VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Available online at journal homepage:

Transcription

1 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Available online at journal homepage: Cost-Effectiveness of Edoxaban vs. Warfarin in Patients with Atrial Fibrillation Based on Results of the ENGAGE AF - TIMI 48 Trial: Taiwanese Perspective Katherine A. Vilain, MPH 1, *, Ming Chin Yang, DrPH 2, Elise Chia Hui Tan, PhD 2, Kaijun Wang, PhD 1, Haiyan Li, MS 1, Wan Hsuan Hsu, MSc 3, Robert P. Giugliano 4, David J. Cohen 1,5, Elizabeth A. Magnuson, ScD 1,5, on behalf of the ENGAGE AF TIMI 48 Investigators 1 Saint Luke s Mid America Heart Institute, Kansas City, MO, USA; 2 Institute of Health Policy and Management, College of Public Health, National Taiwan University, Taipei, Taiwan; 3 Daiichi Sankyo Taiwan Ltd., Taipei, Taiwan (R.O.C.); 4 TIMI Study Group; Brigham and Women s Hospital, Boston, MA; 5 Kansas City School of Medicine, University of Missouri, Kansas City, MO, USA ABSTRACT Background: Novel anticoagulants, such as factor Xa inhibitors, have entered clinical practice as alternatives to warfarin for the prevention of stroke and systemic embolic event (SEE) in patients with atrial fibrillation (AF). It is not known whether edoxaban, the fourth-to-market factor Xa inhibitor approved for this indication, will be cost-effective in Taiwan, where the cost of warfarin monitoring is prohibitive. Methods: A Markov model projecting lifetime results of edoxaban 60 mg/30 mg dose-reduced versus warfarin in patients with nonvalvular AF, based on the ENGAGE AF TIMI 48 trial, found edoxaban to be of high value relative to warfarin, from the perspective of the US health care system. We applied Taiwan-specific cost inputs to this model structure to assess the relative cost-effectiveness of edoxaban versus warfarin from the perspective of the Taiwanese health care system. Event rates and hazard ratios from the ENGAGE AF TIMI 48 East Asian subpopulation were explored in sensitivity analyses. Results: Edoxaban was found to be highly cost-effective compared with warfarin, based on guidelines proposed by the World Health Organization (WHO), with a base case incremental cost-effectiveness ratio of $12,902 per qualityadjusted life year gained. These results were robust to variation of key model parameters, including assumptions regarding the cost and qualityof-life impact of stroke and bleeding events, and assuming East Asianspecific (as opposed to full-trial-population) rates for combinations of ischemic stroke, SEE, and major bleeding. Conclusions: Despite its higher acquisition cost, edoxaban is an economically attractive alternative to warfarinforthepreventionofstrokeandseeinpatientswithafintaiwan. Keywords: atrial fibrillation, cost-effectiveness, decision-analytic modeling, edoxaban, factor Xa inhibitors, health economics, Markov model, Taiwan, warfarin. & 2017 Published by Elsevier Inc. on behalf of International Society for Pharmacoeconomics and Outcomes Research (ISPOR). The authors have indicated that they have no conflicts of interest with regard to the content of this article. * Address Correspondence to: Katherine A. Vilain, Saint Luke s Mid America Heart Institute, 4401 Wornall Road, Kansas City, MO 64111, USA. kvilain@saint-lukes.org The TIMI Study Group, Brigham and Women s Hospital, Boston, MA includes the following: E. Braunwald (Study Chairman), C. McCabe (Director), E. Antman (Principal Investigator), R. Giugliano (Co- Principal Investigator), C. Ruff (Co-Investigator), S. Crugnale, L. Grip (Project Managers), J. Durgin (Assistant Project Manager). Daiichi Sankyo Inc, Parsippany, NJ (sponsor). Clinical Development: T. Bocanegra, M. Mercuri, I. Patel, A. Duggal, J. Hanyok, R. van Kranen, S. Whitaker, J. Dave. Regulatory Affairs: D. Morgan. Clinical Pharmacology: J. Mendell-Harary. Risk Management: Y. Choi. Biostatistics: M. Shi. Project Management: W. Maxwell, A. Inoguchi. Data Coordinating Center, Randomization and Site Management: Quintiles Incorporated, Research Triangle, NC (CRO) Global Operations: C. Giordano (Global Operations Head). Project Management: J. Norris (Senior Project Director), S. Patel (Project Manager). Biostatistics: J. Betcher (Project Statistician), G. Selicato (Functional Lead Statistician), M. Ennis (Independent Unblinded Statistician). SteeringCommittee:D.Alexopoulos,D.Atar,P.Aylward,N.Babilonia,M.Bergovec,J.J.Blanc,R.Botero,J.Camm,S.A.Chen,N.Chung,S. Connolly, R.Corbalan, A. Dalby, R. DeCaterina, M. Dorobantu, T. Duris, M. R. Essop, M. Ezekowitz, A. Garcia-Castillo, A. Goudev, P. Grande, J. Halperin, C. Hassager, H. Heidbuchel, M. Horna, H. Huikuri, S. Juul-Möller, R. Kiss, Y. Koretsune, J. Y. LeHeuzey, B. Lewis, P. Merlini, M. Metra, V. Mitrovic, T. Moccetti,J.Morais,J.C.Nicolau,M.Nieminen,M.Ostojic,A.Oto,T.OudeOphuis,E.Paolasso,A.Parkhomenko,T.Pedersen,D.Roy,M.Ruda,W. Ruzyllo,R.Senior,B.SomaRaju,G.Sotomora,J.Spinar,P.Sritara,J.Voitk,A.Waldo,J.Weitz,H.White,T.Yamashita,Y.Yang,J.L.Zamorano. Data Monitoring Committee: F. Verheugt (Chair), J. Anderson, K. Bauer, J. D. Easton, S. Goto, A. Skene. ClinicalEventsCommittee:S.Wiviott(Chair),E.Awtry,C.Berger,A.Desai,E.Gelfand,C.Ho,D.Leeman,M.Link,F.Ruberg,J.Vita,N.Rost,S. Silverman, W. Goessling, N. Greenberger, W. Maddrey $36.00 see front matter & 2017 Published by Elsevier Inc. on behalf of International Society for Pharmacoeconomics and Outcomes Research (ISPOR).

2 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Because of the challenges associated with treatment with the vitamin K antagonist (VKA) warfarin, novel oral anticoagulants (NOACs) have been introduced in the last decade as a new option for the reduction of ischemic stroke risk in the context of nonvalvular atrial fibrillation (AF). The ENGAGE AF TIMI 48 trial (ClinicalTrials. gov number NCT ) demonstrated that both the higher dose (60 mg/30 mg reduced dose) and lower dose (30 mg/15 mg reduced dose) once-daily regimens of the NOAC edoxaban were superior to warfarin with respect to the prevention of stroke or systemic embolic event (SEE), and both regimens were associated with significantly lower rates of bleeding and death from cardiovascular causes [1]. Edoxaban 60 mg/30 mg reduced dose was approved for use in AF by the Health Authority of Taiwan in February In Taiwan s single-payer health care system, prescription of warfarin and time in therapeutic range (TTR) for AF are relatively low because of the burden of warfarin monitoring costs; the average TTR in Taiwan is below 30% and is believed to undermine warfarin s stroke prevention benefits [2,3]. Higher warfarin adoption and improved maintenance, such as that seen in the ENGAGE AF TIMI 48 trial, could enhance the competitiveness of warfarin against the newer, expensive NOACs, including edoxaban, for patients with AF. However, Taiwanese doctors may be more willing to prescribe edoxaban for this indication because it does not require intense monitoring of the international normalized ratio (INR) and dose adjustment. Further, warfarin relative to no treatment is associated with increased risk of intracranial hemorrhage (ICH) in all races, with the greatest magnitude of risk among nonwhites (hazard ratio [HR] 4.06 for Asians) [4]. It is not known whether reduced rates of ischemic and bleeding events, as seen in the ENGAGE AF TIMI 48 trial, render edoxaban a cost-effective management approach relative to warfarin from the perspective of the health care system in Taiwan. Fig. 1 Representation of the decision model. M represents a Markov process with 20 total health states [9 alive states, each on or off treatment, and 2 absorbing (death) states]. Cycle length is 3 months. Patients remain in their current health state unless a worse event occurs, with the probabilities of these events contingent upon the treatment option and whether patient is currently on or off treatment. Only one event may occur per cycle. The branch from Well represents the potential events, and other health states have branches with similar structures (not shown). CRNM clinically relevant non-major, EDO edoxaban, ECH - extracranial hemorrhage, ICH - intracranial hemorrhage, IS - ischemic stroke, MI - myocardial infarction, SEE - systemic embolic event, S/P - status post, TIA - transient ischemic event.

3 76 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Table 1 Baseline patient characteristics in the ENGAGE AF-TIMI 48 trial. Baseline patient characteristics Base case value Age (years), mean 72 % male 62.3 CHADS 2 score (stratum 1: 2 3; 2: 4 6) 77:23 VKA naive (yes/no) 41:59 Baseline utility Data on the cost-effectiveness of NOACs for AF in Taiwan are limited in the literature; the only such study at the time of publication found dabigatran cost-effective compared with warfarin in Taiwan [5]. In a previously conducted Markov model based cost-effectiveness analysis, based on the ENGAGE AF TIMI 48 trial patient population and clinical results, high-dose edoxaban (60 mg/30 mg reduced dose, hereafter referred to as edoxaban) was found to be an economically attractive alternative to warfarin for the prevention of stroke and SEE in patients with AF, from the perspective of the US health care system [6]. We performed an evaluation of the cost-effectiveness of edoxaban relative to warfarin from the perspective of the Taiwanese health care system, based on results from the ENGAGE AF TIMI 48 trial. Materials and Methods A state-transition (Markov) cohort model incorporating key clinical events and health states associated with the ENGAGE AF TIMI 48 trial clinical endpoints was developed to simulate lifetime health care costs and patient outcomes of edoxaban versus warfarin. Details of the model structure are published in the US perspective cost-effectiveness manuscript [6]. The model was based to the greatest extent possible on the observed results of the ENGAGE AF TIMI 48 trial and was programmed using TreeAge Pro 2015 (TreeAge, Inc, Williamstown, MA). Consistent with our approach to the US analysis, from the Taiwanese perspective, the model cohort was characterized to reflect the demographic characteristics of the entire ENGAGE AF TIMI 48 trial population (mean age 73 years; 58.2% male). Similarly, clinical inputs were based on the full trial population randomized to either high-dose edoxaban or warfarin. We applied acute and Table 2 Hazard ratios, edoxaban versus warfarin, for events based on the ENGAGE AF TIMI 48 trial. Clinical event Base case HR (SE) * Ischemic stroke (0.113) Intracranial hemorrhage (0.072) Major non-ich bleed (0.066) Clinically Relevant Non-major non-ich bleed (0.034) Minor non-ich bleed (0.047) MI (0.122) TIA (0.196) SEE (0.278) Cardiac death other than fatal bleeds and fatal (0.061) thrombotic events HR, hazard ratio; ICH, intracranial hemorrhage; ITT, intention-totreat; MI, myocardial infarction; SE, standard error; SEE, systemic embolic event; TIA, transient ischemic attack. * From on-treatment analysis. from ITT analysis; for bleeding events and thrombotic events to be considered fatal, death had to be within 7 days of the event. long-term utilities and Taiwanese costs to events and health states to assess the lifetime incremental cost-effectiveness in terms of cost per quality-adjusted life year (QALY) gained. A diagram of the model is provided in Figure 1. In the ENGAGE AF TIMI 48 trial, edoxaban was associated with even greater efficacy relative to warfarin in patients who were vitamin K antagonist (VKA) naïve versus patients who were VKA experienced at baseline, in part because of increased incidence of clinical events and bleeding during the initial VKA titration period [7]. Forty-one percent of all trial patients were VKA naïve at baseline. Variation in cost-effectiveness results according to prior VKA-use status was explored in subgroup analyses. One-way and probabilistic sensitivity analyses were conducted on a range of model inputs, including application of Other East Asia (Taiwan, Korea, and China)-specific risks for stroke/see and major bleeding [8]. Binary CHADS 2 score category (low ¼ 2 3 and high ¼ 4 6) at baseline, a stratification factor in the trial, was addressed in separate cost-effectiveness analyses [9]. Characteristics of the patient population in the ENGAGE AF TIMI 48 trial are shown in Table 1. Model Inputs Risk of clinical events Age-adjusted ischemic and bleeding event rates for warfarintreated patients in the ENGAGE AF TIMI 48 trial were used to estimate event-specific transition probabilities [6]. Supplementary Table 1 in the online Appendix shows the logistic regression results for estimation of warfarin-specific event risks, each of which was modeled as a function of patient age. An ontreatment analysis of data from the ENGAGE AF TIMI 48 trial (Table 2) was used to derive event-specific hazard ratios for edoxaban versus warfarin. The underlying mortality risk applied in the model was based on age-specific and sex-specific long-term mortality rates from the life tables for the Taiwan- Fuchien area [10]. A calibration factor of 2.23 was applied to these rates to account for the increased risk of mortality associated with AF in the Taiwanese population [11,12]. Case-fatality rates following thrombotic events (stroke, myocardial infarction [MI], SEE) and bleeding events (ICH, major extracranial hemorrhage) captured in the model were derived from the ENGAGE AF TIMI 48 trial results. Increased risks for ischemic stroke following a stroke or transient ischemic attack, and for death following stroke (by severity), MI, or SEE, as well as after experiencing more than one type of event, were applied as relative risk ratios obtained from published clinical literature and are shown in the online Supplementary Table 2. Treatment interruptions and discontinuation As described in the US-perspective analysis, the Markov model incorporated periods of treatment interruption, including base case assumptions of a 5% annual spontaneous interruption rate for reasons unrelated to cardiovascular or bleeding events, 100% permanent discontinuation following ICH, and 25%:75% permanent/temporary (6-week) discontinuation/interruption following major extracranial hemorrhage. Patients undergoing permanent discontinuation of oral anticoagulant therapy were assumed to receive treatment with aspirin, with event-specific transition probabilities informed by results of published trials comparing warfarin to aspirin (online Supplementary Table 2). Separate off- OAC versions of each health state in the model were used to capture permanent discontinuation. In the rare cases where spontaneous permanent discontinuation occurred for a patient with no history of bleeding and an ischemic event then occurred, the patient would then be assumed to have resumed warfarin therapy (irrespective of randomized treatment arm) and would transition to 1 of 5 on-warfarin postischemic event states, with warfarin-specific event rates and ischemic/mortality hazards. From

4 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Table 3 Utility estimates for disease states and acute events. State utility * Transient disutility References Baseline utility NVAF patients (ENGAGE AF TIMI 48 trial) Decrement for each year of age increase [12,13] Ischemic stroke Mild 0.75 [14,15] Moderate 0.65 Assumption Severe 0.39 [14,15] Hemorrhagic stroke Mild 0.75 [14,15] Moderate 0.65 Assumption Severe 0.39 [14,15] Other ICH [10,16] Major ECH [10,16] CRNM bleed [10,16] Minor bleed [12,17] MI 0.87 [14] SE 0.87 Assumption TIA [12,17] Warfarin treatment [18 20] Edoxaban treatment Assumption Aspirin treatment [18 20] CRNM, clinically relevant non-major; ECH, extracranial hemorrhage; ICH, intracranial hemorrhage; MI, myocardial infarction; NVAF, nonvalvular atrial fibrillation; SEE, systemic embolic event; TIA, transient ischemic event. * State utilities listed represent the multiplicative impact on baseline utility; for patients in health states characterized by 4 1 event of different types, the impact of the second event on utility is multiplicative. For example, for a patient in the S/P MI þ SEE state in Figure 1, utility is first adjusted by a factor of 0.87 for the MI, and an additional factor of 0.87 for the SEE. Applied (subtracted) only in cycle in which event occurs. Decrement per year, accrued incrementally every cycle. these states, treatment could again be interrupted either spontaneously or following a bleeding event. This structure allowed the model to reflect a more clinically realistic treatment pathway. Effect of treatment on cardiovascular mortality A key element driving the cost-effectiveness results of the USperspective analysis was that the impact of edoxaban versus that of warfarin on mortality was captured through two separate effects: 1) the downstream effect of nonfatal ischemic events and hemorrhagic stroke explicitly included in the model on subsequent mortality; and 2) a direct effect relative to warfarin on cardiovascular (CV) mortality distinct from these explicitly modeled events. Among CV deaths not attributed to fatal events (defined as death occurring within 7 days of an event), a majority (684 of 1112) in the ENGAGE AF TIMI 48 trial occurred while the patient was off treatment. Interruptions in treatment largely resulted from major, nonfatal bleeds, for which anticoagulation was frequently halted for Z4 days; 53% of patients resumed treatment with the study drug, and 10% never took another dose of OAC [13]. Edoxaban was associated with a statistically significant reduction in the risk of CV death in the trial primarily because of the prevention of fatal bleeding events but also because of the prevention of nonfatal bleeds, which caused more treatment interruption in the warfarin arm of the trial. This finding led to the decision to model the impact of edoxaban on CV deaths other than fatal thrombotic and bleeding events based on results from an intention-to-treat (ITT) analysis as opposed to on-treatment analysis. An on-treatment analysis would have failed to capture the portion of the edoxaban effect because of the prevention of events that occur off therapy (which are more frequent in the warfarin arm). The ITT analysis generated an HR that was applied to the non event-related CV mortality rate for the edoxaban arm of the model to capture this unique effect. Utilities and costs Health state utilities, as well as disutilities associated with thrombotic and bleeding events, were obtained from clinical literature and are summarized in Table 3. Treatment-associated utilities were also assigned, with the utility associated with warfarin treatment being lower than that for edoxaban or aspirin. The cohort utility upon entering the model was set at 0.836, to reflect the mean baseline utility score of all subjects in the ENGAGE AF TIMI 48 trial. Two different sets of health care costs associated with events were applied: the first in the cycle in which the event occurred ( acute event costs ) and the second (for events considered to have downstream costs) per cycle in the associated postevent state ( postevent costs ). Acute event costs were derived from average total reimbursements for health care claims from the Taiwanese National Health Insurance for the first 3 months after the event occurred (the date of event as the index date and then follow-up to the 90th day). Postevent state cycle costs were derived similarly, as the average annual cost following the acute event period (e.g., from day 91 on). Event and postevent costs were estimated by restricting claims data to patients who experienced each event type, according to diagnoses defined by the International Classification of Disease, 9th revision, Clinical Modification (ICD-9-CM) as they appeared in outpatient and hospitalization claims obtained from Taiwan s National Health Insurance Research Database (NHIRD), which includes reimbursements for outpatient, ambulatory, and hospital inpatient care, as well as dental services. Acute event-related costs and long-term annual costs for the postevent health states (Table 4)

5 78 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Table 4 Cost estimates for disease states and acute events. were expressed in 2015 US dollars (USD, assuming the 2014 exchange rate of New Taiwanese dollars [NTD]to the USD). Warfarin and aspirin treatment costs were derived from the average unit cost of their respective drug classes (warfarin: Anatomical Therapeutic Chemical (ATC) code B01AA03; aspirin: ATC code B01AC06) per the NHIRD. Warfarin dosing was assumed to average 5 mg/day. The cost of an INR monitoring test was based on the weighted average cost per reimbursed examination of coagulation time from the National Health Insurance Administration and the NHIRD [14]. INR monitoring costs were incurred at a rate of eight monitoring events in the first 3-month cycle, and three per cycle thereafter. Aspirin was assumed to average a daily dose of 81 mg. Edoxaban was priced at a base case of NTD $89 (USD$2.73) per daily dose, with an alternate price of NTD$100 (USD$3.07) per daily dose explored in sensitivity analyses. Treatment-related costs are shown in Table 5. Cost-Effectiveness Analysis Annual costs associated with postevent state (USD$) Acute costs (for cycle in which event occurs; USD$) * Ischemic stroke Mild 1,116 1,692 Moderate 1,825 3,126 Severe 2,050 6,946 Hemorrhagic stroke Mild 858 3,892 Moderate 1,785 6,758 Severe 2,419 13,850 Other ICH (nonfatal) 10,201 Major ECH (fatal or 3,256 nonfatal) CRNM bleed 2,057 Minor bleed 846 MI (fatal or nonfatal) 6,793 7,197 SEE (fatal or nonfatal) 4,307 4,202 TIA 1,090 Ischemic stroke þ 3,379 hemorrhagic stroke MI þ SEE 9,724 MI þ TIA 4,396 Fatal ischemic stroke 5,523 Fatal hemorrhagic stroke 5,511 Fatal other ICH 8,630 All costs from Taiwan s National Health Insurance Research Database (NHIRD) and converted to USD. ECH, extracranial hemorrhage; ICH, intracranial hemorrhage; MI, myocardial infarction; NVAF, nonvalvular atrial fibrillation; SEE, systemic embolic event; TIA, transient ischemic event;, USD, United States dollars. * Cycle in which event occurs also incurs one-sixteenth of the annual postevent state costs (column 1), under the assumptions that the event occurs, on average, halfway through the cycle, and the acute care costs cover through the third quarter of that cycle. The incremental cost-effectiveness ratio (ICER) for edoxaban versus warfarin was calculated by dividing the difference in mean total health care costs between the edoxaban and warfarin groups by the corresponding difference in mean QALY. For the purposes of cost-effectiveness analysis, all future life-years and QALYs were discounted at 3% per year, consistent with current guidelines [15]. The thresholds applied to express relative costeffectiveness in a context specific to the Taiwanese economy are explained in the following section. Sensitivity Analyses One-way sensitivity analyses were carried out to examine the impact of specific model assumptions and patient characteristics on the cost-effectiveness results; results from these analyses are summarized graphically as tornado diagrams, which rank the variables assessed in terms of the magnitude of their influence on the ICER. The overall impact of model parameter uncertainty was assessed by using probabilistic sensitivity analysis (PSA), whereby all model parameters were sampled randomly from their appropriate distributions over 1000 runs. Results from these analyses are presented graphically as cost-effectiveness acceptability curves. Distributional assumptions for the PSA are available in the online Appendix, Supplementary Table 3. In addition, we provide the percentage of the joint distribution of cost and effectiveness differences from each PSA that corresponds to ICERs of o 1 Taiwanese gross domestic product (GDP)/person ( ¼ USD$22,704), 1 3 Taiwanese GDP/person, and 4 3 Taiwanese GDP/person, a metric proposed by the WHO to characterize interventions as highly cost-effective, cost-effective, and not cost-effective, respectively [16]. A higher rate of ischemic stroke or SEE has been documented in East Asian patients with AF treated with warfarin compared with Western patients [17]. To assess the cost-effectiveness implications of population differences in outcomes seen in the ENGAGE AF TIMI 48 trial, a key set of sensitivity analysis incorporated HRs (applied to base case warfarin-specific event rates) for ischemic and bleeding events from the East Asian (Taiwan, China, and Korea) trial sub-population [8]. For ischemic stroke and SEE, the HR was 0.31 (95% confidence interval [CI] ); for major extracranial bleed, the HR was 0.39 (95% CI ). Subgroup Analyses Subgroup analyses were performed for a binary CHADS 2 score grouping (2 3 vs 4 6), as well as for the strata by prior VKA experience (with VKA-experienced state defined as VKA use for Z 60 days at the time of study entry). Parameters were identical to the base case for the CHADS 2 analyses, except that both the age and sex distributions, and the risk of stroke and SEE for patients on warfarin, were estimated separately using subgroup-specific data from the ENGAGE AF TIMI 48 trial. For the VKA subgroups, we re-estimated event rates for patients randomized to warfarin, as well as HRs for the effect of edoxaban on events for which the interaction between treatment and prior VKA use was significant at P o Only for ischemic stroke and clinically relevant Table 5 Treatment costs (USD). Treatment Costs Reference Warfarin $8.65/month Taiwan s NHI drug list Warfarin monitoring $36.80/cycle 1 * Taiwan s NHI fee $13.80/cycle 2þ schedule Edoxaban $2.73/day Assumption Aspirin $1.02/month Average price from Taiwan s NHI drug list All costs from Taiwan s National Health Insurance Research Database (NHIRD). NHI, National Health Insurance. * Higher rate of INR testing in first month: 8 INRs/cycle 1; 3 INRs/ cycle thereafter at $4.60/INR.

6 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Table 6 Cost-effectivenessofedoxabanforthebasecaseand selected subgroups and sensitivity analyses. Cost in USD QALYs ICER Results from PSA * EDO WAR ΔE-W EDO WAR ΔE-W % ICERs o1 x TW GDP % ICERs 1-3 x TW GDP % ICERs 43 x TW GDP BASE CASE All Patients (CHADS 2 2) 11,718 7,672 4, ,902 84% 14% 0% SUBGROUP ANALYSIS CHADS ,438 7,984 4, ,241 82% 14% 2% CHADS ,342 7,594 3, ,745 84% 12% 2% VKA-naïve 11,641 7,552 4, ,167 91% 7% 1% VKA-experienced 11,967 7,672 4, ,387 78% 17% 3% SENSITIVITY ANALYSIS Other CV death HR set to 1 11,353 7,672 3, ,251 0% 95% 5% MI HR set to 1 11,845 7,672 4, ,949 81% 14% 3% Other CV Death and MI HRs set to 1 11,472 7,672 3, ,474 0% 91% 9% Edoxaban cost NTD 100/tab 12,373 7,672 4, ,993 79% 17% 2% Post-stroke costs 50% of base case 11,179 7,104 4, ,994 84% 12% 2% 75% :25% permanent:temporary 11,725 7,672 4, ,912 84% 12% 2% discontinuation following ICH No drug treatment-related disutility 11,718 7,672 4, ,116 75% 18% 3% Life Years gained instead of QALY gained 11,718 7,672 4, ,756 80% 13% 3% ITT (vs. on-treatment) HR estimates 11,901 7,650 4, ,517 80% 16% 2% East Asian vs. Japanese HRs for IS & SE 11,635 7,672 4, ,548 95% 4% 1% East Asian vs. Japanese HRs for major bleed 11,096 7,672 3, ,843 88% 9% 2% East Asian vs. Japanese HRs for IS & SEE and major bleed 11,008 7,672 3, ,616 97% 3% 0% ΔE-W = difference (edoxaban warfarin), CV = cardiovascular, EDO = edoxaban 60 mg/day, ICER = incremental cost-effectiveness ratio, ITT = intention-to-treat, GDP = Gross Domestic Product, HR = hazard ratio, ICH = intracranial hemorrhage, MI = myocardial infarction, QALY = quality-adjusted life-year, PSA = probabilistic sensitivity analysis, TW = Taiwan, USD = United States Dollars, VKA = vitamin K antagonist, WAR = warfarin Results from 1000 runs sampling once from each distributional model input. Total may sum to o100% when there were runs where EDO cost more than and was less effective than WAR (i.e., was inferior to WAR). TW GDP per capita in 2014 was US$22, Per Capita GNI. In: Directorate General of Budget, Accounting and Statistics (DGBAS) of Executive Yuan, R.O.C.; non-major bleed were the treatment VKA interactions significant at this threshold (see the online Supplementary Table 4 for the VKA-specific event rates and Supplementary Table 5 for the VKAspecific HRs applied in the model). Results Base Case and Associated Probabilistic Sensitivity Analyses Model results for the base case analysis are summarized in row 1 of Table 6. A strategy of lifetime treatment results in average treatment-related and event-related costs of USD$11,718, compared with $7672 for warfarin, yielding incremental costs of $4046 for edoxaban. QALYs for edoxaban were 5.860versus for warfarin, yielding an incremental QALY gain of and an ICER of $12,902 per QALY gained. Results from PSA for the base case are summarized in the last three columns of Table 6; 84%, 14%, and 0% of the joint distribution of cost and effectiveness differences fell below the lines representing very cost-effective, cost-effective, and not cost-effective strategies, respectively, in the Taiwanese setting according to GDP/person (these proportions sum to o 1 because 2% of PSA runs resulted in inferiority of edoxaban vs warfarin; that is, where edoxaban was both more costly and less effective than warfarin). Figure 2 shows the distribution of results from the PSA on the cost-effectiveness plane, and Figure 3 provides the cost-effectiveness acceptability curve demonstrating the percentage of PSA runs falling at or under a continuously increasing threshold of cost-effectiveness. One-Way Sensitivity Analyses We explored the effect of different unit costs for edoxaban on costeffectiveness results. Assuming daily edoxaban costs of NTD$50, NTD$75, NTD$100, and NTD$125 (USD$1.53, 2.30, 3.07, and $3.83) per day, 96%, 90%, 79% and 64% of ICER estimates for edoxaban are o 1 Taiwanese GDP/person/QALY gained, respectively. Results from other sensitivity analyses are presented in Table 6. Results from a series of one-way sensitivity analyses examining the impact of variations in key model inputs around their base case values are presented in terms of a tornado diagram in Figure 4. Setting the HR for the effect of edoxaban versus warfarin on the risk of CV death other than fatal thrombotic and bleeding events equal to 1.0 had the single greatest impact on the ICERs increasing it to $37,251/QALY gained for edoxaban (not pictured in Figure 4, as inclusion obscures visibility of other values in the tornado diagram). The price of edoxaban had the second greatest influence on the cost-effectiveness results, followed by age, with cost-effectiveness improving as age increases; this was followed by the price of warfarin. Assumptions regarding the cost of warfarin monitoring and the costs and utilities applied to postischemic and posthemorrhagic stroke states had relatively little influence on the costeffectiveness results.

7 80 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) A series of sensitivity analyses was conducted using published HRs for the (non-japanese) East Asian subgroup of the ENGAGE AF TIMI 48 trial population [8]. Shimada et al. [8] found that the median TTR (INR ) for East Asian patients outside of Japan enrolled in the ENGAGE AF TIMI 48 trial was 56%, which was lower than the median TTR of 68.4% for the overall population in the trial. Hence, the relative clinical and economic benefit of edoxaban versus warfarin could differ by geographical region. Application of the East Asian HRs for ischemic stroke/see resulted in greater cost differences between edoxaban and warfarin than in the base case, and higher QALYs for edoxaban than in the base case; consequently, the ICER was reduced to $9340 per QALY gained with edoxaban. Applying the major bleedspecific HRs decreased edoxaban costs and the ICERs, and increased edoxaban QALYs as well, but to a lesser extent than the ischemic event HRs. Applying the HRs for ischemic stroke, SEE, and major bleed in a third sensitivity analysis resulted in an ICER of US$8932 for edoxaban. Fig. 2 Joint distribution of cost and effectiveness differences from probabilistic sensitivity analysis: edoxaban vs. warfarin. Subgroup Analyses Incremental cost-effectiveness results were similar for the two CHADS 2 score subgroups ($13,241 and $12,745 per QALY gained for CHADS and CHADS 2 4 6, respectively; Table 6). Average costs and QALYs for both treatment groups were lower in the Fig. 3 Cost-effectiveness acceptability curve, base case results for the comparison of edoxaban to warfarin, from probabilistic sensitivity analysis. Fig. 4 1-way sensitivity analysis (tornado diagram) on ICER, edoxaban vs. warfarin.

8 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Fig. 5 CHADS and CHADS subgroups: Cost-effectiveness acceptability curves from probabilistic sensitivity analyses. CHADS subgroups because of a lower baseline age (mean age 69.7 years) in the lower CHADS 2 subgroup than in the higher CHADS 2 group (mean age 73.5 years); the additional life-years for the younger cohort offset the cost savings associated with the lower-risk category. Cost-effectiveness acceptability curves from these PSAs for each of the CHADS 2 subgroups and the base case are presented in Figure 5. Results stratified by prior VKA use reveal ICERs that are more favorable for edoxaban for VKA-naïve patients ($11,167 vs $14,387 for patients with prior VKA use; Table 5). Median age for VKA-naïve patients was 71 years; median age for VKA-experienced patients was 72 years. Figure 6 shows the cost-effectiveness acceptability curves for both prior VKA use subgroups and the base case. Discussion This is the first evaluation of the cost-effectiveness of edoxaban versus warfarin from the Taiwanese perspective. We took a purposefully conservative approach to informing the model with respect to key model inputs, using data from the pivotal the ENGAGE AF TIMI 48 trial to the greatest extent possible. Our analysis yielded a base case ICER of $12,902 per QALY gained with edoxaban versus warfarin. The 2003 WHO guidelines to costeffectiveness analysis proposed that treatments with ICERs o 1 GDP/person per gained be considered highly cost-effective, those with ICERs between 1 and 3 GDP/person per QALY gained be considered cost-effective, and those with ICERs 4 3 GDP/ person per QALY gained be considered not cost-effective. 16 Accordingly, at a GDP of $22,704, results from this analysis support edoxaban, at the price of NTD$89/day, as highly costeffective treatment alternatives to warfarin for the reduction in risk of ischemic stroke and SEE in patients with AF in Taiwan. These results are robust to variation in key model parameters, including assumptions regarding the cost and quality-of-life impact of stroke and bleeding events. In sensitivity analyses, the most influential factor was the relative reduction in non event-related CV mortality with edoxaban, which was demonstrated in the trial, followed by the price of edoxaban and then Fig. 6 VKA-naïve and VKA-experienced strata: Cost-effectiveness acceptability curves from probabilistic sensitivity analyses.

9 82 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) patient age. Results were favorable across both CHADS 2 score stroke risk categories. An evaluation of cost-effectiveness of a chronic anticoagulation therapy over the lifetime of patients with AF must take into consideration the impact of mortality. Differential treatment effects with respect to mortality endpoints have a large impact on life expectancy gains and can thus greatly influence ICER estimates. The US-perspective cost-effectiveness analysis and associated model, on which this Taiwanese analysis was based, was the first cost-effectiveness evaluation of an NOAC to incorporate a treatment effect on CV mortality not directly related to thrombotic or bleeding events [18]. Analysis of the ENGAGE-AF TIMI 48 data revealed that a majority of the CV deaths in the trial did not occur within 7 days of an acute ischemic or bleeding event and that a majority of these other CV deaths occurred off treatment. Deaths that occur off treatment, and the greater efficacy of edoxaban with respect to the risk of CV death off (vs on) treatment are difficult to interpret; however, a plausible mechanism is that the greater risk of bleeding in the warfarin arm led to cessation of therapy and a subsequent increased risk of death from cardiovascular causes [13]. The ICER estimate was particularly sensitive to the inclusion of the impact of edoxaban versus warfarin on the risk of CV death other than fatal bleeds or thrombotic events. This is the only component of the relative treatment effect included in the model that is obtained from an ITT, as opposed to ontreatment analysis, as it is largely derived from the benefit of edoxaban with respect to preventing CV death following treatment discontinuation related to a major bleeding event. Treatment effects following treatment discontinuation would not be captured in an on-treatment analysis. The price of edoxaban and patient age were the next most influential factors on the cost-effectiveness results. The ICER was more favorable for older patients because stroke and bleeding risks increase with age, as was confirmed in a recent analysis of the ENGAGE trial patients, which found a significant absolute risk difference in bleeding and hemorrhagic stroke rates in these patients 4 75 [19]. Because AF increases with age, and because older adults have worse outcomes of bleeding and hemorrhagic stroke, the true ICER is likely to be even more favorable than our model predicts. Hence, targeting a treatment with a favorable profile with respect to the reduction of stroke and bleeding in patients at higher risk of those events will have a greater incremental benefit per unit cost. We considered the possibility that HRs for the impact of edoxaban versus warfarin on the rates of individual categories of events might differ according to prior VKA use, as clinical results from the ENGAGE AF TIMI 48 trial suggested greater relative benefit with edoxaban for VKA-naïve patients. Costeffectiveness results for VKA-naïve patients were slightly more favorable than those for patients with prior VKA use. These findings are possibly related to challenges associated with establishing and maintaining INR control in VKA-naïve patients, resulting in a greater relative benefit of edoxaban compared with warfarin in VKA-naïve patients. Selection bias may also play a role in that most VKA-experienced patients enrolled in the trial would be known to be warfarin tolerant (i.e., previous failure on warfarin might make a patient less likely to be enrolled), and those randomized to the warfarin arm would be likely to do relatively well on it. Edoxaban is the fourth NOAC to be introduced in Taiwan for the prevention of stroke and SEE in patients with AF. TTR (INR ) has been shown to be relatively low in Taiwan compared with Western countries, both in clinical trials and in practice [3,5,20]. Furthermore, warfarin use is lower in eligible patients in Taiwan than in the West [5]. Taiwan s AF population may be encouraged to adopt NOACs because of their lower associated risks, as compared with those of warfarin. However, warfarin prices and warfarin monitoring costs are very low in Taiwan compared with other countries.theicerofnoacs,suchasedoxabaninafistherefore highly relevant to clinical decision making in Taiwan. A costeffectiveness analysis of dabigatran based on the RE-LY trial and Taiwan s National Health Insurance data was performed by using two real-world scenarios, reflecting 1) all eligible patients receiving warfarin and assumed to maintain the INRs seen in Taiwan; and 2) the percentage of patients receiving warfarin, aspirin, and no treatment set to reflect real-life prescribing practice; in both scenarios, dabigatran was found to be highly cost-effective relative to the alternative treatment(s) ($280 and $10,551 per QALY gained, respectively) [5,20]. Lin et al.[3] published cost-effectiveness results of a Markov model comparing apixaban (using data from the ARISTOTLE and AVERROES trials), dabigatran 110 mg and 150 mg (from the RE- LY trial), rivaroxaban (from the ROCKET-AF trial), and warfarin (for which clinical event rates were pooled from ARISTOTLE, RE-LY, and ROCKET-AF)forstrokepreventioninAFintheTaiwanesesetting. The model was applied to a hypothetical cohort with parameters based on NHIRD demographic data for patients diagnosed with primary or secondary AF, but key costs inputs were US specific. The current analysis of edoxaban is informed by data from the ENGAGE AF TIMI 48 trial, in which INR levels were maintained at higher rates than in real clinical practice (70% for Korea, China, and Taiwan combined) [8]. This suggests that our results are conservative and that edoxaban relative to warfarin is likely to be of high value in clinical practice. We also applied Taiwan-specific costs for all model inputs to improve the relevance of the cost-effectiveness results to the Taiwanese setting. Although empirical data were used to inform our model to the greatest extent possible, this cost-effectiveness analysis is not without limitations; many model parameters had to be obtained from other sources. In addition, o 5% of patients in the ENGAGE AF TIMI 48 trial were enrolled in non-japanese East Asia (and o 2% were enrolled in Taiwan). However, a series of sensitivity analyses applying combinations of event HRs derived from the non-japanese East Asian subpopulation of the ENGAGE AF TIMI 48 trial [8] all resulted in lower ICERs associated with edoxaban versus warfarin, so the base case model is likely to be conservative in terms of application to the Taiwanese population. Conclusions Despite the higher acquisition cost, edoxaban 60 mg/ 30 mg reduced dose daily is consistent with high value relative to warfarin for the prevention of stroke and SEE in patients with AF, from the perspective of the Taiwanese health care system. These results were robust to variation of key model parameters, including assumptions regarding the cost and quality-of-life impact of stroke and bleeding events and assuming risks of ischemic stroke, SEE, and major bleed equal to those derived from the East Asian-specific subpopulation of the ENGAGE AF TIMI 48 trial. Results were favorable across both CHADS 2 score stroke risk categories and both prior VKA use strata and were more favorable in older patients. Acknowledgments The authors would like to acknowledge the pivotal work by Daichii Sankyo and the TIMI group in conducting the ENGAGE AF TIMI 48 trial. Source of financial support: This study was funded by a research grant from Daiichi Sankyo, Taiwan. Publication of the study results was not contingent on the sponsor s approval or censorship of the manuscript.

10 VALUE IN HEALTH REGIONAL ISSUES 12C (2017) Supplemental material Supplemental material accompanying this article can be found in the online version as a hyperlink at vhri or, if a hard copy of article, at healthjournal.com/issues (select volume, issue, and article). REFERENCES [1] Giugliano RP, Ruff CT, Braunwald E, et al. Edoxaban versus warfarin in patients with atrial fibrillation. N Engl J Med 2013;369: [2] Suarez J, Piccini JP, Liang L, et al. International variation in use of oral anticoagulation among heart failure patients with atrial fibrillation. Am Heart J 2012;163: [3] Lin LJ, Cheng MH, Lee CH, et al. Compliance with antithrombotic prescribing guidelines for patients with atrial fibrillation a nationwide descriptive study in Taiwan. Clin Ther 2008;30: [4] Shen AY, Yao JF, Brar SS, et al. Racial/ethnic differences in the risk of intracranial hemorrhage among patients with atrial fibrillation. J Am Coll Cardiol 2007;50: [5] Chang CH, Yang YH, Chen JH, Lin LJ. Cost-effectiveness of dabigatran etexilate for the prevention of stroke and systemic embolism in atrial fibrillation in Taiwan. Thromb Res 2014;133: [6] Magnuson EA, Vilain K, Wang K, et al. Cost-effectiveness of edoxaban vs warfarin in patients with atrial fibrillation based on results of the ENGAGE AF-TIMI 48 trial. Am Heart J 2015;170: [7] O Donoghue ML, Ruff CT, Giugliano RP, et al. Edoxaban vs. warfarin in vitamin K antagonist experienced and naive patients with atrial fibrillation. Eur Heart J 2015;36: [8] Shimada YJ, Yamashita T, Koretsune Y, et al. Effects of regional differences in asia on efficacy and safety of edoxaban compared with warfarin: insights from the ENGAGE AF-TIMI 48 trial. Circ J 2015;79: [9] Gage BF, Waterman AD, Shannon W, et al. Validation of clinical classification schemes for predicting stroke: results from the National Registry of Atrial Fibrillation. JAMA 2001;285: [10] Department of Statistics. Abridged life table in Taiwan-Fuchien area, Available from: [11] Chien KL, Su TC, Hsu HC, et al. Atrial fibrillation prevalence, incidence and risk of stroke and all-cause death among Chinese. Int J Cardiol 2010;139: [12] Dorian P, Kongnakorn T, Phatak H, et al. Cost-effectiveness of apixaban vs. current standard of care for stroke prevention in patients with atrial fibrillation. Eur Heart J 2014;35: [13] Giugliano RP, Ruff CT, Wiviott SD, et al. Mortality in patients with atrial fibrillation randomized to edoxaban or warfarin: insights from the ENGAGE AF-TIMI 48 trial. Am J Med 2016;129:850 7.e2. [14] National Health Insurance Research Database (NHIRD) in Taiwan [database on the Internet]. National Health Research Institutes. Available from: [15] Gold M, Siegel J, Russell L, Weinstein M,editors. Cost-effectiveness in health and medicine. New York: Oxford University Press, [16] Anderson JL, Heidenreich PA, Barnett PG, et al. ACC/AHA statement on cost/value methodology in clinical practice guidelines and performance measures: a report of the American College of Cardiology/American Heart Association Task Force on Performance Measures and Task Force on Practice Guidelines. Circulation 2014;129: [17] Albertsen IE, Rasmussen LH, Overvad TF, et al. Risk of stroke or systemic embolism in atrial fibrillation patients treated with warfarin: a systematic review and meta-analysis. Stroke 2013;44: [18] Giugliano RP, Ruff CT, Wiviott SD, et al. Reduction in bleeding with edoxaban vs warfarin linked to lower all-cause mortality in 21,105 patients randomized in the ENGAGE AF-TIMI 48 trial. Eur Heart J 2014;35(Suppl):867. [19] Kato ET, Giugliano RP, Ruff CT, et al. Efficacy and safety of edoxaban in elderly patients with atrial fibrillation in the ENGAGE AF-TIMI 48 trial. J Am Heart Assoc 2016;5: pii: e [20] Connolly SJ, Ezekowitz MD, Yusuf S, et al. Dabigatran versus warfarin in patients with atrial fibrillation. N Engl J Med 2009;361: