Cost-effectiveness of rivaroxaban versus warfarin for stroke prevention in non-valvular atrial fibrillation in the Japanese healthcare setting

Abstract

Aims: This article aimed to examine the cost-effectiveness of rivaroxaban in comparison to warfarin for stroke prevention in Japanese patients with non-valvular atrial fibrillation (NVAF), from a public healthcare payer’s perspective.

Materials and methods: Baseline event risks were obtained from the J-ROCKET AF trial and the treatment effect data were taken from a network meta-analysis. The other model inputs were extracted from the literature and official Japanese sources. The outcomes included the number of ischaemic strokes, myocardial infarctions, systemic embolisms and bleedings avoided, life-years, quality-adjusted life-years (QALYs), incremental costs and incremental cost-effectiveness ratio (ICER). The scenario analysis considered treatment effect data from the same network meta-analysis.

Results: In comparison with warfarin, rivaroxaban was estimated to avoid 0.284 ischaemic strokes per patient, to increase the number of QALYs by 0.535 per patient and to decrease the total costs by ¥118,892 (€1,011.11) per patient (1 JPY = 0.00850638 EUR; XE.com, 7 October 2019). Consequently, rivaroxaban treatment was found to be dominant compared to warfarin. In the scenario analysis, the ICER of rivaroxaban versus warfarin was ¥2,873,499 (€24,446.42) per QALY.

Limitations: The various sources of data used resulted in the heterogeneity of the cost-effectiveness analysis results. Although, rivaroxaban was cost-effective in the majority of cases.

Conclusion: Rivaroxaban is cost-effective against warfarin for stroke prevention in Japanese patients with NVAF, giving the payer WTP of 5,000,000 JPY.

Keywords:

• Cost-effectiveness analysis
• non-valvular atrial fibrillation
• rivaroxaban
• stroke prevention
• Japan

JEL CLASSIFICATION CODES:

• I11
• I15

Introduction

Atrial fibrillation (AF) is the most common disorder of the heart rhythm. It is estimated that in Japan the prevalence of AF ranges from 0.6 to 1.6%¹. It is associated with an approximately five-fold increase in the risk of a stroke and a two-fold increase in the risk of all-cause mortality².

Prevention of strokes in patients with non-valvular atrial fibrillation (NVAF) is based on anticoagulants. Currently in Japan, warfarin and non-vitamin K oral anticoagulants (NOAC) such as dabigatran, rivaroxaban, apixaban and edoxaban are available for stroke prevention. The Japanese Circulation Society (JCS) guidelines recommend these treatments in patients with a high risk of ischaemic stroke defined as CHADS₂ ≥2 and either recommend or suggest considering their use in patients with an intermediate risk (CHADS₂ = 1)^3,4. Rivaroxaban once-daily is an oral, direct, highly selective inhibitor of factor Xa with predictable pharmacokinetics and pharmacodynamics, with few relevant drug interactions and without the requirement to monitor coagulation parameters.

The efficacy and safety of rivaroxaban for the prevention of strokes in patients with NVAF were demonstrated mainly in phase III randomized controlled trials, in which rivaroxaban was compared to warfarin. The ROCKET AF trial is a randomized phase III, double-blind, multicentre trial, involving 14,264 patients with NVAF^5,6, that compared rivaroxaban once-daily with an adjusted dose of warfarin.

Since Japanese patients were not included in the ROCKET AF study, this trial was replicated in Japan as the J-ROCKET AF study with an adjusted dose of rivaroxaban for Japanese patients. It is a prospective, randomized, double-blind, phase III trial, among 1,280 patients with NVAF. Patients were randomized to receive rivaroxaban (15 mg once daily) or warfarin dose-adjusted according to the Japanese guidelines⁷. In both studies, rivaroxaban demonstrated a risk reduction in the event rate for strokes and systemic embolism (SE). The non-inferiority of rivaroxaban to warfarin was also confirmed^8,9. Currently, rivaroxaban is the most frequently used anticoagulant agent in Japan. Unlike warfarin, rivaroxaban is taken once daily does not require frequent coagulation status monitoring, dose adjustments, dietary and alcohol restrictions. It also has less drug–drug and drug–food interactions than warfarin. Although previous analyses in other countries indicated that rivaroxaban is more cost-effective compared to warfarin^8–17. Clinical practices in Japan such as lower anticoagulation targets, healthcare costs and efficacy and safety of rivaroxaban in the Japanese population are different from other countries. Therefore, an economic evaluation of the treatment was conducted with rivaroxaban in a Japanese healthcare setting. The aim of this analysis was to examine the cost-effectiveness of rivaroxaban compared to that of warfarin in patients with NVAF in Japan.

Methods

Model overview

An analysis was performed using a previously published Markov model^9,16,17. This model reproduces the management and consequences of AF, mainly as the occurrences of cardio- and cerebrovascular events. The structure of the model was found to be appropriate for Japan due to the same treatment strategies being available and the same method (CHADS₂ score) used to evaluate stroke severity. The Markov model was adapted to the Japanese setting by populating the model with local inputs in respect to baseline patients’ characteristics, mortality background, costs and utility values.

The analysis was conducted from a public healthcare payer’s perspective as recommended by the Japanese Health Economics Guidelines (JHEG)¹⁸. The time horizon for the model was set at 30 years, representing patient lifetime. Costs and outcomes were discounted at a 2% annual rate as recommended by the JHEG¹⁸. The model was run using Microsoft Excel 2010.

Patient population

NVAF patients with a moderate to high risk of strokes defined by CHADS₂ score of 2 or higher were entered into the model. The baseline characteristics of patients were retrieved from the J-ROCKET AF patient population. The average age of patients was 71.1 years. At the baseline, the distribution of the risk of strokes among patients was as follows: 16.6% of patients had a moderate risk (i.e. CHADS₂ score of 2) and 83.4% of patients had a high risk (i.e. CHADS₂ score of ≥3). Patients were treated either with rivaroxaban (10 or 15 mg daily) or warfarin (3 mg daily on average) as the comparator. After discontinuation of an initial treatment with rivaroxaban or warfarin, patients were assumed to take aspirin.

Model structure

After patients entered the model with a stable uncomplicated AF, they could transition to the following health states: ischaemic stroke (IS), SE, intracranial bleeding, minor extracranial bleeding, major extracranial bleeding, myocardial infarction (MI) and death. Patients were able to progress between states in the model according to transition probabilities. Recurrence of a stroke was possible as those in a post-stroke state could return to a stroke state, which was considered to be a “recurrent” stroke. Patients discontinuing treatment progressed into a mirror health state, which was populated with event rates for an untreated population. The cycle duration was 3 months. The model structure is presented in Figure 1.

Figure 1. Model structure. Permanent events are boxed; transient events are non-boxed. AF: atrial fibrillation; IC: intracranial; MI: myocardial infarction; Tx: treatment.

Model input parameters

Inputs were obtained from rivaroxaban clinical trials, a network meta-analysis (NMA), literature and official Japanese sources.

Clinical data

The baseline event risks were obtained from the J-ROCKET AF trial enrolling Japanese patients to reflect the risks of the Japanese population. The breakdown of a minor as opposed to a major stroke was done according to an analysis of a nationwide prospective survey conducted by Kimura et al.¹⁹ This study focussed on the characteristics of patients with acute IS registered in the Japan Multicenter Stroke Investigators’ Collaboration registry. The severity of a stroke considered in the model was distributed as follows: 59.3% were minor strokes and 40.7% were major strokes based on the National Institutes of Health Stroke Scale (NIHSS) score with NIHSS ≤ 15 for mild and NIHSS ≥ 16 for severe strokes.

Treatment effect data

A Bayesian NMA was performed, which considered a broad network of evidence from randomized controlled trials of therapies for the prevention of strokes in NVAF patients. First, a systematic literature review was conducted. In total, 32 trials were identified, including both the ROCKET AF and J-ROCKET AF trials. Treatment effect data were obtained from the NMA considering the J-ROCKET AF trial and warfarin trials. The treatment effect evaluated in Japanese patients was preferred in the case of heterogeneity in clinical results between Japanese patients and patients from other countries as advised by the research group on economic evaluation for Japanese public medical benefits JHEG¹⁸: the ROCKET AF trial was not considered in the base case analysis because rivaroxaban pharmacokinetics are different between the Japanese population and others and lower anticoagulation targets and doses of rivaroxaban are used in Japanese clinical practice⁷. However, the ROCKET AF study results were considered in a scenario analysis.

The NMA was restricted to the following treatments: placebo, warfarin and rivaroxaban. The outcomes of interest included IS, haemorrhagic stroke, MI, SE, major bleeding and extracranial and intracranial bleeding. The NMA used odds ratios (OR) as the measure of relative treatment effect and assumed that treatment effects on the OR scale were multiplicative and exchangeable between trials. The statistical model used is described elsewhere^20–23.

Persistence and discontinuation

Discontinuation rates from the J-ROCKET AF were used in the model: 6.7% for the initial cycle and 3.6% for the subsequent cycles for rivaroxaban, while 8.3% for the initial cycle and 4.9% for the subsequent cycles for warfarin.

Mortality

For general population mortality rates, the life tables published by the Ministry of Health, Labour and Welfare (MHLW) were used²⁴.

Specific event mortality was considered in the model. This included event-related mortality and post-event mortality comprising case fatality of major strokes, minor strokes, major bleeding, minor bleeding, intracranial bleeding, SE and MI, as well as post-major stroke, post-minor stroke, post-intracranial bleeding and post-MI mortality. The considered rates are summarized in Table 1.

Table 1. Case fatality and mortality rates.

Health states	%	Low value (%)	High value (%)	Source
Case fatality of a major stroke	9.9	7.4	12.4	Stroke Databank 2015^33
Case fatality of a minor stroke	0.0	0	0	Assumption
Post-major stroke mortality rate	3.5	2.6	4.4	Lip et al.^1
Post-minor stroke mortality rate	0.0	0.0	0.0	Assumption
Case fatality of major bleeding	1.6	1.2	1.9	ROCKET trial safety^9^,^34
Case fatality of minor bleeding	0.0	0.0	0.0	Assumption
Case fatality of intracranial bleeding	14.7	11.0	18.4	Stroke Databank 2015^33
Post-intracranial bleeding mortality rate	4.9	2.6	4.4	Lip et al.^1
Case fatality of SE	0.0	0	0	Assumption
Case fatality rate of MI	9.7	7.3	12.1	ROCKET trial safety^9^,^34
Post-MI mortality rate	2.7	0.0	6.8	Hoit et al.^35

MI: myocardial infarction; OR: odds ratio; SE: systemic embolism.

Resource use and costs

The following cost categories were considered for the economic evaluation: costs of drug acquisition, monitoring visits and event treatment costs. Drug acquisition unit costs for Japan were obtained from Yakka search Version 3.0²⁵. Information on doses was obtained from the Pharmaceutical and Medical Devices Agency²⁶.

Treatment costs were mainly obtained from cost parameters used by Kamae et al. who evaluated the cost-effectiveness of apixaban compared to warfarin for stroke prevention in Japanese NVAF patients²⁷. In this study, two main sources of event treatment costs were used: the Medical Data Vision Co Ltd (MDV) database analysis by Kamae et al., which provided the actual costs of a stroke, bleedings, MI and SE from admission until discharge²⁷, and a follow-up survey of Japanese stroke patients admitted to a university hospital conducted by Hattori et al., which provided the utility and medical costs at one year after onset²⁸. Costs were inflated to the year 2017 using the revision rates published by the MHLW^29,30. The inputs are summarized in Table 2.

Table 2. Overview of drug acquisition, monitoring visits, and event treatment costs.

Event	Cost ￥ (€)	Low value	High value	Source
Acquisition
Rivaroxaban 10 or 15 mg	507.71 (€4.32) (weighted average)/day	456.939 (€3.89)	558.481 (€4.72)	Yakka search Version 3.0^25
Warfarin 0.5–5.0 mg	28.80/day (€0.24)	25.92 (€0.22)	31.68 (€0.27)	Yakka search Version 3.0^25
Aspirin 100 mg	5.6/day (€0.05)	5.04 (€0.04)	6.16 (€0.05)	Yakka search Version 3.0^25
Monitoring visits
Warfarin monitoring visit	2,150 (€18,2826)	1,935 (€16,4544)	2,365 (€20,1109)	Shirobon net^36
Check-ups for other therapy	720 (€6.12)	648 (€5.51)	792 (€6.73)	Shirobon net^36
Stroke treatment costs
Acute treatment—minor	1,014,868 (€8,628.39)	952,860 (€8,101.20)	1,078,803 (€9,172.88)	MDV claims data reported Kamae et al. (mild ischaemic stroke acute care—per episode)^27
Acute treatment—major	1,705,690 (€14,503.85)	1,600,639 (€13,610.57)	1,814, 037 (€15,424.65)	MDV claims data reported by Kamae et al. (moderate and severe ischaemic stroke acute care—per episode) that were weighted by the distribution of patients at hospital discharge reported in the stroke databank 2015 report^27
Follow-on care (per quarter)—minor	0	0	0	Assume no follow-on care for minor stroke
Follow-on care (per quarter)—moderate and major	1,023,310 (€8,700.61)	973,397 (€8,276.23)	1,074, 462 (€9,136.19)	Costs reported by Kamae et al.^27 which used data from Hattori et al.^28 (moderate and severe ischaemic stroke maintenance care per month) that were weighted by the distribution of patients at hospital discharge reported in the stroke databank 2015 report
Rehabilitation costs per day	0	0	0	Rehabilitation costs are captured in stroke treatment costs (MDV claims data reported by Kamae et al.^27)
Bleeding treatment costs
Acute treatment—minor	348,240 (€2,960.91)	313,416 (€2,665.46)	383,064 (€3,257.78)	Stroke care unit hospitalisation cost per day (stroke care unit admits patients with cerebral infarction, cerebral haemorrhage, subarachnoid haemorrhage, and other patients approved by the doctor)^36
				LoS was extracted from Takashima et al.^37, where they reported for the LoS for subarachnoid haemorrhage. The first quartile, i.e. 6 days was assumed to be representative of minor bleeds
Acute treatment—major	666,402 (€5,667.21)	562,202 (€4,779.27)	779,332 (€6,625.09)	MDV claims data reported by Kamae et al. for other major bleedings^27
Acute treatment—intracranial bleeding	2,019,872 (€17,169.83)	1,545,936 (€13,141.16)	2,557,022 (€21,733.86)	MDV claims data reported by Kamae et al.: minor = 1,448,758; moderate = 2,131,870; severe = 2,085,764^27
				Weighted average in function of the distribution of severity level based on the analysis by Sorensen et al.^38
Follow-on care (per quarter)	1,071,885 (€9,110.675)	1,012,591 (€8,606.82)	1,132,848 (€9,628.98)	Costs reported by Kamae et al.^27 which used data from Hattori et al.^28
				Mild haemorrhagic stroke chronic care (per month): 157,117
				Moderate haemorrhagic stroke chronic care (per month): 336,541
				Severe haemorrhagic stroke chronic care (per month): 383,444
				Weighted average in function of the distribution of severity level based on Sorenson 2011
Rehabilitation costs (per quarter)	0	0	0	Assumption validated by clinical experts: Dr Tanahashi recommended that the duration of rehabilitation for intracranial bleeding to be longer than 14 days. However, this duration impacts only the costs and costs were taken as aggregated values from Kamae et al.^27 Therefore, the duration of rehabilitation is not necessary
SE treatment costs
Acute treatment	1,111,879 (€9,450.04)	1,028,899 (€8,744.78)	1,194,860 (€10,154.60)	Reported by Kamae et al.^27 which used data from Tanihata et al.^39 and Evans et al.^40 for systemic embolism acute care per episode
MI treatment costs
Acute treatment	2,136,214 (€18,154.76)	1,922,573 (€16,339.25)	2,360,950 (€20,064.26)	MDV claims data reported by Kamae et al.^27 MI acute care—per episode
Follow-on care (per quarter)	0	0	232,276 (€1,973.97)	Assumption validated by clinical experts, the high value was referenced from MDV claims data of MI chronic care per month reported by Kamae et al.^27

LoS: length of stay; MI: myocardial infarction; MDV: Medical Data Vision Co Ltd; OR: odds ratio; SE: systemic embolism.

Utilities

Japanese sources for utility were used when available (Table 3).

Table 3. Utility parameters.

Utility data	Mean	Low	High	Source
Anticoagulation
Stable—not on therapy	0.8305	0.7475	1.0000	Shiroiwa et al.^41
Stable—on warfarin therapy	0.8175	0.7475	0.9200	Shiroiwa et al.^41
Utility decrement used for Warfarin	0.0130	0	0.08	Dorian et al.^42; Lip et al.^43
Stable—on other therapy	0.8305	0.7475	1.0000	Shiroiwa et al.^41
Stable—initiating warfarin therapy	0.8175	0.7475	0.9200	Shiroiwa et al.^41
Utility decrement used for Initiating warfarin	0.0130	0	0.08	Assumption
Embolic events
Minor stroke	0.7840	0.7580	0.8090	Hattori and Katayama^28
Post-minor stroke	0.7840	0.7580	0.8090	Assumption—same as minor stroke
Major stroke	0.0920	0.0700	0.1170	Hattori and Katayama^28
Post-major stroke	0.5066	0.4559	0.9136	Hallan^44
SE	0.6270	0.5890	0.6640	Hattori and Katayama^28
Bleeding events
Minor bleeding	0.7725	0.7345	0.8015	Dorian et al.^42; Lip et al.^43
Major bleeding	0.6665	0.5925	0.7495	Dorian et al.^42; Lip et al.^43
Intracranial bleeding	0.6795	0.5925	0.7495	Dorian et al.^42; Lip et al.^43
Post-intracranial bleeding	0.5066	0.4559	0.9136	Assumption as post major stroke
Myocardial infarction
MI	0.2205	0.1985	0.2426	Dorian et al.^42; Lip et al.^43
Post-MI	0.7308	0.6578	0.8039	Sanders et al.^45

MI: myocardial infarction; OR: odds ratio; SE: systemic embolism.

Base case analyses

The outcomes of the model included a number of IS and MI events and bleedings avoided, life-years, quality-adjusted life-years (QALYs), incremental costs and incremental cost-effectiveness ratio (ICER) per QALY gained. The ICER per QALY gained is calculated as the difference in total costs (incremental cost) divided by the difference in QALY (incremental effect) between rivaroxaban and warfarin.

Sensitivity analyses

Sensitivity analyses evaluated the impact of assumptions used in the model and variability surrounding model inputs.

Deterministic sensitivity analysis

A series of one-way sensitivity analyses were run for the base case in order to determine the significant drivers of cost-effectiveness. All parameters were included in the variation.

Probabilistic sensitivity analysis

All parameters with the second order uncertainty were tested in the probabilistic sensitivity analysis with 2,000 iterations. Parameters that did not carry the second order uncertainty were excluded: discount rates and the time horizon, unit costs from published reference lists and patient characteristics, such as age and co-morbidities.

Scenario analysis

In a scenario analysis, treatment effect data were derived from the same network meta-analysis, but the two clinical trials on rivaroxaban, regardless the trial location, were considered (J-ROCKET AF trial and the global ROCKET AF study).

Results

Base case

Results of the base case analysis are presented in Table 4. In comparison with warfarin, rivaroxaban was estimated to save ¥118,892 (€1,011.11; 1 JPY = 0.00850638 EUR; XE.com, 7 October 2019), to avoid 0.284 IS events, and to increase the number of QALYs by 0.535 per patient. As a result, rivaroxaban treatment was found to be dominant compared to warfarin over a lifetime horizon.

Table 4. Results of cost-effectiveness analysis.

	Base case			Scenario analysis
	Rivaroxaban	Warfarin	Difference	Rivaroxaban	Warfarin	Difference
Total costs, ¥ (€)	¥6,754,603 (€57,442.22)	¥6,873,496 (€58,456.27)	−¥118,892 (€1,011.11)	¥7,680,947 (€65,319.27)	¥6,854,783 (€58,293.52)	¥826,164 (€7,025.90)
Total QALYs	10.03	9.50	0.535	9.79	9.50	0.288
Total LY	12.63	12.20	0.429	12.44	12.21	0.230
Total ISs	0.58	0.87	0.284	0.79	0.86	0.072
Total bleedings	1.27	1.10	−0.170	1.20	1.11	−0.094
Total MI	0.05	0.02	−0.03	0.02	0.02	0
ICER, ¥ (€) (per QALY)			Dominant			¥2,873,499 (€24,440.31)

ICER: incremental cost-effectiveness ratio; IS: ischaemic stroke; LY: life-years; QALY: quality-adjusted life-years.

Sensitivity analyses

Deterministic sensitivity analysis

Figure 2 presents a Tornado diagram of the key drivers of incremental costs and QALYs. The top key drivers of incremental QALYs and costs were relative risks of MI, intracranial bleeding, minor bleeding, IS and SE for rivaroxaban, relative risk of intracranial bleeding, minor bleeding and IS for warfarin and utility value for stable patients on other therapy.

Figure 2. Results of one-way sensitivity analysis (Tornado diagrams). AF: atrial fibrillation; IC: intracranial; MI: myocardial infarction; RR: relative risk; SE: systemic embolism.

Probabilistic sensitivity analysis

With the lifetime horizon, 34.5% of the iterations fell within the northeast quadrant of the cost-effectiveness plot and 6.6% within the southwest quadrant, indicating the necessity to evaluate its cost-effectiveness to meet the willingness-to-pay thresholds. A total of 19.8% of the iterations fell within the northwest quadrant, indicating that the treatment is dominant as well as 39.2% that fell within the southeast quadrant. This was illustrated in the cost-effectiveness plane in Figure 3.

Figure 3. Cost-effectiveness plane.

The cost-effectiveness acceptability curve shows that there was a probability of 64% for rivaroxaban to be cost-effective compared to warfarin, at a willingness-to-pay of ¥5,000,000 (€42,537) per QALY gained (Figure 4). The willingness-to-pay threshold was estimated at ¥1,400,000 (€11,909.83) with a 50% probability for rivaroxaban to be cost-effective compared to warfarin.

Figure 4. Cost-effectiveness acceptability curve based on QALY.

Scenario analysis

In the scenario analysis, the number of QALYs over the lifetime horizon was 9.50 and 9.79 respectively, for warfarin and rivaroxaban treatments. The total costs were estimated to be ¥6,854,783 (€58,311.15) and ¥7,680,947 (€65,339.03), for warfarin and rivaroxaban, respectively. Rivaroxaban treatment was found to avoid 0.072 IS events per patient. This resulted in an ICER of ¥2,873,499 (€24,443.97) per QALY, indicating that rivaroxaban treatment was found to be cost-effective compared to warfarin treatment.

Discussion

In this cost-effectiveness analysis conducted in the Japanese healthcare setting, rivaroxaban treatment was found to be dominant in comparison to warfarin over a lifetime horizon in the base case analysis.

Because of differences in the dose of rivaroxaban, only Japanese clinical results of rivaroxaban were considered in the base case analysis, as clinical practice such as anticoagulation targets and results of efficacy and safety between Japanese and multinational populations might exist. Nevertheless, clinical results of other populations were tested through the scenario analysis as well and considered both Japanese J-ROCKET AF trial and global ROCKET AF trial of rivaroxaban in the NMA; yet, it was confirmed that rivaroxaban is a cost-effective treatment. Heterogeneity in this model is rooted in the use of various sources of data. The deterministic sensitivity analysis showed that heterogeneity was mainly rooted in the relative risks, which were calculated based on ORs of the meta-analysis and baseline risks from the J-ROCKET AF trial. Table 5 details the large confidence intervals of ORs used in the sensitivity analysis. As some events are rare and confidence intervals are large, considerable heterogeneity around the relative risks appeared and led to heterogeneity in the cost-effectiveness analysis results. However, in the majority of cases, rivaroxaban treatment was found to be cost-effective in comparison to warfarin in the sensitivity analyses.

Table 5. Results of the network meta-analysis (treatment effect data for base case analysis and scenario analysis).

	Baseline event risk (%)	Low value (%)	High value (%)	OR [95% CI] of warfarin vs. placebo (base case and scenario)	OR [95% CI] of rivaroxaban vs. placebo (base case and scenario)
IS	1.40	0.85	2.18	0.36 [0.23, 0.54] 0.36 [0.23, 0.56]	0.14 [0.04, 0.44] 0.28 [0.12, 0.56]
MI	0.07	0.01	0.33	0.42 [0.11, 1.36] 0.43 [0.12, 1.36]	1.61 [0.1, 58.87] 0.38 [0.09, 1.89]
SE	0.07	0.01	0.30	0.46 [0.1, 1.58] 0.47 [0.1, 1.65]	0.47 [0.01, 36.24] 0.13 [0.01, 1.28]
Major extracranial bleeding	0.03	0.01	0.08	3.62 [1.16, 12.96] 3.59 [1.3, 10.77]	3.1 [0.17, 56.72] 3.42 [0.49, 28.05]
Minor extracranial bleeding	0.72	0.95	1.42	4.4 [1.21, 20.33] 4.4 [1.4, 17.57]	4.46 [0.23, 122.59] 4.78 [0.43, 49.65]
Intracranial bleeding	0.07	0.04	0.13	4.21 [0.81, 31.56] 4.22 [0.82, 31.37]	2.24 [0.29, 22.33] 2.71 [0.51, 20.68]

95% CI: 95% confidence interals; IS: ischaemic stroke; MI: myocardial infarction; OR: odds ratio; SE: systemic embolism.

The current model was used in previous published cost-effectiveness analyses of rivaroxaban against warfarin^8,9,17. This model was found appropriate as the most parsimonious solution that captures both stroke severity and recurrence. Conclusions from this study are in line with previous analyses from the payer’s perspective in the United States, Italy, France, Norway, Singapore, Belgium, Greece and Germany^9,11–17 and societal perspective in the US and Portugal^8,10. Costs were found higher for the rivaroxaban treatment compared to the warfarin treatment with a variation in levels of difference depending on healthcare settings. The only exception was for the analysis for the Greek setting where the total lifetime cost was found lower for rivaroxaban treatment resulting in a cost saving of €239¹⁷. The number of QALY was found higher for the rivaroxaban treatment compared to the warfarin treatment^8–17. Rivaroxaban treatment was found to be cost-effective compared to the warfarin treatment in those studies^8–17.

The model is in line with previous Japanese pharmacoeconomic analyses in AF^27,31,32. Our model structure was similar to the published analyses which consider the same health states. This model was populated with inputs, mainly costs and utility inputs that were previously used in the published analyses^27,32. In terms of comparison of the results with previously published Japanese cost-effectiveness analysis models, the use of other NOACs such as dabigatran and apixaban was also found to be more cost-effective against warfarin treatment in Japan^27,31.

The use of various sources of data does carry with it some limitations. Other countries’ data had to be used when domestic data were not available. For example, some utility data from foreign countries were used; however, that data were not found to be drivers of the results. Some assumptions had to be made as well. Post-minor stroke was assumed to have the same utility as a minor stroke because the value could not be identified through literature reviews; testing the variation of this value with deterministic and probabilistic sensitivity analysis also confirmed that this assumption on the model results was minor. More real-world data from the Japanese population, such as baseline event risks and treatment discontinuation, could strengthen this study. However, these inputs were not available for NVAF patients with a CHADS₂ score of 2 or higher or limited to some events. Therefore, for consistency, it was preferred to keep the same data sources. Costs were obtained from Kamae et al. who calculated the costs using data from the MDV database²⁷. This database collected data from 131 hospitals, which represent approximately 9% of Japanese hospitals with diagnosis procedure combination (DPC) systems. However, MDV data is considered to be generalizable to the Japanese population due to its similarity with the MHLW patient distributions.

The strength of this model lies in the adjustment to the local environment, which resulted in the precise evaluation for Japan. Whenever feasible, the inputs were adapted to the Japanese setting, tailoring the model to the Japanese population.

Conclusion

The present analysis suggests that rivaroxaban is cost-effective against warfarin for stroke prevention in Japanese patients with NVAF, giving the payer WTP of 5,000,000 JPY.

Transparency

MH, NT and RG are paid consultants to Bayer Yakuhin, Ltd. for this study and outside of the study. MH has received remuneration for attending meetings (presentations) from Nippon Boehringer Ingelheim Co., Ltd. and Bayer Yakuhin, Ltd. outside of the study. NT has received remuneration for attending meetings (presentations) from Nippon Boehringer Ingelheim Co., Ltd., Bayer Yakuhin, Ltd., Bristol-Myers Squibb, Pfizer, Daiichi Sankyo, outside of the study. GK and JD are employees of Creativ-Ceutical, a health care consulting firm contracted by Bayer Yakuhin, Ltd. SA as an employee of Bayer Yakuhin, Ltd.

The peer reviewers on this manuscript have received an honorarium from JME for their review work. In addition, a reviewer on this manuscript has disclosed that they have received grant funding from all major producers of NOACs. The reviewers have no other relevant financial relationships or otherwise to disclose.

Author contributions

GK and JD contributed to the literature search and analysis. All authors contributed to the study design, interpretation of data, critical revision of the intellectual content and final approval of the version to be published, and all authors agree to be accountable for all aspects of the work.

Acknowledgements

The authors would like to thank Emi Watanabe Fujinuma for project management support. The authors acknowledge that medical writing support was provided by Małgorzata Biernikiewicz of Creativ-Ceutical, with funding from Bayer Yakuhin, Ltd.

Additional information

Funding

This study was funded by Bayer AG and Bayer Yakuhin Ltd.