Economic Analysis of Apixaban Therapy for Patients With Atrial Fibrillation From a US Perspective

Key Points

Question

Is apixaban therapy cost-effective relative to warfarin therapy from a US perspective in patients with atrial fibrillation and at least 1 additional risk factor for stroke?

Findings

This patient-level analysis of the ARISTOTLE trial found no difference between treatments in health care costs (excluding anticoagulation costs) but significantly longer expected survival with apixaban therapy compared with warfarin therapy. Over a lifetime, apixaban treatment cost an additional $53 925 per quality-adjusted life-year gained.

Meaning

In patients with atrial fibrillation, lifetime anticoagulation with apixaban therapy rather than warfarin therapy meets current US norms for reasonable value in health care.

Abstract

Importance

The Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation (ARISTOTLE) trial reported that apixaban therapy was superior to warfarin therapy in preventing stroke and all-cause death while causing significantly fewer major bleeds. To establish the value proposition of substituting apixiban therapy for warfarin therapy in patients with atrial fibrillation, we performed a cost-effectiveness analysis using patient-level data from the ARISTOTLE trial.

Objective

To assess the cost and cost-effectiveness of apixaban therapy compared with warfarin therapy in patients with atrial fibrillation from the perspective of the US health care system.

Design, Setting, and Participants

This economic analysis uses patient-level resource use and clinical data collected in the ARISTOTLE trial, a multinational randomized clinical trial that observed 18 201 patients (3417 US patients) for a median of 1.8 years between 2006 and 2011.

Interventions

Apixaban therapy vs warfarin therapy.

Main Outcomes and Measures

Within-trial resource use and cost were compared between treatments, using externally derived US cost weights. Life expectancies for US patients were estimated according to their baseline risk and treatment using time-based and age-based survival models developed using the overall ARISTOTLE population. Quality-of-life adjustment factors were obtained from external sources. Cost-effectiveness (incremental cost per quality-adjusted life-year gained) was evaluated from a US perspective, and extensive sensitivity analyses were performed.

Results

Of the 3417 US patients enrolled in ARISTOTLE, the mean (SD) age was 71 (10) years; 2329 (68.2%) were male and 3264 (95.5%) were white. After 2 years of anticoagulation therapy, health care costs (excluding the study drug) of patients treated with apixaban therapy and warfarin therapy were not statistically different (difference, −$60; 95% CI, −$2728 to $2608). Life expectancy, modeled from ARISTOTLE outcomes, was significantly longer with apixaban therapy vs warfarin therapy (7.94 vs 7.54 quality-adjusted life years). The incremental cost, including cost of anticoagulant and monitoring, of achieving these benefits was within accepted US norms ($53 925 per quality-adjusted life year, with 98% likelihood of meeting a $100 000 willingness-to-pay threshold). Results were generally consistent when model assumptions were varied, with lifetime cost-effectiveness most affected by the price of apixaban and the time horizon.

Conclusions and Relevance

Apixaban therapy for ARISTOTLE-eligible patients with atrial fibrillation provides clinical benefits at an incremental cost that represents reasonable value for money judged using US benchmarks for cost-effectiveness.

Trial Registration

clinicaltrials.gov Identifier: NCT00412984

This patient-level analysis of the ARISTOTLE randomized clinical trial assesses the cost and cost-effectiveness of apixaban therapy compared with warfarin therapy in patients with atrial fibrillation from the perspective of the US health care system.

Introduction

Patients with atrial fibrillation (AF) face an increased risk of ischemic cerebrovascular events and systemic embolism, and daily oral anticoagulation therapy is recommended for at-risk patients. Until recently, warfarin has been the anticoagulant of choice, reducing thromboembolic risk by two-thirds at minimal expense. However, benefits of warfarin therapy come at a price: patients must undergo regular monitoring owing to the drug’s narrow therapeutic range, deal with potential interactions of warfarin with food and other medications, and accept an elevated risk of bleeding. Adoption of warfarin therapy in the community has been less than optimal, reaching about half to two-thirds of eligible patients.

In an effort to improve the quality of anticoagulation therapy, several non–vitamin K oral anticoagulants (NOACs) have been developed. The Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation (ARISTOTLE) trial reported that apixaban therapy, an oral direct factor Xa inhibitor, was superior to warfarin therapy in preventing stroke and all-cause death while at the same time causing significantly fewer major bleeds. However, this superiority in both efficacy and safety domains does not by itself establish the value of apixaban therapy to society. To make a persuasive case for investment of societal resources to provide this therapy, the efficiency of producing these benefits must compare favorably with well-accepted benchmarks of good value for money in health care.

In this study, we used patient-level resource use and clinical data collected in the ARISTOTLE trial to assess the cost and cost-effectiveness of treating patients with AF with apixaban therapy compared with warfarin therapy from the perspective of the US health care system.

Methods

In the ARISTOTLE trial, 18 201 patients with AF and 1 or more additional risk factors for stroke were randomized to apixaban therapy or warfarin therapy in 39 countries between 2006 and 2010. The trial protocol can be found in Supplement 1. Approval of appropriate ethics committees was obtained at all sites. All patients provided written informed consent.

Compared with warfarin therapy, apixaban therapy significantly reduced the primary end point of stroke or systemic embolism (1.27% vs 1.60%; hazard ratio, 0.79; 95% CI, 0.66-0.95) and reduced all-cause death (3.52% vs 3.94%; hazard ratio, 0.89; 95% CI, 0.80-0.998) and major bleeding (2.13% vs 3.09%; hazard ratio, 0.69; 95% CI, 0.60-0.80). For the US economic evaluation, a patient-level survival model was developed using the overall cohort and applied to the 3417 US patients. Costs were based on resource use in the US cohort. This approach assumes that effectiveness is independent of geographic region, as found in the ARISTOTLE trial, but accounts for observed regional variation in resource use and patient risk.

Within-Trial Resource Use

Resource use data collected in the ARISTOTLE trial included dates of inpatient and outpatient care for clinical events; number of nights in intensive care, stepdown unit, or ward; dates of invasive cardiac procedures; and days receiving treatment. Procedures were assigned to reported hospital encounters using dates. Encounters were classified hierarchically as bleeding-related, cardiovascular (without bleeding), noncardiovascular, or unknown. Patients receiving warfarin therapy were assumed to have 8 prothrombin time measurements with clinical follow-up in the first 3 months and monthly monitoring visits thereafter. Patients receiving apixaban were assumed to have semiannual follow-up. Although nonacute health care services were not tracked in the ARISTOTLE trial, use of such services can be affected by occurrence of stroke. To value nonacute care, a longitudinal record of stroke status was created for each patient using dates of disabling and nondisabling stroke.

Within-Trial Cost

External cost weights were developed to value resource use collected in the ARISTOTLE trial. Weights for hospital-based services were estimated using encounter-level cost data from the Premier Research Database (2008-2010) for patients with a diagnosis of AF. Using a generalized linear model with an identity link and gamma distribution function, weights for major components of inpatient and observation encounters were estimated and used to predict a cost weight and associated standard error for each ARISTOTLE hospitalization. Weights for major procedures without a corresponding hospital encounter, outpatient procedures, emergency department visits, and outpatient visits were based on a random sample of corresponding encounter costs from the Premier subset. The costs of physician services, including daily hospital care, procedures, visits, and office-based anticoagulant monitoring, were based on the 2014 Medicare Fee Schedule. Anticoagulant costs were based on the 2014 National Average Drug Acquisition Cost. Cost weight distributions for nursing home care, home health services, medications, and supplies were derived for disabling and nondisabling stroke status using the 2004 National Nursing Home Survey and the 2008 Medical Expenditure Panel Survey data for respondents reporting coronary heart disease or stroke risk factors.

Patient-level costs were calculated by applying cost weights to resource use during trial follow-up. Mean costs were estimated in 3-month intervals using inverse probability weighting methods that accounted for administrative (noninformative) censoring.

Posttrial Projections

Survival

We assumed anticoagulation as observed in the trial would continue over a lifetime. A 2-stage approach was used to model life expectancy. Survival through 2 years was estimated with within-trial data (overall cohort) using a standard time-based Cox proportional hazards model, with adjustment for baseline covariates and treatment. Survival beyond 2 years was estimated using an age-based model, which treats the hazard of death as a function of age rather than time to fully exploit the patient-years of follow-up in the overall ARISTOTLE cohort. Cox proportional hazards methods for left-truncated and right-censored data were used, with adjustment for baseline prognostic factors and stratified by treatment. The time-based and age-based survival models were combined to estimate each patient’s survival curve. Life expectancies, calculated from the area under survival curves, were averaged to obtain the mean predicted life expectancy in each treatment group.

Quality-of-Life Adjustment

Age-based quality-of-life adjustment factors for patients with AF were based on previously developed preference weights for chronic conditions. Expected survival was reduced by a factor of 0.81 at age 67 years, and an age-related adjustment of −0.0003 per year was applied.

Cost

Health care costs beyond trial follow-up were based on experience in years 2 and 3 of follow-up. Apixaban was valued using proprietary rates for the estimated patent-life of 10 years, after which time we assumed generic apixaban would be prescribed. Projected costs of inpatient, outpatient, home health, and nursing home care beyond the study were based on inverse probability-weighted cost estimates for years 2 and 3, estimated as described above (eFigure 1 in Supplement 2).

Statistical Analyses

Baseline characteristics of the US cohort were compared with those of patients enrolled elsewhere, and within-trial resource use and cumulative costs through 2 years were compared between treatment groups. Incremental cost-effectiveness ratios were calculated as the between-group difference in mean lifetime cost divided by the between-group difference in mean quality-adjusted life expectancy. The base case included US patients and assumed (1) a lifetime horizon with projected cost and effectiveness mirroring trial experience, (2) a comprehensive definition of cost (including nonacute care and all procedures), (3) 10 years of proprietary apixaban therapy ($9.47 per day) followed by generic apixaban therapy at 20% of proprietary cost in the treatment group, (4) a warfarin therapy cost of $0.09 per day, and (5) an annual discount rate of 3% for cost and survival. In sensitivity analyses, we excluded estimates of nonacute care costs associated with stroke, excluded proxy costs for major procedures without associated encounters in the case report form, removed the substitution of generic for proprietary apixaban, removed the quality-of-life adjustment, restricted the time horizon over which benefits and costs could accrue, varied the discount rate, and estimated cost and effectiveness for the overall trial population. We also explored the cost-effectiveness of treatment in subgroups of interest (CHA₂DS₂-VASc status and prior vitamin K antagonist [VKA] use).

In each scenario, sampling uncertainty was characterized using nonparametric bootstrap techniques, resampling patients with replacement 1000 times. Uncertainty regarding input costs was incorporated by drawing a value at random for each cost from its respective distribution within each bootstrap iteration. Confidence intervals were calculated for differences using the normal approximation, with standard errors estimated from the bootstrap. Incremental cost-effectiveness was displayed on the cost-effectiveness plane and summarized using cost-effectiveness acceptability curves. All costs were expressed in 2014 US dollars. Analyses were performed using SAS version 9.4 (SAS Institute).

Results

Baseline Characteristics

Compared with patients enrolled elsewhere, US patients were slightly older and were more likely to have histories of diabetes and myocardial infarction (eTable 1 in Supplement 2). Risk factors less common in the US cohort included heart failure and prior stroke, transient ischemic attack, or systemic embolism. Prior VKA use was more common in the United States. Baseline characteristics in the US cohort were balanced between treatment groups (eTable 2 in Supplement 2).

Within-Trial Resource Use and Cost

Over a median follow-up of 1.8 years, one-fourth of patients were hospitalized (Table 1). Almost half of hospitalizations were cardiovascular (excluding bleeding). Although the likelihood of hospitalization overall did not differ significantly between treatments, admissions for bleeding appeared slightly less likely with apixaban therapy compared with warfarin therapy (apixaban: 8.2%; 95% CI, 6.9-9.5; vs warfarin: 10.0%; 95% CI, 8.5-11.4; P = .07). The number of hospitalizations followed a similar pattern. Underlying these averages are high hospitalization rates of patients experiencing stroke, which are difficult to discern in treatment-level comparisons owing to the low absolute incidence of stroke (apixaban: 1.2 admissions per patient with stroke vs 0.38 without stroke; warfarin: 1.5 admissions per patient with stroke vs 0.43 without stroke).

Table 1. Within-Trial Resource Use in US Cohort by Treatment^a.

Variable	Apixaban (n = 1720)	Warfarin (n = 1697)	Differenceb	P Valuec
Patients admitted, No. (%)	457 (26.6)	477 (28.1)	−20 (−1.5)	.31
Bleeding related	141 (8.2)	169 (10.0)	−28 (−1.8)	.07
Cardiovascular (no bleed)d	221 (12.8)	223 (13.1)	−2 (−0.3)	.80
Noncardiovascular	164 (9.5)	186 (11.0)	−22 (−1.5)	.17
Unknown	30 (1.7)	36 (2.1)	−6 (−0.4)	.42
Hospitalizationse	0.39 (0.80) [0,0,1]	0.45 (0.91) [0,0,1]	−0.06 (−0.11 to 0.0)	.05
Bleeding related	0.10 (0.35)	0.12 (0.39)	−0.02 (−0.05 to 0.0)	.10
Cardiovascular (no bleed)d	0.16 (0.46)	0.17 (0.50)	−0.01(−0.04 to 0.02)	.53
Noncardiovascular	0.11 (0.38)	0.14 (.45)	−0.02 (−.05 to 0.0)	.10
Unknown	0.02 (0.15)	0.02 (0.14)	0.0 (−0.01 to 0.01)	.69
Time in hospital, de	2.03 (7.6) [0,0,1]	2.17 (6.1) [0,0,1]	−0.14 (−0.60 to 0.33)	.56
Ward	1.06 (4.7)	1.13 (3.7)	−0.07 (−0.36 to 0.21)	.61
Stepdown	0.57 (2.8)	0.58 (2.4)	−0.01 (−0.19 to 0.16)	.88
Intensive care	0.41 (2.3)	0.46 (2.3)	−0.05 (−0.20 to 0.10)	.53
Emergency department visitse	0.11 (0.40)	0.13 (0.42)	−.02 (−0.05 to 0.01)	.14
Bleeding related	0.04 (0.25)	0.06 (0.26)	−0.02 (−0.03 to 0.0)	.07
Cardiovascular (no bleed)d	0.02 (0.15)	0.02 (0.14)	0.0 (−0.01 to 0.01)	.85
Noncardiovascular	0.04 (0.20)	0.04 (0.21)	0.0 (−0.02 to 0.01)	.81
Unknown	0.01 (0.11)	0.02 (0.13)	0.0 (−0.01 to 0.0)	.30
Outpatient visitse,f	0.34 (0.79) [0,0,0]	0.39 (0.81) [0,0,1]	−0.04 (−0.10 to 0.0)	.11
Patients With Major Procedures, No. (%)
Ablation
Linked with admission	8 (0.5)	7 (0.4)	1 (0.1)	.82
Without corresponding admission	24 (1.4)	35 (2.1)	−11 (−0.7)	.13
Pacemaker
Linked with admission	6 (0.4)	10 (0.6)	−4 (−0.2)	.30
Without corresponding admission	57 (3.3)	56 (3.3)	1 (0.0)	.98
Implantable cardioverter defibrillator
Linked with admission	3 (0.2)	2 (0.1)	1 (0.1)	>.99
Without corresponding admission	21 (1.2)	16 (0.9)	5 (0.3)	.43
Resynchronization device
Linked with admission	2 (0.1)	3 (0.2)	−1 (−0.1)	.69
Without corresponding admission	12 (0.7)	13 (0.8)	−1 (−0.1)	.81
Surgical mazeg
Linked with admission	2 (0.1)	2 (0.1)	0	>.99
Without corresponding admission	12 (0.7)	8 (0.5)	4 (0.2)	.39
CABG and/or valve surgery
Linked with admission	8 (0.5)	4 (0.2)	4 (0.2)	.26
Without corresponding admission	8 (0.5)	5 (0.3)	3 (0.2)	.42
Percutaneous coronary intervention
Linked with admission	23 (1.3)	23 (1.4)	0	.96
Without corresponding admission	32 (1.9)	28 (1.7)	4 (0.2)	.64

Abbreviation: CABG, coronary artery bypass graft surgery.

^aObserved data taken from the Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation trial.

^bApixaban group minus warfarin group.

^cPearson χ² or Fisher exact test for proportions and t test for means.

^dCardiovascular includes admissions with ischemic stroke, transient ischemic attack, systemic embolism, myocardial infarction, syncope, deep vein thrombosis, pulmonary embolism, unstable angina, ischemic chest pain, heart failure, or cardiovascular procedures (and without bleed).

^eMeans (SDs) and percentiles (25th, 50th, 75th) for group values; for differences, mean (95% CI). Percentiles not shown if all equal zero in both groups.

^fIncludes visits for treatment of clinical events; visits for clinical management of chronic disease are excluded.

^gWith or without CABG and/or valve surgery.

Overall, time in the hospital with apixaban therapy was marginally lower, but not statistically different, than with warfarin therapy (mean, 2.03 vs 2.17 days; difference, −0.14; 95% CI for difference, −0.60 to 0.33). The cost of hospitalizations during follow-up averaged $4757 with apixban therapy compared with $5122 with warfarin therapy (eTable 3 in Supplement 2). Emergency department visits for bleeding were less frequent with apixaban therapy than warfarin therapy (mean, 0.04 vs 0.06; difference, −0.02; 95% CI for difference, −0.03 to 0.0), as were outpatient visits for clinical events (mean, 0.34 vs 0.39; difference, −0.04; 95% CI for difference, −0.10 to 0.0). Major procedures were fairly infrequent and similar between treatments (Table 1). Consistent with patterns of resource use, mean cumulative cost of inpatient, outpatient, and emergency department care (excluding unlinked procedures) at 2 years was marginally lower, but not statistically different, with apixaban therapy compared with warfarin therapy (difference, −$172; 95% CI, −$1440 to $1097) (Table 2). The cost of procedures without an associated hospital encounter in the case report form was slightly higher in the apixaban group compared with the warfarin group but again was not statistically different (difference, $376; 95% CI, −$1172 to $2463). Mean estimated nursing home and home health care costs, which reflect stroke care, were slightly lower with apixaban therapy (difference, −$214; 95% CI, −$604 to $177). Overall, there was little evidence of a reduction through 2 years that would offset the incremental cost of apixaban therapy itself, which approached $6000 (difference in health care costs, −$60; 95% CI, −$2728 to $2608).

Table 2. Cumulative 2-Year Within-Trial Costs for US Cohort.

Cost	Discounted 2-y Cost, Mean (95% CI), US$a			P Value
	Apixaban (n = 1720)	Warfarin (n = 1697)	Differenceb
Health care cost
Inpatient, emergency department, and outpatient	5439 (4426 to 6452)	5611 (4847 to 6375)	−172 (−1440 to 1097)	.79
Procedures without encounter in case report form	3734 (2067 to 5402)	3359 (2103 to 4614)	376 (−1712 to 2463)	.72
Nursing home and home health	493 (268 to 717)	706 (387 to 1025)	−214 (−604 to 177)	.28
Medications and supplies	6020 (5643 to 6397)	6070 (5691 to 6450)	−50 (−585 to 484)	.85
Total health care cost without study drug	15 686 (13 572 to 17 800)	15 746 (14 118 to 17 374)	−60 (−2728 to 2608)	.96
Anticoagulation cost
Anticoagulant monitoring and managementc	251 (247 to 256)	1125 (1105 to 1144)	−873 (−893 to −853)	<.001
Study drug	5915 (5793 to 6036)	56 (55 to 58)	5858 (5737 to 5979)	<.001
Total anticoagulant therapy cost	6166 (6040 to 6292)	1181 (1161 to 1202)	4985 (4857 to 5113)	<.001
Total cost	21 852 (19 733 to 23 971)	16 927 (15 299 to 18 555)	4925 (2253 to 7597)	<.001

^aCalculated using inverse probability weighting to account for administrative censoring.

^bApixaban group minus warfarin group.

^cIncludes medical management of atrial fibrillation in both treatment groups.

Extrapolated Costs

Projected health care costs beyond 2 years (excluding anticoagulant therapy) did not differ between groups (mean difference, −$45; 95% CI, −$1635 to $1544 annually before discounting). After discounting, comprehensive health care costs over a lifetime averaged $73 663 with apixaban therapy compared with $70 465 with warfarin therapy. Lifetime anticoagulant monitoring cost was $2980 lower with apixaban therapy ($1164 vs $4144); when combined with lifetime anticoagulant drug cost ($21 769 vs $247), total anticoagulant therapy cost was $18 542 higher with apixaban therapy than warfarin therapy ($22 934 vs $4392). In total, the incremental cost of apixaban therapy compared with warfarin therapy was estimated to be $21 740 ($96 596 vs $74 856; difference, $21 740; 95% CI for difference, $7342 to $36 139).

Life Expectancy

Projected life expectancy of US patients postrandomization was significantly longer with apixaban therapy than warfarin therapy (12.13 vs 11.40 life-years; difference, 0.73; 95% CI, 0.47 to 0.99). These gains in life-years compounded over time, with a difference of just 0.01 apparent after 2 years (eFigure 2 in Supplement 2). After discounting survival at 3%, projected life expectancy with apixaban therapy was 0.50 years greater (9.83 vs 9.33 years; 95% CI for difference, 0.32 to 0.68) (Table 3). Quality-adjusted life-years (QALYs) were 0.40 greater with apixaban therapy compared with warfarin therapy (7.94 vs 7.54 QALYs; 95% CI for difference, 0.26 to 0.55).

Table 3. Cost, Life Expectancy, and Cost-effectiveness by Scenario^a.

Scenario	Mean (95% CI), US$			P Value	ICER	Probability of Cost-effectiveness of Apixabanc
	Apixaban Group	Warfarin Group	Differenceb			$50 000	$100 000
Base case
Comprehensive cost with patent expiration	96 596 (85 650 to 107 543)	74 856 (65 503 to 84 210)	21 740 (7342 to 36 139)	.003	NA	NA	NA
QALY	7.94 (7.84 to 8.05)	7.54 (7.44 to 7.64)	0.40 (0.26 to 0.55)	<.001	53 925	39.3	97.8
Sensitivity analyses
Life-years (base cost)	9.83 (9.71 to 9.96)	9.33 (9.21 to 9.46)	0.50 (0.32 to 0.68)	<.001	43 420	67.5	99.6
Excluding procedures without encounters (base QALYs)	79 128 (72 432 to 85 824)	59 273 (53 953 to 64 592)	19 856 (11 304 to 28 408)	<.001	49 250	50.4	99.5
Excluding nonacute care costs and procedures without encounters (base QALYs)	47 481 (41 502 to 53 460)	26 448 (22 813 to 30 083)	21 033 (14 036 to 28 030)	<.001	52 171	42.2	98.6
Alternate unit cost of apixaban (base QALYs)
25% of current daily cost	80 270 (69 456 to 91 084)	74 856 (65 503 to 84 210)	5413 (−8884 to 19 711)	.46	13 427	98.7	1.0
50% of current daily cost	85 712 (74 859 to 96 565)	74 856 (65 503 to 84 210)	10 856 (−3472 to 25 183)	.14	26 927	92.3	1.0
75% of current daily cost	91 154 (80 257 to 102 052)	74 856 (65 503 to 84 210)	16 298 (1937 to 30 659)	.03	40 426	72.0	99.5
No apixaban patent expiration (base QALYs)	102 322 (91 300 to 113 343)	74 856 (65 503 to 84 210)	27 465 (13 010 to 41 921)	<.001	68 125	13.5	91.8
Restricted time horizons
Within-trial
Comprehensive cost	21 852 (19 733 to 23 971)	16 927 (15 299 to 18 555)	4925 (2253 to 7597)	<.001	NA	NA	NA
QALYs	1.544 (1.542 to 1.547)	1.537 (1.535 to 1.1540)	0.007 (0.004 to 0.011)	<.001	670 589	0	0
5 y
Comprehensive cost	46 390 (41 770 to 51 009)	35 457 (31 541 to 39 373)	10 932 (4876 to 16 988)	<.001	NA	NA	NA
QALYs	3.50 (3.48 to 3.51)	3.46 (3.45 to 3.47)	0.04 (0.02 to 0.06)	<.001	282 468	0	0.3
Discount rate
0%
Comprehensive cost	116 282 (102 719 to 129 844)	90 513 (78 950 to 102 077)	25 768 (7946 to 43 591)	.005	NA	NA	NA
QALYs	9.79 (9.64 to 9.95)	9.20 (9.06 to 9.35)	0.59 (0.38 to 0.80)	<.001	43 682	66.5	99.6
6%
Comprehensive cost	82 193 (73 108 to 91 277)	63 478 (55 708 to 71 248)	18 715 (6760 to 30 699)	.002	NA	NA	NA
QALYs	6.65 (6.57 to 6.72)	6.36 (6.29 to 6.44)	0.29 (0.18 to 0.39)	<.001	65 175	20.2	91.9
Overall cohort used in analysis
Comprehensive cost with patent expiration	92 480 (87 284 to 97 676)	72 510 (67 944 to 77 076)	19 970 (13 053 to 26 888)	<.001	NA	NA	NA
QALYs	7.79 (7.75 to 7.83)	7.49 (7.45 to 7.53)	0.30 (0.24 to 0.36)	<.001	66 366	4.2	99.8

Abbreviations: ICER, incremental cost-effectiveness ratio; NA, not applicable; QALY, quality-adjusted life-year.

^aCosts and benefits discounted at 3% annually unless otherwise indicated.

^bApixaban group minus warfarin group.

^cPercentage of 1000 bootstrap samples with ICER less than specified willingness-to-pay threshold.

Cost-effectiveness

Under base case assumptions, the additional cost of anticoagulation therapy for a patient using apixaban rather than warfarin was $53 925 per QALY gained, with a 98% likelihood of meeting a $100 000 willingness-to-pay threshold (Figure 1) (Table 3). The ratio was fairly stable when assumptions regarding cost were varied (Table 3). Excluding unlinked procedures improved the ratio slightly to $49 250, while excluding nonacute health care costs associated with stroke along with unlinked procedures moved the ratio back toward the base case ($52 171) (Figure 2A). Of cost parameters examined, apixaban price was the most influential; reducing the proprietary price to 75%, 50%, and 25% of the 2014 value reduced the incremental cost-effectiveness ratio to $40 426, $26 927, and $13 427 per QALY gained, respectively, and dramatically increased the likelihood of cost-effectiveness at a $50 000 threshold (Figure 2A). Conversely, continued use of proprietary drug after generic availability increased the cost per QALY gained to $68 125. Ratios exceeded contemporary norms for cost-effectiveness when the time horizon was changed from a lifetime perspective to 5 years ($282 468 per QALY gained) or to the median trial period of 2 years ($670 589 per QALY gained) (Figure 2B).

Figure 1. Distribution of Lifetime Incremental Cost and Effectiveness.

Estimates of incremental cost and quality-adjusted life-years (QALYs) for the base case (1 black circle for each of 1000 bootstrap samples). The solid red circle indicates the point estimate. Difference indicates apixaban group minus warfarin group.

Figure 2. Cost-effectiveness Acceptability Curves: Varying Assumptions Regarding Cost, Effectiveness, and Patient Risk.

A, Effect of varying assumptions regarding the cost of health care services and apixaban therapy on the probability of cost-effectiveness at a range of willingness-to-pay thresholds. B, Effect on cost-effectiveness of removing quality-of-life adjustment, restricting the time horizon, and basing analysis on the total study cohort. C, Cost-effectiveness of apixaban therapy for different levels of thromboembolic risk. QALY indicates quality-adjusted life-year.

Exploratory analyses stratified by CHA₂DS₂-VASc scores suggest that while absolute life expectancy declines with increased risk of thromboembolism, cost-effectiveness is relatively stable (Figure 2C) (eTable 4 in Supplement 2). History of VKA use had no discernable effect on results.

Estimated life expectancy for the US cohort was slightly longer than for non-US patients, owing to differences in baseline demographic and clinical risk factors. Basing life expectancy estimates on the overall cohort resulted in a slightly lower, though more precise, estimate of QALYs gained with apixaban therapy (Table 3). When combined with an estimate of cost for the overall cohort based on US cost weights, the cost-effectiveness ratio rose slightly to $66 366 per QALY gained (Figure 2B) (Table 3).

Discussion

Our analysis suggests that anticoagulation therapy for patients with AF using apixaban rather than warfarin increases average quality-adjusted life expectancy at an additional cost that falls within current US norms for reasonable value in health care. This result stems primarily from gains in life-years accumulated over a lifetime of therapy, as we did not find convincing evidence of a meaningful reduction in health care costs to offset the additional ongoing cost of apixaban therapy. Contributing factors may include the low underlying stroke rate, which muted stroke-related savings, and costs incurred in years of life gained with apixaban therapy.

Estimates of cost-effectiveness were not greatly affected by changes in comprehensiveness of health care cost estimates, quality-of-life adjustment, or discount rates. The most influential determinant of incremental cost was the price of apixaban itself. Reductions in proprietary price and substitution of generic alternatives when available would improve the cost-effectiveness profile of apixaban therapy. Restricting the time horizon of the analysis reduced the incremental QALYs and cost-effectiveness dramatically, clearly demonstrating that the absolute benefits of lowering chronic risk build over a lifetime. Although exploratory in nature, analyses of cost-effectiveness stratified by thromboembolism risk and prior VKA use found no major differences. Assessing the value of a long-term therapy can be challenging when cost and benefits must be projected beyond available evidence. We relied primarily on trial experience to support estimation of ongoing resource use and overall mortality risk for patients with AF as they age. This holistic approach differs from the decision analytic frameworks of previous studies examining the cost-effectiveness of apixaban therapy, which incorporated estimates of risks of individual clinical events and their associated costs that were derived from multiple sources. These modeling studies varied considerably with respect to model assumptions, risk of underlying population, and perspective, as did their estimates of incremental cost and benefit. However, a common finding was that apixaban therapy meets currently accepted thresholds for cost-effectiveness.

When considering alternatives to warfarin, patients and physicians have a choice of 3 NOACs in addition to apixaban: rivaroxaban, edoxaban, and dabigatran. All are as effective as warfarin in preventing stroke, none require regular monitoring of clotting time, and all have been judged cost-effective relative to warfarin, although only apixaban has the joint advantage of significantly reducing the risks of both stroke and major bleeding. However, unlike warfarin, none has an available test for anticoagulation level, and only dabigatran has an approved antidote for use in emergencies, although a universal Factor Xa reversal agent is currently under US Food and Drug Administration review. Perhaps more important to payers, health systems, and patients, all NOACs cost considerably more than warfarin. The additional monthly out-of-pocket expense for NOACs varies dramatically with insurance coverage, averaging $2 for dual Medicare/Medicaid enrollees, $30 for those commercially insured, $70 for Medicare patients with initial Part D coverage, and $130 for Medicare patients in the Part D coverage gap. Clearly, while NOACs meet current thresholds for cost-effectiveness (relative to warfarin) from the perspective of the health care system, medication choice when starting or continuing anticoagulation must depend on the clinical and economic risk profiles of individual patients. An additional consideration is the distribution of the financial burden of NOACs relative to their clinical benefit across the health care sector, as responsibility for affected service areas, such as formularies, anticoagulation clinics, and stroke rehabilitation, may be spread across multiple entities.

Limitations

This analysis had some limitations. First, because cost data were not collected in the ARISTOTLE trial, health services were valued with externally derived unit costs. However, unit cost estimates were based on costs of patients with AF treated at geographically diverse US hospitals, uncertainty of estimates was incorporated in analyses, and estimated costs reflect intensity of care as reported in the trial. Second, some procedures reported on the case report form did not link to health care encounters. However, these procedures occurred infrequently and at similar rates between groups, and their inclusion, while increasing the imprecision of estimates, did not affect results appreciably. Third, nonacute care services commonly provided to patients following a stroke were not captured in the ARISTOTLE trial. Because the absolute rate of stroke in the study was low, including external estimates of these costs had little effect on results. Fourth, productivity losses and informal care were not collected in the ARISTOTLE trial and were excluded. A broader analytical perspective that included patient and caregiver time associated with warfarin monitoring would likely improve the cost-effectiveness of apixaban therapy. Fifth, quality-of-life adjustments did not reflect effects of nonfatal stroke, yielding a conservative estimate of the cost of effectiveness apixaban therapy. Sixth, most ARISTOTLE patients had prior exposure to VKAs, and results reflect this patient mix. However, no interaction between prior VKA use and outcomes was found in the ARISTOTLE trial, nor were costs appreciably affected by prior VKA status. Seventh, consistent with study objectives, cost-effectiveness was estimated for US patients. While estimates lack precision, they capture clinical risk and resource use of the US cohort. Consistency with overall study outcomes was maintained by using the total cohort to estimate effects of clinical risk and treatment on survival. The consistency of results was confirmed in a sensitivity analysis that estimated cost-effectiveness for the overall cohort using US cost weights. Eighth, limited survival data were available for the uppermost tail of the age distribution. However, the contribution of life-years in this range to life expectancy was minor. Our estimates of life expectancy are comparable to recent estimates for patients with AF with similar baseline risk treated with warfarin therapy. Finally, as with all lifetime cost-effectiveness analyses, assumptions were required regarding long-term treatment and its effects. We based extrapolations on trial experience, assuming that effectiveness and adherence observed over time would reflect the chronic state. Large-scale, long-term follow-up of patients in the community could provide data with which to validate results in the future.

Conclusions

Apixaban therapy for ARISTOTLE-eligible patients achieves clinical benefits at an incremental cost that is within a range considered acceptable in the US health care system.

Supplement 1.

Trial Protocol.

Supplement 2.

eTable 1. Characteristics and study drug exposure in patients enrolled in the United States and outsides the United States.

eTable 2. Baseline characteristics of US cohort by treatment.

eTable 3. Cost of hospitalizations during trial follow-up by clinical category and treatment.

eTable 4. Lifetime cost and cost-effectiveness in the US cohort by subgroup.

eFigure 1. Comprehensive costs through 3 years of trial follow-up.

eFigure 2. Estimated treatment-specific survival curves.