Real-world comparison of all-cause hospitalizations, hospitalizations due to stroke and major bleeding, and costs for non-valvular atrial fibrillation patients prescribed oral anticoagulants in a US health plan

Abstract

Aims: To compare the risk of all-cause hospitalization and hospitalizations due to stroke/systemic embolism (SE) and major bleeding, as well as associated healthcare costs for non-valvular atrial fibrillation (NVAF) patients initiating apixaban, dabigatran, rivaroxaban, or warfarin.

Materials and methods: NVAF patients initiating apixaban, dabigatran, rivaroxaban, or warfarin were selected from the OptumInsight Research Database from January 1, 2013–September 30, 2015. Propensity score matching (PSM) was performed between apixaban and each oral anticoagulant. Cox models were used to estimate the risk of stroke/SE and major bleeding. Generalized linear and 2-part models were used to compare healthcare costs.

Results: Of the 47,634 eligible patients, 8,328 warfarin-apixaban pairs, 3,557 dabigatran-apixaban pairs, and 8,440 rivaroxaban-apixaban pairs were matched. Compared to apixaban, warfarin patients were associated with a significantly higher risk of all-cause (hazard ratio [HR] = 1.30; 95% confidence interval [CI] = 1.21–1.40) as well as stroke/SE-related (HR = 1.60; 95% CI = 1.23–2.07) and major bleeding-related (HR = 1.95; 95% CI = 1.60–2.39) hospitalization; rivaroxaban patients were associated with a higher risk of all-cause (HR = 1.15; 95% CI = 1.07–1.24) and major bleeding-related hospitalization (HR = 1.71; 95% CI = 1.39–2.10); and dabigatran patients were associated with a higher risk of major bleeding hospitalization (HR = 1.46, 95% CI = 1.02–2.10). Warfarin patients had significantly higher major bleeding-related and total all-cause healthcare costs compared to apixaban patients. Rivaroxaban patients had significantly higher major bleeding-related costs compared to apixaban patients. No significant results were found for the remaining comparisons.

Limitations: No causal relationships can be concluded, and unobserved confounders may exist in this retrospective database analysis.

Conclusions: This study demonstrated a significantly higher risk of hospitalization (all-cause, stroke/SE, and major bleeding) associated with warfarin, a significantly higher risk of major bleeding hospitalization associated with dabigatran or rivaroxaban, and a significantly higher risk of all-cause hospitalization associated with rivaroxaban compared to apixaban. Lower major bleeding-related costs were observed for apixaban patients compared to warfarin and rivaroxaban patients.

Keywords:

• Cardiac arrhythmias
• anticoagulants
• major bleeding
• stroke
• costs
• cost analysis

Introduction

Atrial fibrillation (AF)—a medical condition characterized by chaotic and irregular electrical activity in the upper chambers of the heart—increases the risk of developing blood clots in the atria, causing an abnormal pattern of blood flow and an increased risk of stroke¹. Non-valvular AF (NVAF) refers to AF that is not associated with rheumatic mitral valve disease, mitral valve repair, or prosthetic heart valves; it accounts for 70% of AF cases^2,3. In 2010, there were an estimated 5.2 million cases of AF; by 2030, the number of AF cases is projected to increase to 12.1 million⁴.

NVAF has long been identified as a significant risk factor for disabling or fatal ischemic stroke and systemic embolism (SE)⁵. Specifically, NVAF is associated with a 5-fold increase in stroke risk and accounts for 15% of all strokes in the US^6,7. The attributable risk of stroke for AF increases with age, from 1.5% for those aged 50–59 years to 23.5% for those aged 80–89 years⁸.

Until recently, vitamin K antagonists (e.g. warfarin) were the main drugs recommended to reduce the risk of stroke/SE among patients with NVAF. However, warfarin is associated with a significant risk of major bleeding and requires regular blood monitoring (international normalized ratio) and dose adjustments; warfarin also has several drug and food interactions⁹. These factors result in both non-adherence and discontinuation, causing only about half of AF patients in the US to receive warfarin therapy as recommended^6,10,11. In recent years, several new drugs have been developed with the goal of decreasing stroke and bleeding risks without the inconvenience or drug and food interactions associated with warfarin⁹. Randomized controlled trials of direct oral anticoagulants (DOACs)—including dabigatran, apixaban, rivaroxaban, and edoxaban—have demonstrated that DOACs are at least as effective as warfarin for stroke risk reduction, and are associated with similar or lower rates of major bleeding^12–15. DOACs offer several advantages over warfarin, as they do not require routine monitoring and have fewer drug and food interactions¹⁶.

Previous real-world claims-based studies have shown varied results for the risk of major bleeding in rivaroxaban and dabigatran patients compared to warfarin patients^17–22. However, apixaban has been consistently shown to have a significantly lower risk of major bleeding compared to warfarin^17,18,23,24. A direct comparison between dabigatran, rivaroxaban, and apixaban showed no significant difference in the risk of stroke/SE; however, apixaban was associated with a significantly lower risk of major bleeding compared to the other two DOACs²⁵. Cost analyses have shown that apixaban patients have lower hospitalization costs compared to warfarin patients^26,27. Although very few cost comparisons have been performed between apixaban and the other two DOACs, evidence shows that medical and hospitalization re-admission costs are lower in apixaban patients, followed by dabigatran and rivaroxaban patients^24,28. There have been limited real-world data comparing effectiveness outcomes between oral anticoagulants (OACs) as well as no stroke- and major bleeding-related costs between OACs²⁹. This study aimed to compare the risks of all-cause hospitalization, hospitalization due to stroke/SE, and major bleeding conditions as well as healthcare costs among OAC treatment-naïve NVAF patients.

Methods

Data source

This retrospective analysis used the OptumInsight Research Database, a large administrative claims database including enrollment information, medical and pharmacy claims for Medicare Advantage, and commercial insurance enrollees from January 1, 2012–September 30, 2015³⁰. The database contains data on >100 million enrollees from geographically-diverse regions across the US. The database also contains enrollee information on demographics (e.g. age, geographic region, and gender), health plan enrollment, and characteristics of health services utilization. These characteristics include facility revenue codes; places of service; International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnoses and procedures; Health Care Common Procedure Coding System and Current Procedural Terminology codes; and prescription claims data for each individual³⁰. OptumInsight data are de-identified and fully compliant with all Health Insurance Portability and Accountability Act privacy and security requirements. The comparative effectiveness research (CER) methods guidance documents aided researchers in designing the study and applying appropriate CER methods^31–34.

Selection criteria

Patients who had ≥1 prescription claim for apixaban, dabigatran, rivaroxaban, or warfarin during the identification period (January 1, 2013–September 30, 2015) were selected. Due to the small sample size, edoxaban was not included in our study. The first OAC pharmacy claim date was designated as the index date. Patients were required to be aged ≥18 years, have continuous health plan enrollment with medical and pharmacy benefits for 12 months pre-index date (baseline period), and have an AF medical claim (ICD-9-CM code: 427.31) during the baseline period³⁵. Patients with evidence of valvular heart disease, valve replacement/transplant, venous thromboembolism (VTE), transient AF (pericarditis, hyperthyroidism, and thyrotoxicity), cardiac surgery prior to or on the index date, or pregnancy during the study period were excluded (Supplemental Table 1). The exclusion criteria were designed to exclude patients with valvular AF, per the indications for the use of DOACs⁹. In addition, VTE patients were excluded to restrict the OAC use for NVAF only. Additionally, patients were excluded if they had an OAC pharmacy claim during the baseline period, >1 OAC prescription on the index date, or a follow-up period of 0 days. Patients were then categorized into four cohorts based on their index prescription: apixaban, dabigatran, rivaroxaban, or warfarin. Patient data were observed from the day after the index date until the earliest of the following dates: treatment discontinuation, switch from treatment, inpatient death, health plan disenrollment, or September 30, 2015. Discontinuation was defined as no evidence of an index prescription for 30 days from the last day of the supply of the last filled prescription. Switching was defined as a prescription for an OAC other than the index OAC prescription within 30 days before or after the discontinuation date³⁶.

Table 1. Baseline characteristics in propensity score matched cohorts.

	Apixaban cohort (n = 8,328)		Warfarin cohort (n = 8,328)		Apixaban cohort (n = 3,557)		Dabigatran cohort (n = 3,557)		Apixaban cohort (n = 8,440)		Rivaroxaban cohort (n = 8,440)
	n/Mean	%/SD	n/Mean	%/SD	n/Mean	%/SD	n/Mean	%/SD	n/Mean	%/SD	n/Mean	%/SD
Age	73.54	10.72	73.37	10.42	70.92	11.41	70.68	11.19	72.84	11.06	72.53	10.79
18–54	420	5.04%	422	5.07%	303	8.52%	302	8.49%	528	6.26%	501	5.94%
55–64	1,203	14.45%	1,233	14.81%	677	19.03%	654	18.39%	1,326	15.71%	1,362	16.14%
65–74	2,556	30.69%	2,546	30.57%	1,196	33.62%	1,186	33.34%	2,585	30.63%	2,583	30.60%
≥75	4,149	49.82%	4,127	49.56%	1,381	38.82%	1,415	39.78%	4,001	47.41%	3,994	47.32%
Gender
Male	4,571	54.89%	4,531	54.41%	2,171	61.03%	2,179	61.26%	4,734	56.09%	4,740	56.16%
Female	3,757	45.11%	3,797	45.59%	1,386	38.97%	1,378	38.74%	3,706	43.91%	3,700	43.84%
US geographic region
Northeast	969	11.64%	970	11.65%	476	13.38%	495	13.92%	966	11.45%	975	11.55%
North central	1,826	21.93%	1,845	22.15%	768	21.59%	766	21.54%	1,797	21.29%	1,832	21.71%
South	3,016	36.22%	3,023	36.30%	1,332	37.45%	1,304	36.66%	3,208	38.01%	3,180	37.68%
West	2,481	29.79%	2,448	29.39%	964	27.10%	972	27.33%	2,436	28.86%	2,421	28.68%
Other	36	0.43%	42	0.50%	17	0.48%	20	0.56%	33	0.39%	32	0.38%
Insurance type
Medicare advantage	5,867	71.22%	6,342	76.15%	2,211	62.16%	2,268	63.76%	5,755	68.19%	5,714	67.70%
Commercial	2,446	29.37%	1,974	23.70%	1,335	37.53%	1,279	35.96%	2,670	31.64%	2,719	32.22%
Dosage
Standard	6,647	79.82%			2,995	84.20%			6,820	80.81%
Reduced	1,681	20.18%			562	15.80%			1,620	19.19%$
Baseline comorbidity
Baseline Deyo-Charlson comorbidity index score	2.37	2.33	2.36	2.33	2.01	2.16	2.03	2.13	2.26	2.29	2.25	2.27
Baseline CHADS2 score	2.37	1.37	2.34	1.35	2.13	1.33	2.16	1.34	2.30	1.36	2.28	1.34
0 = low risk	532	6.39%	558	6.70%	314	8.83%	305	8.57%	584	6.92%	583	6.91%
1 = moderate risk	1,811	21.75%	1,732	20.80%	943	26.51%	910	25.58%	1,960	23.22%	1,937	22.95%
2 = high risk	2,534	30.43%	2,702	32.44%	1,040	29.24%	1,085	30.50%	2,565	30.39%	2,673	31.67%
>2 = high risk	3,451	41.44%	3,336	40.06%	1,260	35.42%	1,257	35.34%	3,331	39.47%	3,247	38.47%
Baseline CHA2DS2-VASc score	3.91	1.83	3.89	1.82	3.49	1.84	3.52	1.84	3.79	1.83	3.77	1.83
0 = low risk	212	2.55%	215	2.58%	155	4.36%	152	4.27%	240	2.84%	240	2.84%
1 = moderate risk	566	6.80%	582	6.99%	395	11.10%	366	10.29%	668	7.91%	678	8.03%
2 = high risk	1,099	13.20%	1,100	13.21%	538	15.13%	564	15.86%	1,199	14.21%	1,212	14.36%
>2 = high risk	6,451	77.46%	6,431	77.22%	2,469	69.41%	2,475	69.58%	6,333	75.04%	6,310	74.76%
HAS-BLED score	2.81	1.31	2.80	1.30	2.56	1.28	2.58	1.29	2.75	1.30	2.74	1.29
0 = low risk	241	2.89%	251	3.01%	157	4.41%	163	4.58%	276	3.27%	263	3.12%
1–2 = moderate risk	3,331	40.00%	3,336	40.06%	1,632	45.88%	1,622	45.60%	3,506	41.54%	3,549	42.05%
>2 = high risk	4,756	57.11%	4,741	56.93%	1,768	49.70%	1,772	49.82%	4,658	55.19%	4,628	54.83%
Baseline prior bleed	1,434	17.22%	1,450	17.41%	495	13.92%	524	14.73%	1,421	16.84%	1,408	16.68%
Baseline prior stroke	1,008	12.10%	1,010	12.13%	359	10.09%	348	9.78%	939	11.13%	939	11.13%
Hemodialysis	17	0.20%	65	0.78%	5	0.14%	0	0.00%	16	0.19%	6	0.07%
Thrombocytopenia	253	3.04%	269	3.23%	92	2.59%	76	2.14%	249	2.95%	218	2.58%
Congestive heart failure	2,289	27.49%	2,280	27.38%	815	22.91%	830	23.33%	2,222	26.33%	2,185	25.89%
Diabetes	2,776	33.33%	2,748	33.00%	1,171	32.92%	1,178	33.12%	2,756	32.65%	2,719	32.22%
Hypertension	7,111	85.39%	7,101	85.27%	2,994	84.17%	2,980	83.78%	7,166	84.91%	7,191	85.20%
Renal disease	2,085	25.04%	2,074	24.90%	652	18.33%	640	17.99%	1,960	23.22%	1,950	23.10%
Myocardial infraction	826	9.92%	853	10.24%	299	8.41%	295	8.29%	801	9.49%	788	9.34%
Dyspepsia or stomach discomfort	1,456	17.48%	1,512	18.16%	575	16.17%	569	16.00%	1,458	17.27%	1,429	16.93%
Peripheral vascular disease	3,849	46.22%	3,853	46.27%	1,529	42.99%	1,547	43.49%	3,836	45.45%	3,782	44.81%
Transient ischemic attack	623	7.48%	585	7.02%	230	6.47%	241	6.78%	590	6.99%	587	6.95%
Coronary artery disease	3,293	39.54%	3,284	39.43%	1,336	37.56%	1,328	37.33%	3,277	38.83%	3,244	38.44%
Baseline medication use
Angiotensin converting enzyme inhibitor	3,268	39.24%	3,260	39.15%	1,347	37.87%	1,352	38.01%	3,255	38.57%	3,231	38.28%
Amiodarone	979	11.76%	965	11.59%	386	10.85%	405	11.39%	981	11.62%	958	11.35%
Angiotensin receptor blocker	1,947	23.38%	1,950	23.41%	778	21.87%	822	23.11%	2,019	23.92%	2,022	23.96%
Beta blockers	4,979	59.79%	4,952	59.46%	2,064	58.03%	2,053	57.72%	5,080	60.19%	5,079	60.18%
H2-receptor antagonist	387	4.65%	382	4.59%	138	3.88%	158	4.44%	379	4.49%	362	4.29%
Proton pump inhibitor	2,285	27.44%	2,343	28.13%	879	24.71%	899	25.27%	2,306	27.32%	2,298	27.23%
Anti-platelets	1,218	14.63%	1,188	14.27%	463	13.02%	466	13.10%	1,227	14.54%	1,235	14.63%
Statins	4,826	57.95%	4,834	58.05%	1,882	52.91%	1,884	52.97%	4,840	57.35%	4,880	57.82%
Inpatient admission	3,668	44.04%	3,679	44.18%	1,377	38.71%	1,421	39.95%	3,649	43.23%	3,618	42.87%

Outcomes

The primary outcomes were first all-cause hospitalization, hospitalization due to stroke/SE, hospitalization due to major bleeding-related conditions, and healthcare costs (including all-cause healthcare, all-cause medical, all-cause pharmacy, all-cause hospitalization, all-cause emergency room [ER]/outpatient, stroke/SE-related medical, and major bleeding-related medical costs). Stroke/SE and major bleeding events were identified using hospital claims that had a stroke/SE or major bleeding code as the first listed ICD-9-CM diagnosis code (Supplemental Table 2). The ICD-9 codes used for stroke/SE and major bleeding were based on validated administrative-claim-based algorithms as well as clinical trial definitions^12,37,38. The hospitalization outcomes were measured independently.

Table 2. Comparisons of all-cause healthcare costs, all-cause medical, pharmacy, hospitalization, and ER/outpatient medical costs after propensity score matching.

All-cause costs PPPM	Apixaban cohort (n = 8,328)	Warfarin cohort (n = 8,328)		Apixaban cohort (n = 3,557)	Dabigatran cohort (n = 3,557)		Apixaban cohort (n = 8,440)	Rivaroxaban cohort (n = 8,440)
	Marginal effect	Marginal effect	p-value	Marginal effect	Marginal effect	p-value	Marginal effect	Marginal effect	p-value
ER/outpatient medical costs	$1,935	$2,312	.014	$1,928	$1,875	.775	$1,948	$1,948	.999
Hospitalization medical costs	$1,010	$1,531	<.001	$1,031	$1,102	.705	$999	$1,237	.044
Medical costs	$2,945	$3,843	<.001	$2,958	$2,977	.942	$2,947	$3,185	.184
Pharmacy costs	$463	$318	<.001	$440	$416	.125	$460	$436	.082
Health care costs	$3,408	$4,161	<.001	$3,399	$3,394	.985	$3,407	$3,621	.235

* All-cause medical costs include all-cause ER/outpatient and hospitalization medical costs; all-cause healthcare costs include all-cause medical and pharmacy costs.

Stroke/SE-related medical costs were defined as the hospitalization costs associated with the first stroke/SE event plus all subsequent stroke/SE costs occurring in the inpatient or outpatient setting. Major bleeding-related medical costs were defined as the hospitalization costs associated with the first major bleed plus all subsequent major bleeding costs that occurred in the inpatient or outpatient setting. All-cause medical costs represent the sum of costs for ER/outpatient and inpatient costs, and total all-cause healthcare costs represent the sum of medical and pharmacy costs. All cost outcomes were measured per patient per month (PPPM) and adjusted to 2015 US dollars using the Consumer Price Index for Medical Care Services³⁹.

Covariates

Demographics, clinical characteristics, and healthcare resource utilization were assessed during the 12-month baseline period and were used to identify differences among the treatment cohorts. Demographic characteristics included age, gender, and US geographic region. Clinical characteristics included the Charlson comorbidity index (CCI) score, CHA₂DS₂ score (congestive heart failure, hypertension, aged ≥75 years, diabetes mellitus, prior stroke or transient ischemic attack or thromboembolism), CHA₂DS₂-VASc score (CHA₂DS₂, vascular disease, aged 65–74 years, sex category), HAS-BLED score (hypertension, abnormal renal and liver function, stroke, bleeding, labile international normalized ratios, elderly, drugs and alcohol), comorbid conditions, and co-medication use. The HAS-BLED bleeding risk score was based on evidence of hypertension, abnormal kidney or liver function, stroke, bleeding, aged >65 years, and drugs/alcohol abuse or dependence⁴⁰. The CHA₂DS₂-VASc stroke risk score was calculated using ICD-9 codes in the claims data as the summed total of the points determined for each diagnosis or characteristic⁴¹. Scores ≥3 and ≥2 were associated with a high bleeding and stroke risk for HAS-BLED and CHA₂DS₂-VASc, respectively⁴².

Statistical analysis

For each study cohort, descriptive analyses were conducted for demographics, comorbidities, and outcomes. Means and standard deviations were reported for continuous variables, and Student’s t-tests were used to detect differences. Percentages were reported for categorical variables, and Chi-square tests were used to detect differences. Differences were considered statistically significant if the p-value was <.05.

The incidence rate of hospitalizations (all-cause, stroke/SE-related, and major bleeding-related) was calculated using a person-time approach: the number of patients with a first episode of an event divided by the 100 person-years (PY) exposure time at risk.

Propensity score matching (PSM) was used between apixaban and the other OACs to create separate matched populations (warfarin-apixaban, rivaroxaban-apixaban, and dabigatran-apixaban). Nearest neighbor without replacement with a caliper of 0.01 was used to match the patients⁴³. Patients were matched 1:1 on the propensity scores generated by multivariable logistic regressions, which were based on age, gender, geographic region, CCI score, CHA₂DS₂-VASc score, HAS-BLED score, prior bleed and stroke, comorbidities, baseline co-medications, and baseline hospitalization. The variables were chosen given their clinical importance to the outcomes measures, which helps reduce the bias in the matching process⁴⁴. The balance of covariates was checked based on standardized differences with a threshold of 10%⁴⁵.

In each PSM population, we used multivariate Cox proportional hazards models to identify the associations between OAC treatment and hospitalizations (all-cause, stroke/SE-related, and major bleeding-related)⁴². The baseline characteristics were balanced in the matched populations; therefore, only treatment status was included as the independent variable. We calculated the hazard ratio (HR) and 95% confidence interval (CI) for each outcome of interest.

The relationship between the study cohort and the all-cause healthcare costs was evaluated using generalized linear models with log-link and a gamma distribution, wherein the dependent variable was cost⁴⁶. Additionally, 2-part models with bootstrapping were used in the analysis of stroke/SE- and major bleeding-related medical costs, given the high proportion of cost fields with 0 values. The primary independent variable was the treatment cohort, with the apixaban cohort as the reference cohort⁴⁷. The cost and difference, 95% CI, and p-value for each cost model were reported.

Sub-group and sensitivity analyses

Sub-group analyses were conducted on specific insurance types (Medicare Advantage and commercial) and dosages (standard and low dose). The balance of baseline characteristics was tested in each sub-group, and when an imbalance was detected (standardized difference >10%), the variable was included in the multivariate model. The risks of hospitalization (all-cause, stroke/SE-related, and major bleeding-related) were compared among the study cohorts, and the statistical significance of the interaction between treatments and sub-groups was evaluated.

Two sensitivity analyses were conducted. First, we censored patients at 6 months. Because apixaban patients had a shorter follow-up time due to the drug’s recent market entry, this sensitivity analysis helped create a more balanced follow-up between treatment groups. Second, we used an intent-to-treat method to extend the follow-up period, following patients from the index date until the study outcome, health plan disenrollment, inpatient death, or end of study period.

Results

Of the 47,634 patients who were eligible for the analysis before PSM, 21,135 (44.4%) were prescribed warfarin, 8,652 (18.2%) were prescribed apixaban, 14,163 (29.7%) were prescribed rivaroxaban, and 3,684 (7.7%) were prescribed dabigatran. After 1:1 PSM, there were 8,328 warfarin-apixaban matched pairs, 3,557 dabigatran-apixaban matched pairs, and 8,440 rivaroxaban-apixaban matched pairs (Figure 1).

Figure 1. Patient selection criteria.

Baseline characteristics

After PSM, each of the three matched cohorts was well balanced in terms of baseline demographics and clinical characteristics. Warfarin-apixaban patients were the oldest on average (aged 73.5 years), followed by rivaroxaban-apixaban (aged 72.8 years) and dabigatran-apixaban patients (aged 70.9 years). The warfarin-apixaban matched cohort also had the highest CCI and CHA₂DS₂-VASc scores among the three cohorts, followed by the rivaroxaban-apixaban and dabigatran-apixaban cohorts (CCI: 2.4, 2.3, 2.0, and CHA₂DS₂-VASc: 3.9, 3.8, 3.5, respectively). Further, the warfarin-apixaban cohort had the highest percentages of prior bleeding and stroke events, followed by the rivaroxaban-apixaban and dabigatran-apixaban cohorts (Table 1).

Hospitalizations—all-cause, stroke, and major bleeding

After PSM, warfarin (HR = 1.30; 95% CI = 1.21–1.40, p < .001) and rivaroxaban (HR = 1.15; 95% CI = 1.07–1.24, p < .001) patients were more likely to have an all-cause hospitalization compared to apixaban patients. Dabigatran patients (HR = 1.11; 95% CI = 0.99–1.25, p = .083) trended towards a higher risk of hospitalization (Figure 2).

Figure 2. Propensity score matched incidence rates and hazard ratios for all-cause hospitalization among apixaban patients matched to warfarin, dabigatran, and rivaroxaban patients.

Warfarin patients (HR = 1.60; 95% CI = 1.23–2.07, p < .001) were associated with a significantly higher risk of hospitalization due to stroke/SE compared to matched apixaban patients. Dabigatran (HR = 1.25; 95% CI = 0.78–2.00, p = .365) and rivaroxaban (HR = 1.18, 95% CI = 0.89–1.57, p = .246) patients were associated with a non-significantly higher risk of hospitalization due to stroke/SE as compared to apixaban patients (Figure 3).

Figure 3. Propensity score matched incidence rates and hazard ratios for hospitalization due to stroke/SE among apixaban patients matched to warfarin, dabigatran, and rivaroxaban patients.

Rivaroxaban (HR = 1.71, 95% CI = 1.39–2.10, p < .001), dabigatran (HR = 1.46, 95% CI = 1.02–2.10, p = .039), and warfarin (HR = 1.95; 95% CI = 1.60–2.39, p < .001) patients were significantly more likely to have a hospitalization due to major bleeding as compared to apixaban patients (Figure 4).

Figure 4. Propensity score matched incidence rates and hazard ratios for hospitalization due to major bleeding among apixaban patients matched to warfarin, dabigatran, and rivaroxaban patients.

Healthcare costs

Warfarin patients were associated with significantly higher all-cause total healthcare costs PPPM compared to apixaban patients ($4,161 vs $3,408, p < .001). Additionally, warfarin patients had significantly higher all-cause medical costs PPPM ($3,843 vs $2,945, p < .001). Inpatient and office visit costs were the main drivers for all-cause healthcare costs. Dabigatran ($3,394 vs $3,399, p = .985) and rivaroxaban ($3,621 vs $3,407, p = .235) patients had similar total all-cause healthcare costs compared to apixaban patients (Table 2).

Warfarin ($137 vs $94; p = .280), rivaroxaban ($134 vs $96, p = .377), and dabigatran ($109 vs $87, p = .679) patients had similar stroke/SE-related medical costs compared to apixaban patients (Figure 5). Rivaroxaban ($209 vs $107, p = .013) and warfarin ($258 vs $105, p = .002) patients had significantly higher major bleeding-related medical costs PPPM compared to matched apixaban patients (Figure 5).

Figure 5. Comparisons of stroke/SE- and major bleeding-related medical costs after propensity score matching.

Sub-group and sensitivity analyses results

Among the matched population, 73.3% (12,209) of the warfarin-apixaban, 63.0% (4,479) of the dabigatran-apixaban, and 67.9% (11,469) of the rivaroxaban-apixaban patients had Medicare Advantage insurance; also, 79.8% (13,294) of the warfarin-apixaban, 85.7% (6,099) of the dabigatran-apixaban, and 77.7% (13,120) of the rivaroxaban-apixaban patients had standard dose treatment. Similar trends for hospitalization risks due to stroke/SE and major bleeding were found for the sub-group analyses by insurance type and dosage (Supplemental Table 3). There were no significant interactions found between insurance type and stroke/SE or major bleeding. Among apixaban and dabigatran patients, a significant interaction was found for dose and hospitalization due to stroke/SE (p = .018). Standard-dose dabigatran was associated with a higher risk of hospitalization due to stroke/SE (HR = 1.78; 95% CI = 1.01–3.14) compared to standard-dose apixaban. Low-dose dabigatran was associated with a similar risk of hospitalization due to stroke/SE (HR = 0.43; 95% CI = 0.15–1.22) compared to low-dose apixaban. For the sensitivity analyses, the results were generally consistent with those of the main analysis. However, there was no significant difference in hospitalization due to major bleeding between dabigatran and apixaban users when censored at 6 months (HR = 1.33; 95% CI = 0.88–2.00, p = .174) and when the intent-to-treat approach was used (HR = 1.20; 95% CI = 0.93–1.54, p = .157) (Supplemental Table 4).

Discussion

Using a claims database representing a US health plan, we found that, compared to apixaban patients, warfarin patients were associated with a higher risk of all-cause, stroke/SE-related, and major bleeding-related hospitalization. Rivaroxaban patients were associated with a significantly higher risk of all-cause and major bleeding-related hospitalization as well as a numerically higher risk of hospitalization due to stroke/SE. Dabigatran patients were associated with a significantly higher risk of hospitalization due to major bleeding as well as a numerically higher risk of all-cause and stroke/SE-related hospitalization. This resulted in significantly higher major bleeding-related medical costs among warfarin and rivaroxaban patients, as well as numerically higher major bleeding-related medical costs among dabigatran patients. Potentially due to the limited follow-up length and small sample size, the stroke/SE-related medical costs were numerically higher but not statistically significant among warfarin, dabigatran, and rivaroxaban patients compared to apixaban patients. Warfarin was associated with significantly higher all-cause total medical and total (outpatient, inpatient, and pharmacy) healthcare costs. Dabigatran and rivaroxaban showed similar total all-cause healthcare costs compared to apixaban.

Our hospitalization results were consistent with published clinical trial data, showing apixaban being superior to warfarin in preventing stroke/SE-related hospitalization and reducing major bleeding-related hospitalization risk^12,48. In the ARISTOTLE trial, apixaban demonstrated a significantly lower risk of major bleeding (HR = 0.69, 95% CI = 0.60–0.80, p < .001) and stroke/SE (HR = 0.79, 95% CI = 0.66–0.95, p = .01)¹². There are currently no DOAC head-to-head clinical trials available.

Our study showed similar results with the real-world Noseworthy et al.²⁵ study comparing apixaban, dabigatran, and rivaroxaban, which showed that apixaban was associated with a lower risk of major bleeding compared to dabigatran and rivaroxaban, and no significant difference of stroke/SE was found among the three DOACs. In our analysis, apixaban had a similar risk of hospitalization due to stroke/SE and lower risk of hospitalization due to major bleeding compared to the other DOACs. Other real-world studies comparing apixaban and warfarin have found that warfarin patients have a higher risk of stroke and major bleeding compared to apixaban patients^{17,18,22,23,49}.

Sensitivity analysis specifically in the Medicare Advantage and commercial insurance enrollees validated our main result: apixaban had lower risks of stroke/SE and major bleeding compared to the other OACs, regardless of coverage type. In general, the Medicare Advantage population did have a higher risk of stroke/SE and major bleeding than the commercial insured population.

Due to its recent market entry, apixaban has not been compared to dabigatran, rivaroxaban, and warfarin in published real-world economic studies evaluating costs in the inpatient and outpatient setting. Recent cost-effectiveness studies with Markov models, using data from clinical trials, have been developed to compare treatment costs between DOACs and warfarin. Numerous studies have reported that all DOACs are cost-effective alternatives to warfarin, with apixaban providing the greatest monetary and quality-adjusted life year (QALY) value compared to dabigatran, rivaroxaban, and warfarin^50,51. For example, the Canestaro et al.⁵¹ study, based on data from the RE-LY, ROCKET-AF, and ARISTOTLE trials using the Markov model, showed that apixaban, rivaroxaban, and dabigatran cost $93,063, $111,465, and $140,557 per additional QALY gained, respectively, compared with warfarin. Another study using the Markov decision model based on the data from the ARISTOTLE trial concluded that apixaban was cost-effective relative to warfarin for secondary stroke prevention, with an incremental cost-effectiveness ratio of $11,400 per QALY⁵². Although deterministic results might not reflect real-world situations, the comparisons provided insight on the cost-effectiveness between OACs.

Additionally, when retrospective claims data were used in the Amin et al.⁵³ study to predict medical costs, patients treated with apixaban instead of warfarin had medical cost reductions of $493 for stroke, $752 for major bleeding (excluding intracranial hemorrhage), and $1,245 for the combined outcome of both events. Vaughan Sarrazin et al.⁵⁴ reported that the per patient-year inpatient hospital costs for stroke were significantly lower for dabigatran ($141, p = .007) and rivaroxaban ($129, p = .002) compared to warfarin ($192) among Medicare beneficiaries. Dabigatran ($246, p = .23) had similar—and rivaroxaban ($270, p = .035) had significantly lower—gastrointestinal hemorrhage inpatient hospital costs compared to warfarin ($229)⁵⁴. Deitelzweig et al.²⁴ compared the all-cause hospitalization re-admission costs of DOACs and found significantly higher costs for rivaroxaban compared to apixaban (difference = $413, p = .003) and numerically higher costs for dabigatran vs apixaban ($142, p = .31) per patient. In addition to hospitalization costs published in previous studies, our study also looked at direct healthcare costs in other settings.

Limitations

This study has several limitations. Because of its observational design, the study demonstrates associations but does not establish causality. Additionally, limitations inherent to administrative claims data apply to this study. There are no head-to-head clinical trials comparing the efficacy and safety of any of the DOACs for reducing the risk of stroke/SE in patients with NVAF. The cost calculation does not imply comparable efficacy, safety, or product interchangeability. A claim for a prescribed medication does not necessarily mean the medication was taken as prescribed; therefore, we cannot measure compliance. Although we could not account for potential residual confounders such as compliance, AF duration, and over-the-counter aspirin use, we could account for differences in the observed demographic and clinical characteristics and adjust them in PSM. Although, as mentioned in King and Nielsen⁴⁴, PSM may not be the best option, where the inferences are degraded by the matching process, the large sample size and observational nature of our study help reduce the bias and generate a more justified matching]. The pre-index period was 12 months; therefore, a patient may not have been treatment-naïve and may have been prescribed an OAC prior to the baseline period. The follow-up time was not uniform, nor was it consistent with the clinical trials. The sensitivity analysis with patients censored at 6 months was conducted to address the issue of imbalanced follow-up time. After performing the sensitivity analysis, the major bleeding and stroke/SE results were generally consistent with those in the main analysis. As is the case for claims-based studies, there may be coding errors or diagnoses entered for administrative processing rather than clinical completeness. Although the OptumInsight Research Database includes patients from across the US, the results from this health plan may not be generalizable to the entire population of NVAF patients. Due to the short follow-up length and the limited number of the stroke/SE events, the interpretation of stroke/SE-related outcomes should be carefully considered. Finally, we focused on treatment-naïve patients; therefore, major bleeding and stroke/SE risks may be different for patients previously treated with warfarin who switched to a DOAC.

Conclusions

In this real-world study of NVAF patients, the use of warfarin was associated with a significantly higher risk of hospitalization due to stroke/SE and major bleeding compared to apixaban, which is consistent with clinical trial data. Compared to apixaban, rivaroxaban and dabigatran were associated with a similar risk of hospitalization due to stroke/SE and a significantly higher risk of hospitalization due to major bleeding. In addition, rivaroxaban and warfarin patients were associated with a significantly higher risk of all-cause hospitalization compared to apixaban patients. Furthermore, there were higher major bleeding-related medical costs for rivaroxaban and warfarin patients. Warfarin patients also had significantly higher all-cause healthcare costs compared to apixaban patients. This study may assist clinicians in determining the appropriate OAC for treatment-naïve NVAF patients and could be informative to formulary decision-makers managing commercial and Medicare populations.

Transparency

Declaration of funding

This work was funded by Pfizer, Inc. and Bristol-Myers Squibb.

Declaration of financial relationships

AA is an employee of the University of California, Irvine, and was a paid consultant to Bristol-Myers Squibb. AA has also served as a consultant and/or speaker for Bristol-Myers Squibb, Pfizer, Boehringer-Ingelheim, and Portola. AK and QZ are employees of STATinMED Research, a paid consultant to Pfizer in connection with this study and the development of this manuscript. KO, OD, and JT are employees of Pfizer, Inc. LV and CP are employees of Bristol-Myers Squibb.

Acknowledgments

Christopher Haddlesey of STATinMED Research provided editorial assistance in the preparation of this manuscript. Juan Du of STATinMED Research provided statistical support for this manuscript.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.