naïve and experienced 45

46 patients. There was no obvious difference between ustekinumab doses or interactions between dose and prior anti-tnfα experience (Table 7). As would be expected, fewer patients in all groups showed ACR50 and ACR70 response; though the relative benefit of ustekinumab increased as proportionally fewer placebo patients met these increasingly stringent response criteria (see Table 6). Estimates of relative ACR50 and ACR70 response were slightly larger for the anti-tnfα naïve subgroup, though again the difference did not indicate any significant subgroup effect (ACR50 z=0.49, p=0.62; ACR70 z=0.32, p=0.75). PsARC: Overall, ustekinumab was also associated with a statistically significant improvement in PsARC response relative to placebo at 24 weeks (58% vs. 35%; RR 1.65, 95% CI 1.40 to 1.95). Despite a marginally greater relative benefit in the anti-tnfα experienced subgroup, there was no good evidence for a difference in treatment effect between anti-tnfα naïve and experienced patients (z=-0.86, p=0.39; RRR 0.81, 95% CI 0.50 to 1.31). A small statistically non-significant trend favoured 90mg over 45mg among anti-tnfα naïve patients (RR 0.88, 95% CI 0.76 to 1.02) but not among anti-tnfα experienced patients (RR 1.18, 95% CI 0.83, 1.69). Given the licenced posology allows 90mg ustekinumab in patients with a body weight greater than 100 kg, the slightly poorer PsARC response to this higher dose observed among the heavier anti-tnfα experienced subgroup (47% for 90mg versus 55% for 45mg) is surprising. However, the overall effects of different ustekinumab doses do not differ significantly, and the data are too sparse to draw any clear conclusions about possible interactions between ustekinumab dose and prior TNFα inhibitor treatment. PASI75: Among patients with 3% BSA psoriasis skin involvement at baseline, ustekinumab was associated with a statistically significant improvement in PASI75 response relative to placebo at 24 weeks (57% vs. 9%; RR 6.51, 95% CI 4.25 to 9.96). The relative risk of PASI75 response was much larger for the anti-tnfα experienced subgroup than for the anti-tnfα naïve subgroup (RR 23.5 vs. 5.6; Table 6). This was driven by the almost complete lack of response among anti-tnfα experienced patients randomised to placebo. However, confidence intervals for the experienced subgroup were extremely wide and the difference between anti-tnfα naïve and experienced patient subgroups was not statistically significant (test of interaction z=-1.41, p=0.16; RRR 0.24, 95% CI 0.03 to 1.75). Regardless of whether patients were anti-tnfα naïve or experienced, the observed effect was similar for 45mg and 90mg ustekinumab doses (Table 7). HAQ-DI: Ustekinumab was also associated ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *** ******** ***** suggesting a beneficial effect on functional status. As with other outcomes, there was no obvious evidence of a different treatment effect in anti-tnfα naïve and experienced patients. 46

47 Radiographic progression: The MS reported (Section 6.5, table 23) ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** However, since baseline radiographic scores were not reported in the MS or response to clarifications, it is difficult to establish whether the observed changes were clinically meaningful. Unlike other outcomes, the manufacturer did not report these data separately for anti-tnfα experienced population. Soft tissue conditions: The response to clarifications confirmed that patients receiving ustekinumab was associated had a ************* *********** ***** ********** ** ********** *** ********** relative to placebo at 24 weeks. The observed effects were similar between anti-tnfα naïve and experienced patient subgroups (see Appendix) Dermatology Life Quality Index: Among patients with 3% BSA psoriasis skin involvement at baseline, median change from baseline DLQI score was ************* ******* for ustekinumab than for placebo at 24 weeks (***** *** *********** ***** *** **see Appendix). This meets most published definitions of a meaningful clinically important difference, 11 though the absolute severity values at baseline and follow-up were not reported so these changes cannot be interpreted in context. SF-36: ************* *********** ******* **** ******** SF-36 physical and mental component scores were reported for all randomised patients, as well as for anti-tnfα naïve and experienced patient subgroups (see Appendix). However, the lack of baseline data precluded interpretation of the clinical meaningfulness of these changes. 47

48 Table 6: ACR, PsARC and PASI75 response rates in PSUMMIT trials Trial results (% responders) Relative risk vs placebo (95% CIs) Placebo UST 45mg UST 90mg Comb. UST 45mg 90mg Comb. ACR 20 response at 24 weeks PSUMMIT 1 Anti-TNFα naïve (N=615) 22.8% 42.4% 49.5% 46.0% 1.86 (1.38, 2.50) 2.17 (1.63, 2.87) 2.01 (1.53, 2.65) PSUMMIT 2 Anti-TNFα naïve (N=132) 28.6% 53.5% 55.3% 54.4% 1.87 (1.08, 3.26) 1.94 (1.33, 3.33) 1.91 (1.14, 3.19) PSUMMIT 2 Anti-TNFα experienced (N= 180) 14.5% 36.7% 34.5% 35.6% 2.53 (1.27, 5.03) 2.38 (1.18, 4.79) 2.45 (1.28, 4.70) PSUMMIT 1 & 2 Anti-TNFα naïve (n= 747) 23.8% 44.4% 50.6% 47.5% 1.86 (1.43, 2.42) 2.13 (1.65, 2.74) 2.00 (1.57, 2.54) PSUMMIT 1 & 2 All randomised patients N=927) 21.9% 42.9% 47.6% 45.2% 1.95 (1.53, 2.50) 2.17 (1.71, 2.76) 2.06 (1.64, 2.59) ACR 50 response at 24 weeks PSUMMIT 1 Anti-TNFα naïve (N= % 24.9% 27.9% 26.4% 2.85 (1.72, 4.70) 3.20 (1.95, 5.24) 3.02 (1.89, 4.84) PSUMMIT 2 Anti-TNFα naïve (N=132) 7.1% 20.9% 31.9% 26.7% 2.93 (0.85, 10.08) 4.47 (1.39, 14.36) 3.73 (1.19, 11.71) PSUMMIT 2 Anti-TNFα experienced (N=180 ) 6.5% 15.0% 15.5% 15.3% 2.33 (0.76, 7.15) 2.41 (0.78, 7.39) 2.36 (0.84, 6.68) PSUMMIT 1 & 2 Anti-TNFα naïve (n= 747) 8.5% 24.2% 28.7% 26.5% 2.86 (1.80, 4.55) 3.39 (2.15, 5.33) 3.12 (2.02, 4.82) PSUMMIT 1 & 2 All randomised patients N=927) 8.1% 22.4% 26.2% 24.3% 2.78 (1.81, 4.27) 3.25 (2.14, 4.95) 3.01 (2.02, 4.50) ACR 70 response at 24 weeks PSUMMIT 1 Anti-TNFα naïve (N=615) 2.4% 12.2% 14.2% 13.2% 5.02 (1.96, 12.87) 5.86 (2.31, 14.83) 5.44 (2.21, 13.39) 48

49 PSUMMIT 2 Anti-TNFα naïve (N=132) 4.8% 9.3% 12.8% 11.1% 1.95 (0.38, 10.10) 2.68 (0.57, 12.57) 2.33 (0.53, 10.18) PSUMMIT 2 Anti-TNFα experienced (N= 180) 1.6% 5.0% 5.2% 5.1% 3.10 (0.33, 28.98) 3.21 (0.34, 29.96) 3.15 (0.39, 25.60) PSUMMIT 1 & 2 Anti-TNFα naïve (n= 747) 2.8% 11.7% 13.9% 12.8% 4.14 (1.85, 9.28) 4.94 (2.24, 10.91) 4.54 (2.11, 9.77) PSUMMIT 1 & 2 All randomised patients N=927) 2.6% 10.4% 12.3% 11.3% 4.03 (1.89, 8.60) 4.77 (2.26, 10.05) 4.40 (2.14, 9.02) PsARC response at 24 weeks PSUMMIT 1 Anti-TNFα naïve (N=615) 37.4% 56.1% 64.7% 60.4% 1.50 (1.21, 1.86) 1.73 (1.41, 2.12) 1.62 (1.33, 1.96) PSUMMIT 2 Anti-TNFα naïve (N=132) 38.1% 55.8% 57.4% 56.7% 1.47 (0.92, 2.34) 1.51 (0.95, 2.38) 1.49 (0.97, 2.28) PSUMMIT 2 Anti-TNFα experienced (N=180) 25.8% 55.0% 46.6% 50.8% 2.13 (1.32, 3.44) 1.80 (1.09, 2.99) 1.97 (1.25, 3.11) PSUMMIT 1 & 2 Anti-TNFα naïve (N=747) 37.5% 56.0% 63.3% 59.7% 1.49 (1.23, 1.82) 1.69 (1.40, 2.03) 1.59 (1.34, 1.90) PSUMMIT 1 & 2 All randomised patients (N=927) 35.2% 55.8% 60.2% 58.0% 1.59 (1.33, 1.90) 1.71 (1.44, 2.04) 1.65 (1.40, 1.95) PASI 75 response at 24 weeks PSUMMIT 1 Anti-TNFα naïve (N=440) 11.0% 57.2% 62.4% 59.9% 5.22 (3.22, 8.47) 5.70 (3.53, 9.19) 5.46 (3.41, 8.76) PSUMMIT 2 Anti-TNFα naïve (N=106) 10.0% 58.3% 62.5% 60.5% 5.83 (1.93, 17.67) 6.25 (2.08, 18.78) 6.05 (2.04, 17.98) PSUMMIT 2 Anti-TNFα experienced (N=135) 2.0% 45.5% 48.8% 47.1% 22.7 (3.18, ) (3.42, ) (3.34, ) PSUMMIT 1 & 2 Anti-TNFα naïve (N=546) 10.8% 57.5% 62.4% 60.0% 5.32 (3.42, 8.29) 5.78 (3.73, 8.97) 5.56 (3.61, 8.57) PSUMMIT 1 & 2 All randomised patients (N=681) 8.8% 55.1% 60.0% 57.6% 6.23 (4.03, 9.62) 6.78 (4.40, 10.44) 6.51 (4.25, 9.96) 49

50 Table 7: Ustekinumab 45mg vs. 90mg. Random effects meta-analyses (relative risk, 95% confidence interval) Outcome (24 weeks) PSUMMIT 1 PSUMMIT 2 (naïve) PSUMMIT 2 (experienced) Pooled PSUMMIT trials ACR (0.69, 1.06) 0.97 (0.66, 1.41) 1.06 (0.65, 1.73) 0.90 (0.76 to 1.07) PsARC 0.87 [0.74, 1.02) 0.97 (0.68, 1.40) 1.18 (0.83, 1.69) 0.94 (0.79, 1.11) PASI (0.76, 1.11) 0.93 (0.65, 1.35) 0.93 (0.59, 1.46) 0.92 (0.79, 1.08) Post-24 week outcome data: The original MS only reported response-over-time data up to 52 weeks for the primary outcome of ACR20. In response to a clarification from the ERG, the manufacturer also provided this data for PsARC and PASI75. These data are reproduced in Figure 1-Figure 5 below. After crossing over to active treatment at 24 weeks, ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** The timing of these changes confirms the basic efficacy of ustekinumab treatment. ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** ********* ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** *********** ***** *** In their response to a point for clarification, the manufacturer stated that post-24 week modified van der Heijde Sharp score data were not yet available, so the longer-term effects of ustekinumab on radiographic progression of joint disease cannot be established. Subgroup analyses: The authors described a series of subgroup analyses, summary data for which were supplied in response to a request from the ERG (clarification A7). The manufacturer confirmed that these analyses were not adjusted for multiple testing. *********** ********** ********* ********* ** ******* *********** ******* ********* ***** ** ************* ******** ******* *************** ** *********** ** ***** *** ** ********************** *** 50

51 *********** ***** *** *********** ***** *** **** The main subgroup analysis comparing anti- TNFα naïve and experienced subgroups has been commented on throughout the preceding sections. 51

52 Figure 1: ACR20 response through week 52 (PSUMMIT1) Figure 2: ACR20 response through week 52 (PSUMMIT2) * 52

53 Figure 3: PsARC response through week 52 (PSUMMIT1) Figure 4: PsARC response through week 52 (PSUMMIT2) 53

54 Figure 6: PASI 75 response through week 52; randomised subjects with 3% BSA psoriasis skin involvement at baseline (PSUMMIT1) Figure 5: PASI 75 response through week 52; randomised subjects with 3% BSA psoriasis skin involvement at baseline (PSUMMIT2) 54

55 4.2.7 ERG commentary on PSUMMIT trials The majority of evidence presented on the efficacy of ustekinumab for PsA in the MS was derived from two methodologically similar trials (PSUMMIT 1 and PSUMMIT 2) comparing 45mg and 90 mg doses of ustekinumab against conventional management (excluding the use of anti-tnfα inhibitors). These trials were of adequate methodological quality, though the interruption of the placebocontrolled phase at week 16 and the termination of the controlled phase altogether at 24 weeks provided only a very brief comparison for a chronic condition such as PsA. However, the available data suggest that ustekinumab is a more effective treatment than conventional management over this short period in terms of both joint and skin response, and that these benefits are likely to persist for at least 52 weeks. No data is available on longer term radiographic progression. PSUMMIT 2 differs from other RCTs of biologics in that it specifically included a subgroup of patients who had received prior treatment with TNFα inhibitors. On the basis of the PSUMMIT trial data, there is no convincing evidence of a substantial difference in the efficacy of ustekinumab between these anti-tnfα experienced patients and patients who had not received any prior anti-tnfα treatment. The relatively large PASI75 benefit observed in the anti-tnfα experienced subgroup may be an artefact of the small number of patients in the analysis and should be confirmed in a properly powered trial in this specific patient population. Any such future trial should clearly distinguish between prior anti-tnfα experience and anti-tnf- treatment failure in order to establish an appropriate comparator. A second- or third line TNF- inhibitor might have been a more clinically meaningful comparator than conventional management for many of the anti-tnfα experienced patients included in PSUMMIT Description and evaluation of adverse event studies As stated in the MS, four data sources were used to investigate adverse events associated with ustekinumab: (1) the PSUMMIT trials; (2) five-year extension of four RCTs of ustekinumab for psoriasis; (3) the Psoriasis Longitudinal Assessment and Registry (PSOLAR) study; and (4) an observational study from the British Society for Rheumatology Biologics Register (BSRBR) for anti- TNFα therapies in PsA. These appear to be appropriate sources to establish longer term safety information on ustekinumab (albeit in a psoriasis population) and anti-tnfα treatment in PsA. Other than the PSUMMIT trials, the manufacturer did not assess the quality of these studies. 55

56 With the exception of a slightly higher proportion of injection site reactions for 90mg observed in the PSUMMIT trials (1% 90mg vs. 0.6% 4mg vs. 0.4% placebo), the included trial data did not suggest any obvious excess in adverse events, serious adverse events or treatment discontinuation for ustekinumab treated patients. The PSOLAR study indicated similar unadjusted rates of major adverse cardiovascular events, and a possible trend toward lower rates of serious infections for ustekinumab relative to anti-tnfα therapy (0.98 per 100 patient-years, 95% CI 0.73 to 1.28), though such observational data can be confounded by differing selection criteria across treatments. The original MS only reported discontinuation rates due to adverse events for PSUMMIT1. The same data for PSUMMIT 2 were provided on request from the ERG (clarification A25). *********** ***** *** ************ ***** *** ************ ***** *** ************ ***** *** ************ ***** *** ************ ***** *** * Table 8: Discontinuation of study agent due to adverse events (PSUMMITtrials) Placebo Placebo -> ustekinumab 45 mg Ustekinumab 45 mg Ustekinumab 90 mg All ustekinumab PSUMMIT 1 through week 24 7/25 (3.4%) 0/58 (0%) 3/205 (1.5%) 3/204 (1.5%) 6/467 (1.3%) PSUMMIT 1 through week 52-3/189 (1.6%) 5/205 (2.4%) 7/204 (3.4%) 15/598 (2.5%) PSUMMIT 2 through Week 24 ****** ******* * ***** ****** ***** ****** ***** ****** PSUMMIT 2 through Week 60 * **** ****** ***** ****** ***** ****** ****** ****** 56

57 4.3 Critique of the indirect comparison and/or multiple treatment comparison and the trials identified and included The decision problem required ustekinumab to be compared with the other biologic agents licensed for the treatment of PsA in UK: adalimumab, etanercept, infliximab and golimumab. The systematic review by the manufacturer identified nine trials of these agents plus the two previously discussed trials of ustekinumbab. All nine trials were placebo controlled with no head to head comparisons of the anti-tnfαs. In the absence of head-to-head comparisons between alternative anti-tnfα agents, it was appropriate that the manufacturer used a network meta-analysis (NMA also known as MTC) approach to estimate the relative efficacy between the five agents of interest: adalimumab, etanercept, infliximab, golimumab and ustekinumab. Trials of all five agents had a common comparator, placebo, which allowed a network to be established. NMA was only undertaken for the anti-tnfα naïve population, which reflects the population included in the all the trials, other than the treatment experienced subgroup in PSUMMIT2. Table 9 presents the ERG s quality assessment of these trials, as such an assessment had not been included in the manufacturer s submission. All nine trials were double-blind placebo controlled RCTs and in general were of good quality. A full discussion of the ustekinumab PSUMMIT trials is presented in Section 4.2. The time point at which response was assessed varied across the trials: the ustekinumab trials assessed response at 24 weeks whereas the trials for the anti-tnfαs first assessed response at 12, 14 or 16 weeks, although assessments were also made at 24 weeks for all agents (although not in all trials). 57

58 Table 9: Quality assessment of included RCTs (from Rodgers et al) 12 Quality assessment criteria Etanercept Infliximab Adalimumab Golimumab Ustekinumab Mease 2000 Mease 2004 IMPACT IMPACT 2 ADEPT Genovese 2007 GO-REVEAL PSUMMIT 1 PSUMMIT 2 Eligibility criteria Y Y Y Y Y Y Y Y Y specified? Power calculation? Y Y Y Y Y Y Y Y Y Adequate sample size? Y Y Y Y Y Y Y Y Y Number randomised Y Y Y Y Y Y Y Y Y stated? True randomisation? Y Y Y Y Y Y Y Y Y Double-blind? Y Y Y Y Y Y Y Y Y Allocation of treatment Y Y Y Y NR Y Y Y Y concealed? Treatment administered Y Y Y Y Y Y Y Y Y blind? Outcome assessment Y Y Y Y Y Y NR Y Y blind? Patients blind? Y Y Y Y Y Y Y Y Y Blinding successful? NR NR NR NR NR NR NR NR NR Adequate baseline details Y Y Y Y Y Y Y Y Y presented? Baseline comparability? Y Y Y Y Y Y Y Y Y Similar co-interventions? Y Y Y Y Y Y Y Y Y Compliance with Y Y Y Y Y Y Y Y Y treatment adequate? All randomised patients Y Y Y Y Y Y Y Y Y accounted for? Valid ITT analysis? Y Y Y Y Y Y Y Y Y > 80% patients in followup Y Y Y Y Y Y Y Y Y assessment? Quality rating Good Good Good Good Good Good Good Good Good Y=Yes; N=No; NR=Not reported 58

59 Table 10: (1 of 2) Summary of trial population baseline characteristics (placebo arm only) Study PSUMMIT 1 PSUMMIT 2 Mease 2000 Mease 2004 IMPACT 1 IMPACT 2 ADEPT Genovese GO-REVEAL Placebo arm U n= 206 U n=104 E n= 30 E n=104 I n=52 I n=100 A n=162 A n=49 G n= 113 Age, mean years (SD) 47.4 (12.3) 47.6 (11.2) 43.5 (median) 47.3 (median) 45.2 (9.7) 45.2 (9.7) 49.2 (11.1) 47.7 (11.3) 47.0 (10.6) Sex, % male Weight, mean kg (SD) 89.5 (22.6) 81.4 (median) (16.5) 88.5 (21.1) - Duration of PsA, mean years (SD) 6.7 (7.5) 8.5 (8.5) 9.5 (median) 9.2 (median) 8.5 (6.4) 7.5 (7.8) 9.2 (8.7) 7.2 (7.0) 7.6 (7.9) Duration of PsO, mean years (SD) 15.9 (12.8) 15.2 (11.8) 17.5 (median) 19.7 (median) 19.4 (11.6) 16.8 (12.0) 17.1 (12.6) 13.8 (10.7) 19.0 (12.9) Number of prior DMARDs: mean (SD) 21% = 0 2% = 0 ***** * * * * **** * * * ***** * *** *** * ** - 50% = 1 19% = 2 38% = 1 48% = % = 1-2 9% = % = 3+ Treatment history % NSAID MTX Corticosteroid 87.8 **** PASI score baseline, mean (SD) 11.7 (10.3) 11.3 (9.3) 6.0 (median) (6.6) 10.2 (9.0) 8.3 (7.2) (9.5) Patients evaluable for PASI at baseline (%) HAQ (0-3) Mean (SD) 1.25 (median) 1.13 (median) 1.2 (median) 1.1 (median) 1.2 (0.7) 1.1 (0.6) 1.0 (0.7) 1.1 (0.6) 1.03 (0.55) Swollen Joint Count, mean (SD) 15.0 (10.2) 13.5 (9.9) 14.7 (median) 15.3 (median) 14.7 (8.2) 14.4 (8.9) 14.3 (11.1) 18.4 (12.1) 13.4 (9.8) Tender Joint Count, mean (SD) 25.1 (15.0) 23.4 (14.9) 19.0 (median) 22.1 (median) 20.4 (12.1) 25.1 (13.3) 25.8 (18.0) 29.3 (18.1) 21.9 (14.7) Patients with 3% BSA psoriasis at baseline; Patients with a baseline PASI score

60 Table 11: (2 of 2) Summary of trial population baseline characteristics (treatment arm only) Study PSUMMIT 1 PSUMMIT 2 Mease 2000 Mease 2004 IMPACT 1 IMPACT 2 ADEPT Genovese GO-REVEAL Treatment arm Age* U 45 mg n= (12.6) U 90 mg n= (11.8) U 45 mg n= (11.2) U 90 mg n= (12.4) E n= (median) E n= (median) I n= (11.1) I n= (12.8) A n=151 A n= 51 G 50mg n=146 G 100mg n= (12.5) 50.4 (11.1) 45.7 (10.7) 48.2 Sex, % male Duration of PsA* Duration of PsO* Number of prior DMARDs* Treatment history, % use NSAID MTX Corticosteroid 6.1 (6.8) 14.9 (13.0) ***** * * ***** * * **** * * 7.0 (7.6) 15.5 (12.1) ***** * * ***** * * **** * * 8.2 (8.6) 15.4 (11.2) ***** * * ***** * * ***** * * (7.5) 14.8 (12.7) ***** * * ***** * * ***** * * (median) 19 (median) (median) 11.7 (9.8) 8.4 (7.2) 18.3 (median) 16.9 (10.9) NR 27% = 0 40% = 1 20% = % = 0 52% = 1 37% = % = % = % = ( 10.9) 9.8 (8.3) 7.5 (7.0) 7.2 (6.8) 7.7 (7.8) 17.2 (12.0) 18.0 (13.2) 17.7 (11.9) 18.4 (12.7) % = % = % = % = PASI score baseline 11.5 (11.8) 10.6 (8.5) 13.4 (13.0) 12.1 (10.3) 10.1(median) CiC 8.6 (6.6) 11.4 (12.7) 7.4 (6.0) (8.6) 11.1 (9.5) Patients evaluable PASI baseline (%) HAQ (0-3)* Swollen Joint Count* Tender Joint Count* 1.25(median) 1.25(median ) 12.5 (7.8) 22.2 (13.9) 12.9 (8.3) 23.2 (13.7) 1.25(median ) 15.0 (9.2) 27.2 (15.4) 1.06(median) 1.3(median) 1.1(median) 1.2 (0.7) 1.1 (0.6) 1.0 (0.6) 0.9 (0.5) 0.98 (0.65) 1.05 (0.62) 14.0 (10.9) 25.9 (15.5) 14.0 (median) 15.9 (median) 22.5 (median) 20.4 (median) * unless otherwise stated Mean (SD); Patients with 3% BSA psoriasis at baseline ; Patients with a baseline PASI score (7.5) 13.9 (7.9) 14.3 (12.2) 18.2 (10.9) 14.1 (11.4) 12.0 (8.4) 23.7 (13.7) 24.6 (14.1) 23.9 (17.3) 25.3 (18.3) 24.0 (17.1) (15.7) 60

61 The validity of the NMA is built on the assumptions that no important differences exist between trials in terms of baseline characteristics such as disease severity. The population characteristics of all nine trials are summarised and compared in Table 10 and Table 11. Of interest is the between study comparability, therefore one table presents treatment arm patient characteristics, the other presents the placebo arm characteristics. For a combined table, including some further characteristics not included in the ERG tables, see manufacturer s points for clarification (C3; pg. 58, Section 6.7; also presented in Appendix 10.2). The trials are generally similar in terms of patients characteristics: age, % male, race, duration of PsA and psoriasis (where reported). Importantly they are generally similar in terms of patients joint disease severity at baseline particularly mean tender joint count and mean swollen joint count. The high standard deviations in each of the trials suggest that the variation between trials is unlikely to be of significance; however these differences are worth noting. The same can be said of the baseline HAQ score, although the use of median values in both the ustekinumab trials and the two etanercept trials make the comparison less transparent. There was a concern about the correlation between baseline HAQ scores and absolute HAQ changes in these PsA patients, given such a high variability of these HAQ values; however as in previous submissions the ERG considered the exchangeability of mean HAQ scores across the included trials in the MTC analysis to be acceptable. There were some differences in the proportions of patients evaluable for psoriasis endpoints at baseline and variation in mean PASI score between the included trials. Where reported, the PSUMMIT 1, PSUMMIT 2, IMPACT 2, GO-REVEAL, and Mease 2004 trials had higher percentages of patients evaluable for psoriasis endpoints at baseline. In contrast, the IMPACT and ADEPT trials had much lower percentages of patients evaluable for psoriasis endpoints. Thus, there might be potential interactions between different patient samples and the treatment effect when estimating the relative efficacy in terms of skin disease response. As noted in Section 4.2, a substantial minority of patients in the PSUMMIT trials had never received prior DMARDs. However, this is not unique to the PSUMMIT trials; prior DMARD exposure was variable across trials, and where reported the mean number of prior DMARDs was typically fewer than two (see Table 12). On the basis of the available information, it would appear that baseline disease severity is broadly comparable across trials, though the MTC population may be less severe than that routinely considered for biologic treatment in practice. 61

62 Table 12 Baseline DMARD use among RCTs included in the MTC Trial Drug evaluated % of patients without any prior DMARD use % receiving 1-2 prior DMARDs Mean number of prior DMARDs at baseline Treatment Control Treatment Control Treatment Control Mease 2000 Etanercept Mease 2004 Etanercept IMPACT Infliximab IMPACT 2 Infliximab ADEPT Adalimumab Genovese Adalimumab GO-REVEAL Golimumab PSUMMIT 1 Ustekinumab **** **** - - PSUMMIT 2 Ustekinumab **** **** - - Despite some limitations mentioned above, overall, in the NMA presented by the manufacturer the degree of clinical heterogeneity between the included trials in terms of joint and skin disease severity and functional status was reasonable and, the assumption of exchangeability between the trials for the purposes of the analyses is considered acceptable. Separate networks were constructed for three outcomes PASI75, PASI90 and PsARC. These three outcomes were considered for two time points weeks and 24 weeks. The period of weeks was used due to the different response criteria utilised in trials for the alternative treatments. These outcomes are appropriate and reflect those used in previous evaluations of PsA treatment. Due to data limitations not every trial measured or reported each of the outcomes at each of the time points and thus not all nine trials were included in all six networks. The trials included in the NMA are listed in Table 13, along with the outcomes for which they provide data. These trials appear to represent the complete relevant body of evidence for the comparators included. The network diagrams are presented in the manufacturer s submission (Figures 10, 11 and 12, pages 74 and 75). 62

63 Table 13: Summary of the trials used* to conduct the MTC (Manufacturer s submission, table 1, pg 73.) Week outcomes 24 Week outcomes Study Treatment arms PASI75 PASI90 PsARC PASI75 PASI90 PsARC PSUMMIT 1 Ustekinumab 45 mg Ustekinumab 90 mg PSUMMIT 2 ADEPT Genovese et al Placebo Ustekinumab 45 mg Ustekinumab 90 mg Placebo Adalimumab 40 mg Placebo Adalimumab 40 mg Placebo Mease et al Etanercept 25 mg Placebo Mease et al Etanercept 25 mg GO-REVEAL Placebo Golimumab 50 mg Golimumab 100 mg Placebo IMPACT 1 Infliximab 5 mg Placebo IMPACT 2 Infliximab 5 mg Placebo *The infliximab RESPOND trial was not included in the mixed treatment comparison because the comparator was MTX. As all other trials used placebo, it was not possible to integrate these data. The NMA was checked by the ERG. The NMA was conducted using R programming; the code was supplied and appears to be appropriate. A Bayesian random effects model was fitted to the data. Following a points for clarification request, the manufacturer confirmed that the model was run for 220,000 iterations with a burn-in of 10,000 and thinning of 100. They further stated that trace plots were examined to see how well the chain was mixing. The model achieved convergence and samples were drawn from the posterior distribution. No fit statistics were produced or considered by the manufacturer. Trace plots were not presented to the ERG. Where possible, the ERG ran its own NMA models in an attempt to validate the manufacturer s results. Full details of these are presented in appendix For adalimumab, etanercept, infliximab and golimumab data were retrieved, where available, from the trial publications. (ADEPT, Genovese et al 2007, Mease et al. 2000, Mease et al. 2004, GO-REVEAL, IMPACT 1, IMPACT 2) For ustekinumab, a separate patient level data analysis was undertaken to obtain PsARC and PASI response rates for a weight-based dosing regimen, i.e. 45mg for patients 63

64 with body weight 100kg; and 90mg for those with body weight >100kg. These data were presented in the submission alongside other NMA data. (Section 6.7, Table 35, pg. 76) To aid clarity they are presented again separately in Table 14. The manufacturer assumed that in practice patients will be treated according the weight-based dosing criteria, and therefore it was deemed appropriate to exclude subjects not treated this way in the PSUMMIT trials. However, it is not clear to the ERG that this is in line with the licence or the manner in which the trials were conducted so such an analysis is not necessarily the most appropriate one. Furthermore, a significant number of subjects in the PSUMMIT trials were not treated according to the weight based dosing rule and therefore a significant amount of data were lost from these analyses. Table 14 Number of patients included in weight-based NMA Trial Arm Number randomised (naïve) Number meeting PASI criteria 12 week total number included in weight-based analysis PASI 75 PASI 90 PsARC PSUMMIT 1 45mg PSUMMIT 1 90mg PSUMMIT 2 45mg PSUMMIT 2 90mg Trial Arm Number randomised (naïve) Number meeting PASI criteria 24 week total number included in weight-based analysis PASI 75 PASI 90 PsARC PSUMMIT 1 45mg PSUMMIT 1 90mg PSUMMIT 1 45mg PSUMMIT 1 90mg The ERG requested that the manufacturer re-run their NMA using the entire intention to treat population, as opposed to the restricted weight-based dosing population. The results for both sets of NMA results are shown in Table 15 for PASI 75 response, Table 16 for PASI 90 response and Table 17 for PsARC response. 64

65 Table 15: PASI 75 - Proportion of responders for the anti-tnfα naïve population, using the weight-based dosing criteria compared with ITT PASI 75 Response PASI 75 Response Treatment WEIGHT BASED ITT Week Week 24 Week Week 24 Mean (%) Standard Error Mean (%) Standard Error Mean (%) Standard Error Mean (%) Standard Error Ustekinumab 45mg ***** **** ***** **** ***** **** ***** **** Ustekinumab 90mg ***** **** ***** **** **** **** ***** **** Golimumab 50mg ***** **** ***** **** ***** **** ***** **** Golimumab 100mg ***** **** ***** **** ***** **** ***** **** Adalimumab ***** **** ***** **** ***** **** ***** **** Etanercept 25mg/50mg **** **** **** **** ***** **** ***** **** Infliximab ***** **** ***** **** ***** **** ***** **** Table 16. PASI 90 - Proportion of responders for the anti-tnfα naïve population, using the weight-based dosing criteria compared with ITT PASI 90 Response PASI 90 Response Treatment WEIGHT BASED ITT Week Week 24 Week Week 24 Mean (%) Standard Error Mean (%) Standard Error Mean (%) Standard Error Mean (%) Standard Error Ustekinumab 45mg ***** **** ***** **** ***** **** ***** **** Ustekinumab 90mg ***** **** ***** **** ***** **** ***** **** Golimumab 50mg ***** **** ***** **** ***** **** ***** **** Golimumab 100mg ***** **** ***** **** ***** **** ***** **** Adalimumab ***** **** ***** **** ***** **** **** **** Infliximab ***** **** ***** **** ***** **** ***** **** 65

66 Table 17. PsARC - Proportion of responders for the anti-tnfα naive population, using the weight based dosing criteria compared with ITT PsARC Response Treatment WEIGHT BASED ITT Week Week 24 Week Week 24 Mean (%) Standard Error Mean (%) Standard Error Mean (%) Standard Error Mean (%) Standard Error Ustekinumab 45mg **** **** ***** **** ***** **** ***** **** Ustekinumab 90mg ***** **** ***** **** ***** **** ***** **** Golimumab 50mg ***** **** ***** **** ***** **** ***** **** Golimumab 100mg ***** **** ***** **** ***** **** ***** **** Adalimumab ***** **** ***** **** ***** **** ***** **** Etanercept 25mg/50mg ***** **** ***** **** ***** **** **** **** Infliximab **** **** ***** **** ***** **** ***** **** In general the results are fairly similar across both the strict weight-based and full ITT populations, with only the 24 week PASI 75 in the ustekinumab 90mg arm showing some variation. The ustekinumab 90mg arm was most affected by the switch to ITT data, which is to be expected given that the weight based dosing criteria resulted in a large portion of subjects in that arm to be excluded. The ERG believe that it is important to be aware of the impact of these alternative analysis, but acknowledge that given the licence it is unlikely that patients <100kg will receive a 90mg dose in clinical practice so the use of the weight-based data for the 90mg arm is reasonable. However, the ERG clinical advisor suggested that it would be expected in practice that patients will be administered ustekinumab 45mg initially regardless of their weight making the full ITT analysis results (PsARC, PASI 75 and 90) for this arm more representative of clinical practice. The impact of using the ITT data for this population is assessed and discussed in Section In addition, the ERG requested that an analysis be undertaken pooling the 45mg and 90mg arms of the PSUMMIT trials. These results are presented in the manufacturer s points for clarification (A21, pg. 40). The pooling of doses does not appear to impact on the effectiveness of ustekinumab relative to its ranking in terms of probability of response. Overall the results of the NMA found that *********** **** had the lowest or one of the lowest response rates for PASI 75 and 90 and PsARC. For PASI 75 response, ********** was the most effective treatment, and for PASI 90 response ********** was the most effective treatment. Across both the weight-based and ITT analyses, ******* achieved the greatest proportion of PsARC responders, followed by *********** ***** *** **********and **********. 66

67 In the two previous submissions (TA199/TA220) a NMA of HAQ conditional on PsARC response was undertaken. For this submission the manufacturer did not undertake this analysis, rather they used the results from the previous TA220 submission to inform the estimates for the comparator drugs. They then utilised the PSUMMIT trial data to inform the ustekinumab estimates. In brief, the ERG opted to undertake a NMA of these data to appropriately incorporate the ustekinumab trials. Overall, despite the heterogeneity between trials the ERG believe that it was appropriate to undertake the network meta-analysis. The results obtained by the manufacturer were robust when compared with the results of the ERG analysis. The comparative results are presented in appendix Conclusions of the clinical effectiveness section While PSUMMIT trial data suggest that ustekinumab (45mg and 90mg) is a more effective treatment than conventional management over 24 weeks in terms of PASI and PsARC response among anti- TNFα naïve patients, the NMA suggests that ustekinumab 45mg has lower PASI and PsARC response rates than all currently available TNFa inhibitors for PsA. Pooling of 45mg and 90mg doses does not appear to impact on the effectiveness of ustekinumab relative to its ranking in terms of probability of response. The treatment effect of ustekinumab did not differ significantly between the anti-tnf naive and experienced patients. Evidence for the relative efficacy of ustekinumab and anti-tnfs in anti- TNF experienced patients is not available. Futhermore **** * **** ***** ****** ** ******** in the anti-tnf experienced subgroup of PSUMMIT could be considered having truly failed anti-tnf therapy (having tried > 3, ****) 67

68 5 Cost Effectiveness This section focuses on the economic evidence submitted by the manufacturer and the supplementary information provided post the ERG s points for clarification. The manufacturer s initial economic submission included: A systematic review of existing economic evaluations of therapies used in the treatment of psoriatic arthritis (MS, Section 7.1) with additional information presented in an appendix (, Appendix 10). A report on the de novo economic evaluation conducted by the manufacturer. The report described the technology; comparators and patient population (MS, Section 7.2); clinical parameters and variables (MS, Section 7.3); the assumptions and sources of evidence used to assess quality of life (MS, Section 7.4); the resource use and unit cost assumptions and sources (MS, Section 7.5); sensitivity analysis (MS, Section 7.6); and the base-case costeffectiveness results (MS, Section 7.7). An Excel-based model comprising the manufacturer s electronic economic model. The ERG has noted that the Excel-based model allows the user to produce the results of scenarios that are not presented or discussed within the main submission. In response to the request for clarification made by the ERG (see manufacturer s response to ERG s points for clarification, Section B), the manufacturer further submitted: An updated version of the previously submitted Excel-based model. Clarification around the resource use estimates, some of the scenario analyses, the natural history of the disease estimates, and the agreement to supply nurses free of charge to administer ustekinumab to patients. The components of the submission were subject to a critical review on the basis of the manufacturer s report and by direct examination of the electronic version of the economic model. A narrative review was undertaken to highlight key assumptions and possible limitations. Section 6 of this report presents additional work undertaken by the ERG to address and explore some remaining uncertainties. 5.1 Overview The manufacturer conducted a review of published cost-effectiveness studies of therapies used in the treatment of PsA. The review identified a number of previous economic models; although no other evaluations which included ustekinumab as a comparator were included. Most of the evaluations 68

69 identified were developed or based on those developed for NICE technology appraisals. 12, 23, 24 A critique of the identified studies was undertaken and presented. The majority of the models adopted the same structure, and the manufacturer chose a similar structure to model the cost-effectiveness of ustekinumab. In line with the NICE scope, ustekinumab was compared to adalimumab, etanercept, infliximab, golimumab and conventional care. The population considered in the economic model was adult patients with active PsA who had failed to achieve an adequate response to previous DMARD therapy. Within this group of patients two distinct populations were considered; these will be discussed in more detail in Section The economic model comprised two elements: an initial short-term decision tree, which evaluates a patient s initial response to the biologic treatment; followed by a longer-term Markov model, which captures the longer term costs and consequences of treatments. The time horizon of the model was 52 years (i.e. lifetime). The decision tree element of the model is structured around treatment response rules: patients achieving an initial PsARC response at 12 weeks for adalimumab, etanercept, infliximab and golimumab, and 24 weeks for ustekinumab were considered to be responders and continued on an active treatment. Based on these responses patients enter the Markov model which evaluates patients through four initial cycles of 12 weeks followed by annual cycles across the model states. Patients not achieving an initial PsARC response are considered non-responders and withdraw to conventional management. After initial response a constant annual withdrawal rate was applied to allow for long-term discontinuation of treatment due to adverse events or lack of efficacy. Full details of the model structure are discussed in Section Health states within the Markov model captured both the joint (PsARC) and skin (PASI) components of PsA, as both of these are related to quality-of-life and costs. PASI score changes were applied dependent on PASI 75 response and health assessment questionnaire (HAQ) scores were applied dependent on PsARC (see Sections and ). HAQ and PASI improvements (i.e. reductions) were estimated using clinical trial data and network meta-analyses. Health-related qualityof-life was measured using quality-adjusted life-years (QALYs). A previously published mapping algorithm was used to relate both HAQ and PASI scores to health-related quality-of-life (i.e. utilities). The mapping algorithm has been used in previous submissions and is discussed further in section Resource use assumptions were based on a previous publication (Rodgers et al., 2011) and the British Society for Rheumatology guidelines. 12 These data were then validated by experts at an advisory board meeting. In addition, health state costs were estimated as a function of HAQ and PASI through 69

70 the use of previously published work. List prices were taken from the BNF (2013) for all treatments. 25 Adverse event costs were not included in the model, as it was reported that adverse events related to treatment tend to be infrequent and minor. These issues are discussed fully in section A probabilistic analysis was undertaken and a number of scenario analyses were conducted to test the impact of uncertainty. Deterministic and probabilistic results were presented separately for two populations: anti-tnfα naïve patients and anti-tnfα experienced patients. A discussion of these results and their limitations are presented in section Table 18 presents a summary of the manufacturer s economic evaluation, with signposts to the relevant sections of the MS. Table 18: Summary of the manufacturer s economic evaluation (and signposts to MS) Approach Source/Justification Location in Population The population were adult patients with active PsA for whom response to previous DMARD therapy had been inadequate, which broadly matched the clinical trials. Two distinct sub-groups were considered (i) anti-tnfα naïve patients; (ii) anti-tnfα experienced patients. The adult patient with active PsA for whom response to previous DMARD therapy has been inadequate is based on the licensed indication for ustekinumab and the NICE scope. The two sub-groups were identified as relevant under the licensed indication; and for the anti-tnfα experienced patients a clear unmet need. Section 2.6; pg 16 Section 7.2; pg 96 Comparators The comparators for anti-tnfα naïve patients included: golimumab, infliximab, etanercept, adalimumab and conventional management. The comparator for anti-tnfα experienced patients was conventional management. For the for anti-tnfα naïve patients the selection was based on the product indications and the relevant NICE guidelines. The selection was also consistent with the scope. For the anti-tnfα experienced patients the use of conventional management as the sole comparator was based on a lack of available RCT evidence regarding secondline use of anti-tnfα therapies. Section 7.2.7; pg 103 to 104 Model, states and events A cost-effectiveness (cost-utility) analysis was undertaken using a combined decision tree and Markov model. An initial treatment response was applied in the decision tree and the resulting distribution of patients was used to populate an eight state Markov model. States included: No initial response (PsARC or PASI 75); Initial PASI 75 response only; Initial PsARC response only; Initial PsARC and PASI 75 response; conventional management; initial PsARC but withdrawn to conventional management; initial PsARC and PASI 75 but withdrawn to conventional management; and death. A similar model structure has been used in 12, 24 previous submission. It differs in the use of a longer decision rule to reflect the anticipated treatment rules associated with ustekinumab. Initial response is evaluated at 12 weeks for all of the comparators and the resulting decision to continue treatment or withdraw to conventional management; however for ustekinumab administration of the drug is likely to be less frequent and it is anticipated that patients will be assessed for response at 24 weeks rather than 12 weeks. Therefore, a decision rule of response at 24 weeks was used for ustekinumab (only). Section 7.2.2/3; pg 96 to

71 Natural history Natural history of the condition was based on an annual worsening of HAQ for those patients who are receiving conventional management. It was assumed that those patients responding to an anti-tnfα or ustekinumab did not experience disease progression. The assumption and data were derived Section 12, 24 from previous submission ; pg 100 Treatment effectiveness Effectiveness was measured as HAQ score change dependent on PsARC response, and PASI score change dependent on PASI 50, 75 and 90 responses. Data from 9 trials were utilised in the evidence synthesis model. 8, In addition, previous NMA results and the PSUMMIT trial results were used to inform these parameters. Section 7.3; pg 105 to 121 HRQoL QALYs are measured as a function of HAQ and PASI. In the base case, an algorithm estimating the utilities based on HAQ and PASI from Rodgers et al/ Yang et al. was used. 24, 34 In addition, an algorithm from Rowen et al was used to map SF-36 data from the two PSUMMIT trails to EQ-5D and OLS regressions were undertaken to estimate the relationship between EQ- 5D as the dependent variable and HAQ and PASI scores as explanatory variables. 35 The use of the algorithm and data from previous submissions was used in the base case to allow comparability. The additional mapping and OLS regressions were undertaken to explore some of the differences in the data. Section 7.4.3; pg 124 to 125 Section 10.20; Appendix 20. Adverse events Not included This was justified on the basis of the lack of consideration in earlier submissions. Section 7.4.8; pg. 128 Section 7.5.7; pg 142. Resource use and costs Resources use associated with active treatment were taken from Rodgers et al. 34 Those associated with conventional management were elicited from the experts attending the advisory board meeting. It is assumed that those treatments given by subcutaneous injection will be self-administered, with the exception of ustekinumab. For ustekinumab it is assumed that 70% of patients will have the injection administered by a nurse. However, the cost of the nurse will be covered by the company. Costs for hospital visits, monitoring tests, and follow-up care were taken from NHS reference costs. As outlined the experts at the advisory board meeting comprised one external clinical expert (a dermatologist) and four external health economic advisors. The assumption of self-administration and resource use were validated by the advisory board. These costs were originally derived from the Kobelt et al study. 36 Section 7.3.5; pg. 109 Section 7.5.5; pg. 137 to 142 Section 7.5.6; pg 141 Drug acquisition costs were taken from the BNF 65 (July 2013). 25 The on-going costs of managing PsA were estimated as a function of HAQ, which was taken from Yang et al. 24 Costs of psoriasis, as a function of PASI, were derived from Yang et al. 24 Discount rates 3.5% per annum for QALYs and costs NICE reference case Section 7.2.6; pg 71

72 103 Sensitivity analysis A number of alternative parameter scenarios were evaluated. These included: initial response criteria, HAQ rebound on withdrawal from treatment, cost assumptions, utility values, withdrawal rates, mortality and weightbased dosing. Probabilistic sensitivity analysis was also undertaken. Some justification was given on the choice of alternative parameter values and assumptions used in the sensitivity analysis. A number of parameters were not assigned distributions or were assigned incorrect distributions, no justification was provided for these omissions. Section 7.6.1; pg 143 Section 10.9; appendix ERG comment on manufacturer s review of cost-effectiveness evidence The MS described the search strategies used to identify cost-effectiveness studies and utilities relevant to this appraisal of Ustekinumab (Stelara) for treating active and progressive psoriatic arthritis. Search strategies were only briefly described in the main submission, however full details were provided on request. The electronic databases MEDLINE and MEDLINE In-Process (via Ovid), EMBASE (via Ovid), the Cochrane Library including CDSR, CENTRAL, HTA, DARE, Science Citation Index (SCI), Cinahl (via Ebsco) and Econ LIT were searched. In addition to the database searches, a further search was performed for relevant documentation from NICE technology appraisals (TAs) of interventions for PsA such. Database searches were performed in October 2012 and subsequently updated in July Search strategies for each database were documented in the MS Section 10.10, appendix 10. No language or date limits were applied to the search. Methodological search filters were included to identify economic studies and utilities in MEDLINE, EMBASE and Cinahl. The searches were appropriate and comprehensive, and included the use of both subject indexing terms and free text searching. Field searching, Boolean operators and truncation were used where required. All NICE required databases were searched though medical society and regulatory body websites were not. The review identified a number of previous economic models; although no other evaluations which included ustekinumab as a comparator were included. Most of the evaluations identified were developed or based on those developed for NICE technology appraisals. 12, 23, 24 A critique of the identified studies was undertaken and presented. The majority of the models adopted the same structure, and the manufacturer chose a similar structure to model the cost-effectiveness of ustekinumab. 72

73 5.3 ERG s summary and critique of manufacturer s submitted economic evaluation The analysis conducted by the manufacturer combines clinical and economic data to evaluate the costeffectiveness of ustekinumab for the treatment of patients with active and progressive PsA. The remainder of this section provides a summary and critique of the de novo model presented in the MS. A summary of the NICE reference checklist with the ERG s comments on whether the manufacturer s de-novo model has been judged to fulfil the NICE reference case is presented in Table 19. Table 19: NICE reference checklist Attribute Reference Case Included in MS Comment on whether de novo evaluation meets requirements of NICE reference case Comparator(s) Alternative therapies in the NHS, including those currently regarded as current best practice YES Meets the scope set by NICE, and includes alternative therapies recommended by NICE. However, sequencing of treatments was not fully addressed and for the evaluation of the anti-tnfα experienced sub-population it is clear that relevant comparators (i.e. second line anti-tnfα treatments) have not been included. Perspective costs NHS and PSS YES Perspective - benefits All health effects on individuals YES The utilities of patients were derived based on changes in both HAQ and PASI scores. Time horizon Sufficient to capture differences in costs and outcomes YES 52 year time horizon appears sufficient Synthesis of evidence on outcomes Systematic review YES Outcome measure QALYs YES Health states for QALY measurement Described using a standardised and validated instrument YES A published mapping algorithm was used to derive EQ-5D utility estimates. Additionally, patient level SF-36D data from the two PSUMMIT trails were mapped to EQ-5D and these were then used to derive an alternative algorithm estimating the relationship between HAQ, PASI and EQ-5D. Benefit valuation Time Trade Off or Standard Gamble YES Source of preference data Representative sample of the public YES Discount rate 3.5% on costs and health benefits YES Equity weighting No special weighting YES Sensitivity analysis Probabilistic sensitivity analysis YES The sensitivity analysis undertaken included probabilistic sensitivity analysis although a number of parameters were not assigned 73

74 distributions in the original submission Population The scope population for ustekinumab is people with active and progressive PsA whose disease has responded inadequately to previous DMARDs or anti-tnfα treatments. The inclusion criterion for all of the clinical trials is failure of two DMARDS. However, as discussed in Section 4.2.4, it is clear that this has not been strictly adhered to in either of the PSUMMIT trials, or the trials of the other biologics. The MS presented separate economic analyses for two distinct populations: Patients who are naïve to treatment with anti-tnfα (anti-tnfα naïve patients); Patients who are exposed to previous treatment with anti-tnfα (anti-tnfα experienced patients). Both populations are within the scope. The anti-tnfα naive population is in line with that used in previous submissions and is the licensed population for ustekinumab. 37, 38 However, the anti-tnfα experienced patients are a group of patients who have not been assessed in previous submissions for PsA. The MS described the unmet need of this group of patients, a group who have been exposed to, but achieved an inadequate response to anti-tnfα therapies. However no clear distinction is made between those who have failed the anti-tnfα class and therefore have a genuine unmet need; and those who have failed one or two anti-tnf treatments but have other options within the class still available to them. The baseline population characteristics for the anti-tnfα experienced population were based on a subgroup of patients from the PSUMMIT 2 trial (n=180). The populations have been fully discussed in Section 4.2.3, where the notable differences between the baseline characteristics have been presented Interventions and comparators This section discusses the use of ustekinumab for both the anti-tnfα naïve and anti-tnfα experienced populations as well as the comparators for each group Anti-TNFα naïve population 74

75 Ustekinumab is a subcutaneous injection available in a pre-filled syringe. For PsA, the recommended dose of ustekinumab is 45mg, with a 90mg dose being possible for patients weighing greater than 100kg. The comparators for the anti-tnfα naïve group include adalimumab, etanercept, golimumab, infliximab and conventional management strategies. Conventional management was not specifically defined, but reflects treatment with non-biologics. Table 20 provides the treatment dosing regimens for the biologic treatments considered in the model. Table 20: Treatment dosing regimens (adapted from Table 5, Section 4.1.1, pg. 20 in the MS) Mechanism of action Treatment Administration Dose/frequency TNFα inhibitors Adalimumab Subcutaneous 40mg every other week Etanercept Subcutaneous 25mg twice weekly or 50mg once weekly Golimumab Subcutaneous 50mg every month or 100mg every month for patients weighing >100kg who do not achieve adequate clinical response after 3 or 4 50mg doses. Infliximab Intravenous 5mg/kg over 1-2 hours, 5mg/kg every 8 weeks thereafter IL-12/IL-23 inhibitor Ustekinumab Subcutaneous 45mg dose at 2 week 0, 4, 16, then every 12 weeks thereafter. A 90mg dose may be used for patients >100kg. All of the comparator treatments were evaluated in RCTs, which were considered comparable to the ustekinumab trials for the anti-tnfα populations. These trials have been discussed in more detail in Section Anti-TNFα experienced population For the anti-tnfα experienced population, the only comparator included in the MS was conventional management strategies. As stated previously, conventional management was not specifically defined, but represents treatment with non-biologics. It is therefore implicitly assumed, within the model, that these experienced patients have exhausted anti-tnf treatment options and have no choice but to revert to non-biologic treatment options (i.e. DMARDs). 75

76 Currently NICE recommends that once patients meet the criteria for prescribing one of the anti-tnfα biologics (i.e. failure of two DMARD treatments), treatment with a biologic should commence. This population represents the anti-tnfα naïve population. Due to similarities in effectiveness NICE recommend that the prescriber commence treatment with the cheapest option first. Treatment should be discontinued if inadequate PsARC response is achieved at 12 weeks. Currently in UK routine practice, on failure of this first treatment, patients move onto a second anti-tnfα and response is judged in the same manner. This sequential approach to prescribing continues until the clinician judges that response is unlikely, the patient discontinues treatment due to an adverse event, or the best possible response for the patient has been achieved. The ERG s clinical expert confirmed that this sequential approach is reflective of the current situation in UK clinical practice. The ERG feels that to compare ustekinumab with only conventional management for the anti-tnfα experienced population is limited and potentially misleading. In only comparing ustekinumab to conventional management, the manufacturer s analysis represents the specific situation in which all possible anti-tnfα treatments have been exhausted, with ustekinumab being used as the final treatment option as an alternative to conventional management. In practice it is expected that up to three biologics are given to each patient; the data from the PSUMMIT 2 experienced population represents a mixed population - some of whom have failed only one anti-tnfα - unrepresentative of a population that have failed all anti-tnf options. The ERG feels that the analysis presented has severe limitations. To consider ustekinumab as a second or third treatment option, anti-tnfαs which have not been failed need to be considered as comparators. To consider ustekinumab as end of line treatment, all (or multiple) anti-tnf options need to have been trialled and failed. What little evidence there is on sequencing suggests that when anti-tnf treatments are used as a second-line treatment there is a degradation of efficacy, with this degradation likely to increase as a patient receives a 3 rd, a 4 th and so on, treatment. 1, There is no evidence to support that this same degradation of effect would not occur with ustekinumab. Thus derivation of efficacy data from a mixed population who are receiving ustekinumab 2 nd, 3 rd and 4 th line, is unlikely to provide appropriate estimates for the effectiveness of ustekinumab as an end of line treatment option. It should be noted that due to a lack of robust evidence and the inclusion of only treatment naïve patients, previous submissions did not include any sequencing of treatments within their base-case models. 23, 37, 38 However, Rodgers et al did attempt a secondary analysis where sequencing of treatment was permitted. 34 These issues will be discussed and explored further in Section

77 Conventional management therapies The MS does not describe in detail what constitutes conventional management therapies, apart from that they can include DMARD therapy and will not include anti-tnfα therapy. The ERG s clinical expert confirmed that conventional management therapies usually consist of DMARD therapy, which would primarily include MTX, sulfasalazine and leflunomide Model structure The model structure presented in the MS is the same for both the anti-tnfα naïve and the anti-tnfα experienced patients. The economic model is composed of two elements. The model begins with an initial short-term decision tree. This can be seen in Figure 7. 77

78 Figure 7: Decision tree structure (Figure 20, Section 7.2.2, pg. 97 in the MS) Here a patient s initial response to the biologic treatment is assessed. Following on from this, patients then enter a longer-term Markov model. This can be seen in Figure 8. Figure 8: Markov model structure (Figure 31, Section 7.2.2, pg. 98 in the MS) 78

79 Patients initially enter the model within the decision tree where they are assigned to either receive a biologic treatment or conventional management, as can be seen in Figure 7. In the base case the initial response to treatment is assessed using PsARC: patients with an adequate PsARC response continue on biologic treatment and patients with an inadequate PsARC response withdraw to conventional management strategies. Other feasible response criteria are assessed in sensitivity analyses. These include additional scenarios where both PASI and PsARC are used to determine a patient s initial response and where either PASI or PsARC can be used to determine a patient s initial response (see section 5.3.9). The distribution of patients at the end nodes of the decision tree, that is the distribution of patients within the biologic treatment, conventional management strategies and dead health states, informs the initial distribution of patients within the longer term Markov model. A graphical depiction of the Markov model is presented in Figure 2. Within the Markov model, patients continue through the model by continuing to respond to treatment, withdrawing from treatment or until death. 79