Costs and Benefits of Antithrombotic Therapy in Atrial Fibrillation in England: An Economic Analysis based on GRASP-AF

Costs and Benefits of Antithrombotic Therapy in Atrial Fibrillation in England: An Economic Analysis based on GRASP-AF Marion Kerr Insight Health Economics for NHS

We would like to acknowledge PRIMIS who helped develop the GRASP-AF tool in partnership with NHS IQ and who freely provided the software tools to run GRASP-AF and upload data. We would also like to acknowledge the practices that have voluntarily uploaded data to the CHART Online database, without whom this report would not have been possible. Acknowledgements 2

Executive Summary 1. We estimate that more than 900,000 people in England (1.74% of the population) have atrial fibrillation (AF). AF is associated with increased risk of stroke and heart failure, and with premature mortality, independent of other known risk factors. 2. We use an extract from the GRASP-AF dataset, covering 13% of the population of England, to estimate England-level prevalence and prescribing patterns. Based on GRASP-AF and evidence from clinical studies, we develop a Markov model to estimate the long-term impact of antithrombotic prescribing in AF on stroke and major bleed incidence, mortality, quality-adjusted life years (QALYs), and National Health Service (NHS) and social care costs. 3. We estimate that around 16,100 strokes a year, including 4,400 fatal strokes, are averted in people with AF who are taking warfarin (or warfarin plus aspirin). Fatal strokes are defined as those in which the patient dies within 30 days. However, disabling strokes also increase the risk of death over the longer term. Owing to the reduction in disabling strokes associated with warfarin, further deaths are averted. Total deaths in people with AF who are taking warfarin (or warfarin plus aspirin) are reduced by an estimated 4,800 over a one-year perspective. However, while these medications reduce stroke risk and mortality, they do increase the risk of bleeding. We estimate that there are 2,800 excess major bleeds a year in people with AF who are taking warfarin or warfarin plus aspirin. 4. Recent National Institute for Health and Care Excellence (NICE) guidelines recommend that the CHA 2 DS 2 -VASc algorithm should be used to assess stroke risk in people with AF, and that anticoagulants should be offered to those with a CHA 2 DS 2 -VASc score of 2 or more, taking bleeding risk into account. However, in our dataset only around half of those who meet the prescribing criteria (53.60%) are taking anticoagulants. For 8.74%, it is recorded that anticoagulants are contraindicated or declined. It is likely therefore that there are substantial numbers of people with AF who would benefit from anticoagulation therapy but are not receiving it. 5. Our model suggests that if warfarin were prescribed for all those in England with a CHA 2 DS 2 -VASc score of 2 or more (apart from those for whom the record states that anticoagulants are contraindicated or declined) a further 11,600 AF-related strokes, including 3,200 fatal strokes, would be averted each year. There would be an additional 1,500 excess bleeds, and the total number of deaths in people with AF, over a one-year perspective, would be reduced by 3,500. 6. Our analysis suggests that warfarin prescribing for people with AF is associated with an overall reduction in the NHS cost of AF-related care, as savings from averted strokes exceed the combined cost of treatment and associated major bleeds. We estimate that, over a lifetime, 0.30 strokes are averted, 0.79 QALYs are gained and net savings of more that 2,400 are generated for the NHS per person with AF treated with warfarin. We estimate that if everyone in England with a CHA 2 DS 2 -VASc score of 2 or more (apart from those for whom the record states that anticoagulants are contraindicated or declined) received warfarin, the NHS would save approximately 124 million a year, relative to current patterns of care. There are further estimated savings to social care. 7. While the focus in this paper is on NHS and social care costs, strokes associated with AF will also entail costs for individuals and their carers, and for society through losses in employment and productivity, and tax and benefit effects. If these effects were considered, it is likely that the overall savings from averted strokes would be higher than those estimated here. Executive Summary 3

Introduction 1. The prevalence of atrial fibrillation (AF) has been estimated at 1-2% of the population in the developed world. 1,2,3,4 AF is associated with increased risk of stroke and heart failure, and with premature mortality, independent of other known risk factors. 5,6 2. Current management of AF is focused on rate/rhythm control and reduction of stroke risk, primarily through anticoagulation therapy. In the English NHS the Quality and Outcomes Framework (QOF) has, since 2006-07, provided financial incentives to general practitioners for the formation of AF registers, and for treatment of the condition with anticoagulation or antiplatelet therapy. The QOF data for 2012-13 indicate that 849,407 people in England had diagnosed AF, a population prevalence of 1.5%. 3. The National Institute for Health and Care Excellence (NICE) has recently issued guidelines recommending that the CHA 2 DS 2 -VASc algorithm should be used to assess stroke risk in people with AF, and that anticoagulants should be offered to those with a CHA 2 DS 2 -VASc score of 2 or more, taking bleeding risk into account. 7 Anticoagulation should also be considered for men with a CHA 2 DS 2 -VASc score of 1, again taking bleeding risk into account. The guidelines recommend that aspirin monotherapy should not be offered solely for stroke prevention to people with atrial fibrillation. 4. The CHA 2 DS 2 -VASc algorithm is a point system in which 2 points are assigned for a history of stroke or transient ischaemic attack (TIA), or age 75; and 1 point each is assigned for age 65 74 years, a history of hypertension, diabetes, recent cardiac failure, vascular disease and female sex. 5. In this paper we examine antithrombotic prescribing patterns for people with AF in England, based on data from 1,016 GP practices uploaded to the Guidance on Risk Assessment and Stroke Prevention for Atrial Fibrillation (GRASP-AF) dataset in 2011 and 2012. We estimate the impact of warfarin, and warfarin plus aspirin prescribing on stroke incidence and on NHS and social care costs, and the potential impact of prescribing warfarin to all those with a CHA 2 DS 2 -VASc score of 2 or more. In a supplementary appendix we also provide illustrative analysis of the impact on stroke incidence and costs of aspirin monotherapy. Methods 1. Details of AF status and care were obtained from GRASP-AF for 1,016 GP practices in England. Together, these practices provide primary care to 7.1 million people, approximately 13% of the population of England. Data were uploaded to GRASP-AF between November 2011 and July 2012. 2. The GRASP-AF dataset provides pseudonymised patient-level data on AF status, age, gender, CHA 2 DS 2 -VASc score, prior stroke status and antithrombotic prescribing for GP practices that choose to participate. 3. The age distribution of patients in practices included in the data extract was compared with that reported for the population of England in Office for National Statistics (ONS) mid-year data for 2012. 8 AF prevalence estimates from the GRASP-AF dataset were applied to the ONS population of England by 5-year age-band. The resulting AF prevalence estimates for England were compared with the prevalence recorded in QOF for 2012-13. 4. The distribution of patients with AF in England in 2012-13 by CHA 2 DS 2 -VASc score and prior stroke status was estimated by applying the GRASP-AF proportions to the 2012-13 AF prevalence estimates for England. The volumes of anticoagulant and antiplatelet prescribing were estimated by applying the GRASP-AF proportions by CHA 2 DS 2 -VASc score and prior stroke status. Introduction Methods 4

5. A Markov model was built to estimate the impact of warfarin monotherapy, and warfarin plus aspirin combination therapy on stroke incidence, mortality, QALYs and NHS and social care costs. The model was also used to estimate the impacts of aspirin monotherapy. However, there is uncertainty regarding the efficacy of aspirin in reducing stroke risk in AF. In a recent meta-analysis, the estimate for the impact of aspirin did not reach statistical significance at the 95% level. 9 For this reason, the results of the aspirin monotherapy analysis are provided in a supplementary appendix to this paper. Care should be taken in interpreting the aspirin results owing to the uncertainty surrounding the meta-analysis estimate. 6. Markov models are commonly used in health economic evaluations, in particular where it is necessary to extrapolate beyond study results, and where diseases present with recurrent events. 10 A Markov model consists of a finite set of health states in which an individual can be found. The model runs over a series of cycles, and patients move between states or stay in the same state over time, with movements defined by transition probabilities. Costs and QALYs are assigned to each health state and earned at the end of each cycle. In this case the Markov states are AF without stroke, AF plus acute ischaemic or haemorrhagic stroke (each sub-divided into disabling, moderate and mild stroke), AF plus prior stroke (again sub-divided into disabling, moderate and mild), major non-intracerebral intracranial bleed with or without prior stroke, major extracranial bleed with or without prior stroke, and dead. The AF plus acute stroke states and bleed states are transitional; individuals can stay in them for only one cycle (though return to these states is possible if repeated strokes or bleeds are experienced). The cycle length is three months. The structure of the Markov model, indicating the states and possible transitions, is shown in Figure 1. Figure 1: Markov model structure AF plus acute ischaemic/ haemorrhagic stroke Mild Moderate Disabling AF without stroke Major bleed in AF without stroke Mild AF plus prior stroke Disabling Moderate Major bleed in AF plus prior stroke Methods Death 5

7. In cost effectiveness analysis it is customary to express costs and effects over the lifetime of a representative patient. Indeed, for the estimation of QALY effects it is necessary to take such a long-term perspective. However, decision-makers often wish also to understand costs over shorter timescales. In this paper we therefore present long-run ( lifetime ) cost and QALY impacts, and also one-year cost impacts. The representative patient is taken to be 78 years old at model outset, based on the median age in GRASP-AF. 8. The baseline risk of ischaemic stroke without antithrombotic treatment was estimated by CHA 2 DS 2 -VASc score for the GRASP-AF population, using risks estimated in a recent Swedish study. 11 9. There are a number of challenges in estimating the baseline risk of major bleeds, and the effect of treatment on this risk. Published estimates of major bleed rates in patients taking oral vitamin K antagonists such as warfarin range from 1.3 to 7.2% a year. 12 These differences may reflect variation in a number of factors, including study populations, stability of international normalised ratio (INR) values, methodology, definition and classification. Relatively few studies have examined bleed rates in patients who are not receiving antithrombotics. Some studies have included intracerebral haemorrhage with stroke, rather than with bleeds, some have included all intracranial bleeds with stroke, while others have classified all intracranial bleeds as major bleeds rather than strokes. In their 2007 meta-analysis, Hart et al. did not estimate the impact of warfarin or aspirin versus placebo or control on the relative or absolute risk of intracranial bleeds, owing to the large proportion of empty cells in individual trial data. 9 10. However, in spite of these difficulties we felt it was important to try to model intracerebral haemorrhages (taken to be equivalent to haemorrhagic strokes) discretely; while such haemorrhages are far less frequent than ischaemic strokes, mortality rates are very high (more than 50% in many studies). 13,14 We estimated the rate of all intracranial bleeds in patients with AF on warfarin at 0.73% per year, based on a weighted average from 16 randomised trials summarized by Lip et al. 12 Most of these studies did not distinguish between intracerebral and other intracranial bleeds. The RE-LY trial (which was included in the 16) reported that 46% of intracranial bleeds were intracerebral. (The rate of intracranial bleeds in this trial was 0.76%). 14 Applying this proportion to the overall estimate derived from the 16 trials, we estimate the rate of intracerebral haemorrhage in warfarin at 0.34%, and the rate of other intracranial haemorrhage at 0.39%. 11. Of the 16 studies summarized by Lip at al., one provided an estimate of the rate of intracranial bleed in the absence of warfarin. 15 This study estimated a crude rate ratio of 1.96 (95% CI 1.25-3.03) for intracranial bleed in those taking warfarin, relative to the non-warfarin group. It should be noted that some of the non-warfarin group were taking aspirin. The study did not provide discrete data for the aspirin and non-antithrombotic sub-groups. A US study of 13,559 people with AF reported that the risk of intracerebral haemorrhage was 1.81 times higher in patients on warfarin than in those who were not on warfarin. 16 Applying these proportions to the estimates of bleed rates in warfarin set out in paragraph 10, we estimate the baseline risk of all intracranial bleeds without warfarin at 0.37%, and the baseline risk of intracerebral bleed at 0.19%. 12. No studies were identified which provided discrete intracerebral haemorrhage risk estimates by CHA 2 DS 2 -VASc score or prior stroke status, though some studies have suggested that the absolute increase in risk for intracranial haemorrhage due to warfarin therapy remains fairly stable across thromboembolic risk categories. 16 GRASP-AF does not classify patients using HAS-BLED or other bleed risk measures. Methods 13. Hart et al. estimated in their meta-analysis that warfarin was associated with an absolute increase in the risk of major extracranial bleeds of 0.3% per year, relative to control or placebo, and this figure was used in the model. 9 14. The impact of warfarin on the combined endpoint of stroke and intracranial bleed risk was estimated by Hart et al. in their 2007 meta-analysis findings at minus 64%. 9 Adjusting that figure for the increase in risk of intracranial bleeds, we estimate the reduction in ischaemic stroke risk to be 66.88% for patients with prior stroke and 69.70% for those without prior stroke. 6

15. We have assumed that ischaemic stroke reduction in the case of treatment with both warfarin and antiplatelets is the same as for warfarin alone. The hazard ratio for major bleeds in combined therapy relative to warfarin monotherapy is taken to be 1.53. 18 It should be noted that this approach may underestimate the risk of intracranial haemorrhage when both warfarin and aspirin are taken. Some studies have suggested that the relative risk of intracranial haemorrhage in warfarin and aspirin combination therapy, relative to warfarin monotherapy, may be up to 2.4. 19 16. We followed the practice of Ali et al. in defining minor bleeds as those that do not require medical intervention. 20 We did not model any cost or outcome impacts for these events. 17. Thirty-day ischaemic stroke mortality rates in AF for those without antithrombotic therapy, for patients on warfarin, and for those on aspirin were estimated from a US study. 21 Thirty-day mortality rates after haemorrhagic stroke and other intracranial bleeds were taken from the RE-LY trial. 14 Hazard ratios for subsequent death after non-fatal ischaemic stroke, haemorrhagic stroke and other major bleeds were taken from the Atrial Fibrillation Clopidogrel Trial with Irbesartan for Prevention of Vascular Events (ACTIVE)-W Trial. 22 Mortality risk in AF in the absence of stroke or major bleed, was estimated from ONS Life Tables by gender and year of age. The overall mortality rates predicted in the model for people on warfarin were validated by comparing with meta-analysis findings on the absolute reduction in all-death risk. 9 18. The NHS costs of the first 90 days of care for mild, moderate, disabling and fatal strokes were estimated from a recent UK study on stroke in AF. 23 The impact of stroke in AF on social care costs for mild, moderate, and disabling strokes was estimated from the same report. The study also provided estimates of the difference between ongoing post-stroke (after 90 days) and pre-stroke NHS costs. However, these estimates were not statistically significant at the 95% level. We therefore excluded them from our base-case analysis, but have included them in sensitivity analysis. 19. The proportion of disabling strokes in AF, by treatment modality, was taken from Hart et al. 9 Of the remainder, the distribution of moderate and mild strokes in patients with AF was taken from the US Framingham study. 24 20. The mean costs of major extracranial bleeds and non-intracerebral intracranial bleeds were estimated from NHS Reference Costs. 25 The weighted mean cost of admission episodes grouped to HRGs FZ38G-P, Gastrointestinal Bleed ( 1,092) was used to estimate the cost of extracranial bleeds. The weighted mean cost of admission episodes grouped to HRGs AA23C-G, Haemorrhagic Cerebrovascular Disorders ( 2,589) was used to estimate the cost of non-intracerebral intracranial bleeds. (While these HRGs are labeled as cerebrovascular, ICD 10 codes I60 and I62.0, covering subarachnoid and subdural haemorrhages respectively, are also grouped to these HRGs). 21. The costs of medications were taken from the British National Formulary (BNF). 26 The mean daily dose of warfarin was assumed to be 4mg. 27 Warfarin tablets are available in 0.5mg, 1 mg, 3mg, and 5mg doses. The cost of the medication varies substantially, depending on the combination of tablets used to administer a given dose, as shown in Table 1. Table 1: Annual cost of warfarin medication, by dose and tablet combination Source: BNF 64 26 Methods Dose and cost Tablet combination 4mg daily dose 8 x 0.5mg 4 x 1mg 1 x 1mg + 1 x 3mg 2 x 0.5mg + 1 x 3mg Annual cost 174.16 44.84 22.42 54.75 6mg daily dose 12 x 0.5mg 2 x 3mg 1 x1mg + 1 x 5mg 2 x 0.5mg + 1 x 5mg Annual cost 261.24 22.42 23.20 55.53 7

22. It is assumed here that 4mg is provided as 1 x 3mg tablet plus 2 x 0.5mg tablets, to allow for dose adjustment. We have also set out the cost of a 6mg dose, for a variety of tablet combinations (Table 1). If 6mg is delivered as 1 x 5mg and 2 x 0.5mg tablets, the cost is similar to that for 4mg delivered as 1 x 3mg tablet plus 2 x 0.5mg tablets ( 55.53, as compared with 54.75). We did not therefore explore in the model separate costs for provision of a 6mg daily dose. The mean daily aspirin dosage was taken to be 75mg, and the annual cost was estimated at 6.19. 23. The annual cost of warfarin management, including anticoagulation testing for dose adjustment, has been estimated in studies at 73.99-115.60 (after conversion to 2012-13 prices). 20,28 The mean value ( 94.80) was used in our base case analysis. It is reported, however, that there is considerable variation in the cost of warfarin management. In sensitivity analysis we therefore used a range of values for warfarin monitoring, as recommended in a recent NICE guideline. 29 24. Quality of life in patients with atrial fibrillation and in those who have had strokes was estimated using the EQ-5D standardised measure of health status. Measures were drawn from studies that used weights derived from UK population preferences. 30,31,32 25. Cost and QALY impacts were estimated for current prescribing patterns for patients with CHA 2 DS 2 -VASc score 2, based on GRASP-AF data. Cost and QALY impacts were also modelled for prescribing of warfarin for all those with CHA 2 DS 2 -VASc score 2 for whom anticoagulants are not contra-indicated or declined. 26. All costs are expressed in 2012-13 prices. Where necessary, costs were updated using the Hospital and Community Health Services Pay and Prices index. Future costs and QALYs are discounted at 3.5% a year. 27. A summary of parameters used in the Markov model is provided in Appendix 1, Tables A1.1 and A1.2. Methods 8

Results Prevalence and prescribing 1. The GRASP-AF dataset recorded 128,293 people with AF in participating practices, a prevalence of 1.81%. The age distribution of patients in these practices is broadly similar to that reported for England in ONS, although a somewhat larger proportion of the GRASP-AF population than the ONS population is aged 65+ (17.56%, as compared with 16.44% in ONS (Figure 2)). Figure 2: Age distribution by quinary age-band, GRASP-AF, and ONS 2012 Percentage of population 8% 7% 6% 5% 4% 3% 2% 1% GRASP-AF ONS 0% 0-4 5-9 10-1 4 15-1 9 20-2 4 25-2 9 30-3 4 35-3 9 40-4 4 45-4 9 50-5 4 55-5 9 60-6 4 65-6 9 70-7 4 75-7 9 80-8 4 85-8 9 90 + Age 2. The prevalence of AF in GRASP-AF ranges from 0.01% in those aged 0-29 years to 15.06% in those aged 80 years. 3. Adjusting for the difference in age distribution, the prevalence of AF in England based on GRASP-AF is estimated at 1.74% in 2012 (929,856 people), as compared with 1.52 % (849,407 people) recorded in QOF in 2012-13. QOF data exclude patients who are considered to have AF resolved. If this group were removed from GRASP-AF the estimated age-adjusted prevalence for England would be 1.51%. Patients with AF resolved are included in the analyses reported in this paper. Results 4. Of AF patients in GRASP-AF, 18.14% had a prior stroke, a further 66.00% had a CHA 2 DS 2 -VASc score 2 without prior stroke, 7.64% were men with CHA 2 DS 2 -VASc score = 1, 2.40% were women with CHA 2 DS 2 -VASc score = 1, and 5.81% had CHA 2 DS 2 -VASc score = 0. 5. Of those with CHA 2 DS 2 -VASc score 2, 53.60% were prescribed anticoagulants. For 8.74%, the record stated that anticoagulants were contra-indicated or had been declined (Table 2). Of those in this group receiving anticoagulants, 99.00% were receiving warfarin. 9

Table 2: Prescribing of anticoagulants for AF patients in GRASP-AF CHA 2 DS 2 -VASc 2 Men with CHA 2 DS 2 -VASc = 1 Patients 107,949 9,810 Anticoagulants Warfarin 53.07% 42.33% Acenocoumarol 0.32% 0.24% Dabigatran etexilate 0.12% 0.19% Phenindione 0.09% 0.06% Rivaroxaban 0.00% 0.01% Apixaban 0.00% 0.00% Anticoagulants contra-indicated 5.44% 2.13% Anticoagulants declined 3.30% 1.81% No anticoagulants and anticoagulants not contra-indicated or declined 37.66% 53.21% 6. Anticoagulant and antiplatelet combination therapy was prescribed for 7.78% of those with CHA 2 DS 2 -VASc score 2, and a further 35.63% were on antiplatelet monotherapy. No antithrombotics were prescribed for 10.76% of the group (Table 3). Table 3: Prescribing of anticoagulants and antiplatelets, GRASP-AF CHA 2 DS 2 -VASc 2 Men with CHA 2 DS 2 -VASc = 1 Anticoagulant only 45.82% 36.80% Anticoagulant and antiplatelet 7.78% 6.04% Antiplatelet only 35.63% 36.22% Neither anticoagulant nor antiplatelet 10.76% 20.94% 7. If it is assumed that the prevalence of AF and the distribution of CHA 2 DS 2 -VASc scores in England is the same as the age-adjusted estimates from GRASP-AF, it is estimated that there were 783,592 people in England in 2012-13 with diagnosed AF and a CHA 2 DS 2 -VASc score 2, and 71,519 men with a CHA 2 DS 2 -VASc score = 1. Results 10

Estimated baseline stroke risk 8. Stroke risk in the absence of antithrombotic therapy was estimated by applying stroke risks for each CHA 2 DS 2 -VASc score (2-9) observed in a recent Swedish study 11 to the GRASP-AF population, as shown in Table 4. Table 4: Stroke risk by CHA 2 DS 2 -VASc score (Source: Friberg et al. 11 ), and distribution of patients in GRASP-AF with and without prior stroke Distribution of people with AF in GRASP-AF (% of those with CHA 2 DS 2 -VASc score 2) CHA 2 DS 2 -VASc score Prior stroke No prior stroke Stroke risk (Friberg et al.) 2 374 (0.35%) 20,051 (18.57%) 2.5 3 1,264 (1.17%) 27,191 (25.19%) 3.7 4 3,550 (3.29% 24,765 (22.95%) 5.5 5 6,578 (6.09%) 10,008 (9.27%) 8.4 6 7,357 (6.82%) 2,371 (2.20%) 11.4 7 3,206 (2.97%) 286 (0.26%) 13.1 8 829 (0.77%) 0 (0.00%) 12.6 9 119 (0.11%) 0 (0.00%) 14.4 9. Given the distribution of patients by CHA 2 DS 2 -VASc score in GRASP-AF, we estimate the overall annual ischaemic stroke risk for the population with CHA 2 DS 2 -VASc score 2 at 4.75% for primary stroke and 9.38% for secondary stroke, in the absence of antithrombotic therapy. This compares with meta-analysis findings for the general AF population (not differentiated by CHA 2 DS 2 -VASc score) of 4.1% for primary stroke and 13% for secondary stroke, again in the absence of antithrombotic therapy. 9 It should be noted, however, that the meta-analysis includes haemorrhagic strokes, whereas the Swedish study does not. 10. If it is assumed that the prevalence of AF and the distribution of CHA 2 DS 2 -VASc scores in England is the same as the age-adjusted estimates from GRASP-AF, the Markov model estimates that, without any antithrombotic therapy, there would have been 45,608 strokes (ischaemic and haemorrhagic) in people with AF with CHA 2 DS 2 -VASc score 2 in England in 2012-13, equivalent to 5.8 for every 100 patients. Results Stroke, QALY and cost impacts per person treated, by treatment modality 11. It is customary in health economic analysis to estimate cost and benefit impacts relative to the next best treatment option. In the case of warfarin in AF, there is uncertainty as to whether the next best option is no treatment or treatment with aspirin. As indicated above, the meta-analysis estimate of the impact of aspirin on stroke risk did not reach statistical significance at the 95% level. 9 Recent European Society of Cardiology (ESC) guidelines state that The evidence for effective stroke prevention with aspirin in AF is weak, with a potential for harm. 4 As indicated above, NICE has recommended that aspirin monotherapy should not be offered solely for stroke prevention to people with atrial fibrillation. 7 11

12. In view of this uncertainty, and the NICE recommendation that aspirin monotherapy should not be offered for stroke prevention in AF, we have taken no treatment as the next best option in this paper. In the Appendix we provide illustrative figures for the potential impact of aspirin on stroke. It should be noted, however, that these estimates are based on the (non-significant) estimates from meta-analysis. We cannot be confident, at the 95% level, that aspirin reduces the risk of stroke in AF. 13. The Markov model estimates that, over a lifetime perspective, 0.30 strokes (combined ischaemic and haemorrhagic) are averted and 0.79 QALYs gained per person treated with warfarin, relative to no antithrombotic treatment. Of averted strokes, more than a quarter (0.08 per person treated) are fatal strokes. There are 0.05 excess bleeds (extracranial and non-intracerebral intracranial). NHS and social care costs related to AF are reduced over both a one-year and lifetime perspective, owing to reduced stroke incidence (Table 5). More details of the NHS cost impact are shown in Table 6. Table 5: Estimated one-year and lifetime stroke, major bleed, QALY and cost impacts per person treated with warfarin monotherapy, relative to no antithrombotic therapy Strokes Bleeds QALYs NHS costs Social care costs One-year impact -0.039 0.006 0.007-412 - 94 Lifetime impact -0.298 0.048 0.794-2,408-1,379 Table 6: Estimated one-year and lifetime NHS cost impacts of treatment, major bleeds and averted strokes per person treated with warfarin monotherapy, relative to no antithrombotic therapy Treatment cost Excess major bleeds Averted strokes One-year impact 139 8-559 Lifetime impact 1,129 66-3,603 14. Fatal strokes are defined as those in which the patient dies within 30 days. However, disabling strokes also increase the risk of death over the longer term. Owing to the reduction in disabling strokes associated with warfarin, further deaths are averted. The model estimates that absolute all-cause one-year mortality is 1.17% lower with warfarin than with no treatment. This estimate may be compared with meta-analysis findings that warfarin is associated with a 1.6% reduction in all-cause annual mortality. 9 15. For those treated with warfarin and aspirin combination therapy, stroke reduction is lower than for those on warfarin monotherapy, owing to the increased risk of haemorrhagic stroke. The rate of other major bleeds is also higher, and NHS net savings are lower than on warfarin monotherapy (Table 7). More details of the NHS cost impact are shown in Table 8. Results 12

Table 7: Estimated one-year and lifetime stroke, major bleed, QALY and cost impacts per person treated with warfarin and aspirin combination therapy, relative to no antithrombotic therapy Strokes Bleeds QALYs NHS costs Social care costs One-year impact -0.037 0.011 0.006-379 - 92 Lifetime impact -0.284 0.081 0.754-2,181-1,305 Table 8: Estimated one-year and lifetime NHS cost impacts of treatment, major bleeds and averted strokes per person treated with warfarin and aspirin combination therapy, relative to no antithrombotic therapy Treatment cost Excess major bleeds Averted strokes One-year impact 145 12-536 Lifetime impact 1,167 100-3,448 Estimated clinical and cost impacts of GRASP-AF prescribing patterns 16. If it is assumed that the prevalence of AF and the distribution of CHA 2 DS 2 -VASc scores in England are the same as the age-adjusted estimates from GRASP-AF, and prescribing patterns are also as in GRASP-AF, it is estimated that 415,834 people with CHA 2 DS 2 -VASc score 2 were on warfarin in 2012-13. Of this number, it is estimated that 356,086 were on warfarin monotherapy, and 59,748 were taking both warfarin and aspirin. For a monotherapy group of this size, it is estimated that 13,906 strokes, including 3,801 fatal strokes, would be averted through warfarin in a year, and 106,231 strokes, including 29,449 fatal strokes, over a lifetime perspective. It is estimated that the number of deaths from all causes would be reduced by 4,185 over a one-year perspective. The model predicts that there would be an additional 2,183 non-intracerebral bleeds in a year and 16,935 over a lifetime. Annual net savings to the NHS from averted strokes are estimated at 147 million over one year, and 858 million over a lifetime perspective. There are additional estimated savings to social care (Table 9). More details of the NHS cost impact are shown in Table 10. Impacts for the warfarin and aspirin combination therapy cohort are shown in Tables 11 and 12. Table 9: Estimated one-year and lifetime stroke, major bleed, QALY and cost impacts for a cohort of 356,086 people treated with warfarin monotherapy, relative to no antithrombotic therapy Strokes Bleeds QALYs NHS costs Social care costs One-year impact -13,906 2,183 2,330-146,747,063-33,450,628 Lifetime impact -106,231 16,935 282,892-857,627,384-490,997,377 Results 13

Table 10: Estimated one-year and lifetime NHS cost impacts of treatment, major bleeds and averted strokes per person for a cohort of 356,086 people treated with warfarin monotherapy, relative to no antithrombotic therapy Treatment cost Excess major bleeds Averted strokes One-year impact 49,581,481 2,891,212-199,219,755 Lifetime impact 401,849,653 23,432,788-1,282,909,825 Table 11: Estimated one-year and lifetime stroke, major bleed, QALY and cost impacts for a cohort of 59,748 people treated with warfarin and aspirin combined therapy, relative to no antithrombotic therapy Strokes Bleeds QALYs NHS costs Social care costs One-year impact -2,232 629 369-22,645,716-5,506,964 Lifetime impact -16,939 4,841 45,032-130,331,196-77,984,729 Table 12: Estimated one-year and lifetime NHS cost impacts of treatment, major bleeds and averted strokes for a cohort of 59,748 people treated with warfarin and aspirin combined therapy, relative to no antithrombotic therapy Treatment cost Excess major bleeds Averted strokes One-year impact 8,659,694 879,786-32,185,196 Lifetime impact 69,721,057 7,083,345-207,135,598 17. Combining the results for warfarin monotherapy and warfarin and aspirin combined therapy, we estimate that 16,137 strokes a year, including 4,396 fatal strokes, are averted in people with AF, on the assumption that current prescribing patterns across England are as observed in GRASP-AF. The number of excess major bleeds is estimated at 2,812 a year. Taking into account also the increased long-term risk of death associated with disabling strokes, total deaths in people with AF who are taking warfarin (or warfarin plus aspirin) are reduced by an estimated 4,820 over a one-year perspective. Estimated clinical and cost impacts of NICE-recommended prescribing Results 18. If it is assumed that the prevalence of AF and the distribution of CHA 2 DS 2 -VASc scores in England are the same as the age-adjusted estimates from GRASP-AF, and prescribing patterns are also as in GRASP-AF, it is estimated that there were 710,938 people with CHA 2 DS 2 -VASc score 2 (excluding those on other anticoagulants and those for whom anticoagulants are contra-indicated or declined). If all of this group were taking warfarin, it is estimated that 27,763 strokes, including 7,589 fatal strokes, would have been averted in 2012-13, and NHS costs associated with AF would have been reduced by 293 million over one year and 1.7 billion over a lifetime perspective, relative to no anti-thrombotic therapy (Tables 13 and 14). This represents an additional 11,626 strokes averted, and 124 million one-year cost reduction ( 724 million over a lifetime perspective), relative to estimated current prescribing patterns. 14

Table 13: Estimated one-year and lifetime stroke, major bleed, QALY and cost impacts for a cohort of 710,938 people treated with warfarin monotherapy, relative to no antithrombotic therapy Strokes Bleeds QALYs NHS costs Social care costs One-year impact -27,763 4,359 4,651-292,985,574-66,785,332 Lifetime impact -212,095 33,811 564,803-1,712,282,662-980,293,204 Table 14: Estimated one-year and lifetime NHS cost impacts of treatment, major bleeds and averted strokes per person for a cohort of 710,938 people treated with warfarin monotherapy, relative to no antithrombotic therapy Treatment cost Excess major bleeds Averted strokes One-year impact 98,991,138 5,772,404-397,749,115 Lifetime impact 802,306,697 46,784,371-2,561,373,729 Sensitivity analyses 19. It is known that the cost of INR management for people taking warfarin is subject to considerable local variation. The Evidence Review Group for a 2012 NICE technology appraisal proposed that three values for annual INR monitoring should be used in sensitivity analysis ( 115.14, 241.54 and 279.36). 29 In sensitivity analyses 1-3 we have used these values, updated to 2012-13 prices ( 117.83 in sensitivity analysis 1, 247.18 in sensitivity analysis 2, and 285.89 in sensitivity analysis 3). 20. In all three scenarios, the estimated net cost impact for the NHS is negative, meaning that estimated savings from averted strokes exceed the cost of AF- associated treatment and monitoring. However, net savings are lower than in the base case (Tables 15 and 16). With the highest estimated level of INR monitoring costs, net annual savings to the NHS for care associated with AF are 43% lower than the base case estimate over a one-year perspective, and 58% lower over a lifetime. Net social care costs are unaffected. Table 15: Sensitivity analyses 1-3: Estimated one-year and lifetime NHS cost impacts per person treated with warfarin monotherapy, relative to no antithrombotic therapy, with annual INR monitoring costs of 115.14, 241.54 and 279.36 Results Net NHS cost impact Base case Sensitivity 1 Sensitivity 2 Sensitivity 3 One-year impact - 412-391 270-234 Lifetime impact - 2,408-2,241-1,297-1,015 15

Table 16: Sensitivity analyses 1-3: Estimated one-year and lifetime NHS cost impacts for a cohort of 710,938 people treated with warfarin monotherapy, relative to no antithrombotic therapy, with annual INR monitoring costs of 115.14, 241.54 and 279.36 per person treated Net NHS cost impact Base case Sensitivity 1 Sensitivity 2 Sensitivity 3 One-year impact - 292,985,574-277,741,418-192,121,293-166,498,138 Lifetime impact - 1,712,282,662-1,592,879,823-922,243,388-721,545,000 21. Another key area of uncertainty is the marginal cost of ongoing NHS care for people with AF who have experienced strokes. A recent study reported that the mean annual cost of post-stroke care was higher than that of pre-stroke care for all levels of stroke severity in people with AF. 23 However, the differences were not significant at the 95% level. In the light of this uncertainty, we did not include ongoing NHS costs in our base case. In sensitivity analysis 4, we use the mean differences between pre- and post-stroke care annual costs from this study, updated to 2012-13 prices ( 413 for mild stroke, 894 for moderate stroke, 3,270 for disabling stroke). Using these estimates of post-stroke costs, the saving to the NHS for AF-associated care increases relative to base case by 26% over a one-year perspective, and 64% over a lifetime. Table 17: Sensitivity analysis 4: Estimated one-year and lifetime NHS cost impacts for warfarin monotherapy, relative to no antithrombotic therapy, including estimated post-stroke NHS costs Per person treated For 356,086 people For 710,938 people One-year impact - 519-184,678,126-368,716,250 Lifetime impact - 3,939-1,402,582,163-2,800,303,680 Results 16

Discussion 1. The estimates of AF-associated prescribing, expenditure and outcomes set out in this paper have been based on extrapolations from the GRASP-AF dataset, which provides details of AF status, age, gender, CHA 2 DS 2 -VASc score, prior stroke status and antithrombotic prescribing. The extract used for this report covered patients in more than a thousand GP practices. These practices together provide care to approximately 13% of the population of England. England-level estimates of prevalence have been standardised to match the national age-distribution as recorded in ONS. GRASP-AF does not provide other demographic data such as ethnicity or deprivation, and it has therefore not been possible to standardise for these variables. 2. Estimates of the impact of prescribing on stroke incidence, QALYs and costs are estimated by means of a Markov model. The outputs of such a model are highly sensitive to the parameters used. For this reason we have aimed to set out details of parameters and sources transparently. These are summarised in the Methods section, and further detail is provided in Appendix 1. We have also carried out sensitivity analyses in areas where variable parameters are subject to particular uncertainty or variability. 3. Our analysis suggests that more than 900,000 people in England have AF (1.74%). Of this group, around 84% have a CHA 2 DS 2 -VASc score 2 (a population prevalence of 1.46%). We estimate that current levels of warfarin (or warfarin plus aspirin) treatment in this group are associated with prevention of around 16,100 strokes a year, including 4,400 fatal strokes. 4. The number of excess major bleeds associated with warfarin, or warfarin plus aspirin prescribing for this group is estimated at around 2,800 a year. (This excludes intracerebral bleeds, which are included in the stroke estimate). Taking into account also the increased risk of death associated with disabling stroke in those who have survived to 30-days, total deaths in people with AF are reduced by an estimated 4,800 over a one-year perspective. The absolute reduction in the one-year death rate for those on warfarin monotherapy is estimated at 1.17%. The reduction in all-cause annual mortality associated with warfarin has been estimated at 1.6% in meta-analysis. 9 It is possible therefore that our model under-estimates the reduction in deaths associated with warfarin prescribing. 5.Over a lifetime perspective, our model estimates that 0.30 strokes are averted and 0.79 QALYs gained per person treated with warfarin. 6. Recent NICE guidelines recommend that patients with a CHA 2 DS 2 -VASc score 2 should be offered anticoagulation therapy. However, only just over half (53.60%) of people with a CHA 2 DS 2 -VASc score 2 in our dataset are prescribed warfarin or other anticoagulants. For a further 8.74% of this group, it is recorded that anticoagulants are contraindicated or declined. There may therefore be substantial numbers of people with AF who would benefit from anticoagulation therapy but are not receiving it. Our model suggests that if warfarin were prescribed for all those with a CHA 2 DS 2 -VASc score of 2, apart from those for whom it is contraindicated or declined, approximately 11,600 further AF-related strokes, including 3,200 fatal strokes, would be averted each year in England. There would be an additional 1,500 excess bleeds, and the total number of deaths in people with AF, over a one-year perspective, would be reduced by a further 3,500. Discussion 7. Our analysis suggests that warfarin prescribing for people with AF is associated with a net reduction in the cost to the NHS of AF-related care. This reduction arises because the saving through averted strokes exceeds the costs of treatment and of associated major bleeds. The one-year saving to the NHS from warfarin (or warfarin plus aspirin) prescribing at the levels shown in our dataset for those who have AF in England is estimated at 169 million, and the lifetime saving at 988 million. If warfarin were prescribed for all those with a CHA 2 DS 2 -VASc score 2, apart from those for whom it is contraindicated or declined, the estimated saving to the NHS on AF-related care is 293 million over one year, and 1.7 billion over a lifetime perspective. There are further estimated savings to social care. 17

8. There is uncertainty regarding the efficacy of aspirin on stroke prevention in AF. Meta-analysis findings report a non-significant 19% reduction in stroke incidence (compared with a 64% reduction with warfarin). 9 NICE has recommended that aspirin monotherapy should not be offered solely for stroke prevention to people with atrial fibrillation. In view of this recommendation, and the statistical uncertainty, we have presented analysis of aspirin in a separate appendix. Based on a 19% reduction in stroke incidence, we estimate that over a one-year perspective aspirin would be associated with a reduction in stroke incidence of 0.015 per person treated, compared with 0.039 for warfarin. Estimated overall NHS costs rise with aspirin treatment, owing to an increase in the proportion of strokes that are haemorrhagic, an increase in other major bleeds, and an increase in the proportion of disabling strokes. In economic parlance, warfarin dominates aspirin as a therapy for stroke prevention in AF. 9. The recent NICE guideline also recommends that anticoagulants should be considered for men with a CHA 2 DS 2 -VASc score of 1. There is little evidence on baseline stroke risk in this group, or on the impact of anticoagulation. We did not feel that the evidence base was sufficiently robust to support estimation of costs and benefits for this group. Further clinical studies are needed on the potential for stroke reduction through anticoagulation in this group. 10. New oral anticoagulants (dabigatran etexilate, rivaroxaban and apixaban) have recently been approved by a number of international bodies, including NICE in England and Wales, for stroke prevention in sub-groups of the AF population. In both cases the product cost is considerably higher than that of warfarin. The manufacturers of these drugs state that there is no requirement for regular anticoagulation testing for dose adjustment with these medications. 11. The GRASP-AF data extract used for our paper covered the period from November 2011 to July 2012, and the NICE recommendations on these new agents were published toward the end of this period. Not surprisingly, therefore, use of these agents in the period covered by our study was very low (approximately 0.22% of total anticoagulant use in people with CHA 2 DS 2 -VASc score 2). In this paper, therefore, we have not modelled the cost or impacts of prescribing of the new oral anticoagulants. It is likely, however, that in the future there will be a need for further study on the impact of these agents on stroke risk in the AF population in England, and associated costs. 12. There is variation in the cost of warfarin monitoring, and these costs have a substantial impact on overall net cost estimates. A recent NICE technology appraisal used three values for warfarin monitoring in sensitivity analyses. The value used in our base case is drawn from study evidence, and is lower than the values used by NICE. We have used the NICE values in our sensitivity analyses. For all of these values, we estimate that warfarin is associated with a reduction in the NHS cost of AF-associated care. 13. Given the variation in the cost of such monitoring around the country, the overall cost impact of warfarin prescribing will also vary from place to place. There may be potential in places where costs are particularly high to provide monitoring services more efficiently, thus increasing the overall cost effectiveness of treatment. 14. There is also a need for further study on the cost of warfarin monitoring in people with AF. Since the cost effectiveness of medications is estimated with reference to the next best option, understanding the true cost of warfarin is important not only for assessment of the costs and savings of warfarin itself, but is also an important factor in determining the cost effectiveness of alternative medications. Discussion 15. It is hoped that this paper will provide a foundation for local analysis and decision-making on antithrombotic prescribing in AF. 18

Appendix 1 Table A1. Transition probabilities No antithrombotics (baseline) Warfarin Warfarin and aspirin Aspirin (See Appendix) Ischaemic stroke (annual) Haemorrhagic stroke (annual) Primary 4.75%, Secondary 9.38% (Estimated by applying CHA 2 DS 2 -VASc risks (Friberg et al. 11 ), to CHA 2 DS 2 -VASc distribution in GRASP-AF) 0.19% Estimated by applying ratio of intracerebral haemorrhage in warfarin and non-warfarin patients in Singer et al 16. to estimated risk with warfarin shown in cell below. Primary 1.44%, Secondary 3.11% (Relative risk reduction: 64% (Hart et al. 9 ), adjusted for impact on haemorrhagic stroke, applied to baseline risks estimated above. 0.34% Overall rate of intracranial bleeds estimated by taking weighted average from 16 studies cited in Lip et al 12. (0.73%), proportion of these that are intracerebral estimated from Hart el al. 14 (46%) As for warfarin Primary 3.81%, Secondary 7.53% (Risk reduction: 19% non significant (Hart et al. 9 ) applied to baseline risks estimated above. Hazard ratio 1.53, relative to warfarin alone (Steinberg et al. 18 ) 0.19% - Overall rate of intracranial bleeds taken from Flaker et al. 34 (0.41%), proportion that are intracerebral assumed as for warfarin Other intracranial bleeds Stroke severity for strokes survived beyond 30 days (Combined ischaemic and haemorrhagic) 0.18% Estimated by applying rate ratio for intracranial bleed in warfarin and non-warfarin patients from Go et al. 15 to risk of intracranial bleed in warfarin estimated by taking weighted average from 16 studies cited by Lip et al. 12, and subtracting estimated haemorrhagic stroke risk for non-warfarin patients 60% disabling (Hart et al. 9 ), of remainder moderate:mild ratio = 1:2.95 (Lin et al. 24 ) 0.39% Overall rate of intracranial bleeds estimated by taking weighted average from 16 studies cited in Lip et al 12 (0.73%), proportion of these that are not intracerebral estimated from Hart et al. 14 (54%) 60% disabling (Hart et al. 9 ), of remainder moderate:mild ratio = 1:2.95 (Lin et al. 24 ) Hazard ratio 1.53, relative to warfarin alone (Steinberg et al. 18 ) 0.22% - Overall rate of intracranial bleeds taken from Flaker et al. 34 (0.41%), proportion that are not intracranial assumed as for warfarin Assumed as for warfarin 65% disabling: (13% reduction in disabling strokes, 29% reduction in non-disabling strokes (Hart et al. 9 )), of remainder moderate:mild ratio = 1:2.95 (Lin et al.24) Appendix 1 30-day ischaemic stroke mortality 24% (Hylek et al. 21 ) 16% for INR <2 (62% of total), 6% for INR 2 (Hylek et al. 21 ) Assumed as for warfarin 15% (Hylek et al. 21 ) 19

Transition probabilities No antithrombotics (baseline) Warfarin Warfarin and aspirin Aspirin (See Appendix) 30-day haemorrhagic stroke mortality 30-day mortality, other intracranial bleeds Assumed as for warfarin 41% (Hart et al. 14 ) Assumed as for warfarin 15% (Hylek et al. 21 ) Assumed as for warfarin 29% (Hart et al. 14 ) Assumed as for warfarin Assumed as for warfarin Excess major extracranial bleed (annual) 0 (reference) 0.3% increase relative to no antithrombotics (Hart et al. 9 ) Hazard ratio 1.53, relative to warfarin alone (Steinberg et al. 18 ) 0.2% increase relative to no antithrombotics (Hart et al. 9 ) Stroke mortality post 30-days (Combined ischaemic and haemorrhagic) Disabling stroke: HR 14.3, relative to no stroke. Non-disabling stroke: no significant difference in mortality risk relative to no stroke (De Caterina et al. 22 ) Non-stroke mortality AF all-cause mortality rate estimated using RR all-cause mortality in AF 1.5 (men). 2.2 (women) (Stewart et al. 5 ) and applied to ONS life tables by year of age, further adjusted to remove stroke mortality (Hylek et al. 21, De Caterina et al. 22 ) Table A1.2: Cost estimates and sources for Markov model NHS costs Social care costs Warfarin Annual cost: Medication = 54.75 (BNF), Management = 94.80 (Ali et al. 20 Jowett et al. 28 ) adjusted for inflation (PSSRU 33 ). Sensitivity analyses 1 3, Management costs: 117.83, 247.18, 285.89 respectively (NICE 29 ) Aspirin Annual cost: Medication = 6.19 (BNF 26 ) Warfarin and aspirin Annual cost: Medication = 60.94 (BNF 26 ), Management = 94.80 (Ali et al. 20 Jowett et al. 28 ) adjusted for inflation (PSSRU 33 ) Fatal stroke 3,312 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Disabling stroke First 90-day cost: 26,240 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ). Sensitivity analysis 4: ongoing annual costs 3,270 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Moderate stroke First 90-day cost: 19,212 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ). Sensitivity analysis 4: ongoing annual costs 894 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Mild stroke First 90-day cost: 3,683 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ). Sensitivity analysis 4: ongoing annual costs 414 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Major intracranial bleed (excluding intracerebral) Major extracranial bleed 2,589 (NHS Reference Costs 25, HRGs AA23C-G) 1,092 (NHS Reference Costs 25 HRGs FZ38G-P) Annual cost: 23,101 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Annual cost: 11,527 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Annual cost: 1,020 (Luengo-Fernandez et al. 23 ), adjusted for inflation (PSSRU 33 ) Appendix 1 20