Comparison of effectiveness and safety of treatment with apixaban vs. other oral anticoagulants among elderly nonvalvular atrial fibrillation patients

Abstract

Objective: To compare the risk of stroke/systemic embolism (S/SE) and major bleeding (MB) of elderly (≥65 years of age) nonvalvular atrial fibrillation (NVAF) patients initiating apixaban vs. rivaroxaban, dabigatran, or warfarin.

Methods: NVAF patients with Medicare Advantage coverage in the US initiating oral anticoagulants (OACs, index event) were identified from the Humana database (1 January 2013–30 September 2015) and grouped into cohorts depending on OAC initiated. Propensity score matching (PSM), 1:1, was conducted among patients treated with apixaban vs. each other OAC, separately. Rates of S/SE and MB were evaluated in the follow-up. Cox regressions were used to compare the risk of S/SE and MB between apixaban and each of the other OACs during the follow-up.

Results: The matched pairs of apixaban vs. rivaroxaban (n = 13,620), apixaban vs. dabigatran (n = 4654), and apixaban vs. warfarin (n = 14,214) were well balanced for key patient characteristics. Adjusted risks for S/SE (hazard ratio [HR] vs. rivaroxaban: 0.72, p = .003; vs. warfarin: 0.65, p < .001) and MB (HR vs. rivaroxaban: 0.49, p < .001; vs. warfarin: 0.53, p < .001) were significantly lower during the follow-up for patients treated with apixaban vs. rivaroxaban and warfarin. Adjusted risks for S/SE (HR: 0.78, p = .27) and MB (HR: 0.82, p = .23) of NVAF patients treated with apixaban vs. dabigatran trended to be lower, but did not reach statistical significance.

Conclusions: In the real-world setting after controlling for differences in patient characteristics, apixaban is associated with significantly lower risk of S/SE and MB than rivaroxaban and warfarin, and a trend towards better outcomes vs. dabigatran among elderly NVAF patients in the US.

Keywords:

• Elderly
• nonvalvular atrial fibrillation
• oral anticoagulants
• apixaban
• rivaroxaban
• dabigatran
• warfarin
• effectiveness
• safety

View correction statement:

Correction to: Steven Deitelzweig et al., Comparison of effectiveness and safety of treatment with apixaban vs. other oral anticoagulants among elderly nonvalvular atrial fibrillation patients

Introduction

Nonvalvular atrial fibrillation (NVAF) is a common cardiac rhythm disorder and has long been identified as a significant risk factor for disabling or fatal ischemic stroke and systemic embolism (SE)^1,2. The prevalence of NVAF increases with age as does NVAF-related stroke risk^1,3. In 2010 it was estimated that AF affected 5.2 million persons in the United States⁴. With the increasing incidence of AF and the growing number of the elderly population, the prevalence of AF is estimated to increase to 12.1 million by 2030⁴.

For more than 50 years, warfarin has been used to prevent the risk of stroke among patients with NVAF, although its use has been associated with several limitations^5,6. In recent years, direct oral anticoagulants (DOACs) have become available in the US market providing alternative options to warfarin for anticoagulation therapy among NVAF patients. The DOACs, apixaban, rivaroxaban, and dabigatran, have all been shown in large randomized clinical trials to be superior or noninferior to warfarin for the prevention of stroke, as well as to have lower or similar rates of bleeding among patients with NVAF^7–9. Furthermore, the DOACs do not require frequent monitoring and lack many of the other limitations associated with warfarin therapy¹⁰.

Although some observational studies have shown that the effectiveness and safety of apixaban, rivaroxaban, and dabigatran relative to warfarin are similar to that reported in clinical trials^11–13, there are no direct comparisons of the DOACs in randomized controlled trials, and only a few in the real-world setting^14–17. This is particularly true for apixaban, which entered the market later than rivaroxaban and dabigatran. Further real-world assessment of apixaban is warranted, especially among the older population who are at higher risk of stroke and bleeding^2,18. The objectives of this study were to evaluate and compare the risk of stroke/systemic embolism (S/SE) and major bleeding (MB) among oral anticoagulant (OAC) treatment naïve elderly NVAF patients in a US population who initiated treatment with apixaban vs. rivaroxaban, dabigatran, and warfarin.

Patients and methods

Study design and data source

This study was an observational, retrospective, cohort analysis conducted using the Humana Research Database. The database comprises claims from more than 20 million members with Medicare Advantage coverage in the US. The database is an integrated source of managed care medical and pharmacy claims and eligibility files containing information on diagnostic and therapeutic services rendered in both inpatient and outpatient settings, prescription drugs dispensed, and demographic characteristics.

Study population

AF patients with Medicare coverage and an age of ≥65 years who had a pharmacy claim of apixaban, rivaroxaban, dabigatran, or warfarin between 1 January 2013 and 30 September 2015 were identified from the data source. Edoxaban was not included in this analysis as its FDA approval was not long before the end of the study period and the patient count was too low for this study. AF patients were identified using ICD-9-CM diagnosis code 427.31, which has been validated to identify AF patients with a median positive predictive value of 89%¹⁹. The date of the first pharmacy claim for the specific OAC was defined as index date. Patients were required to have an AF diagnosis within 12 months prior to or on the index date (baseline period), and to have 12 months of continuous health plan enrollment. Patients with claims indicative of diagnoses of valvular heart disease, venous thromboembolism, transient AF, cardiac surgery, hyperthyroidism and thyrotoxicity, or pregnancy during the baseline period were excluded. To ensure patients were OAC treatment naïve, patients were excluded if they had a pharmacy claim for warfarin, apixaban, dabigatran, rivaroxaban, or edoxaban during the baseline period. Patients with claims for >1 type of OAC on the index date were also excluded. Eligible patients were assigned to the apixaban, rivaroxaban, dabigatran, or warfarin cohort based on their newly initiated OAC. The end of the follow-up for patients within the study cohorts was defined as the earliest of the following dates: 90 days after index OAC treatment discontinuation (the end of drug days of supply), date a patient switched from the index OAC treatment to another OAC, date of health plan disenrollment, or 30 September 2015 (end of study period). The selection of patients for study cohorts treated with apixaban, rivaroxaban, dabigatran, and warfarin are shown in Figures 1–3.

Figure 1. Selection of patients for study cohorts treated with apixaban and rivaroxaban. AF: atrial fibrillation; NVAF: nonvalvular atrial fibrillation; PSM: propensity score matching.

Figure 2. Selection of patients for study cohorts treated with apixaban and dabigatran. AF: atrial fibrillation; NVAF: nonvalvular atrial fibrillation; PSM: propensity score matching.

Figure 3. Selection of patients for study cohorts treated with apixaban and warfarin. AF: atrial fibrillation; NVAF: nonvalvular atrial fibrillation; PSM: propensity score matching.

Propensity score matching

Multivariate logistic regression was used to generate propensity scores with covariates including patient characteristics of age, gender, race, US geographic region, Charlson Comorbidity Index (CCI) score, CHA₂DS₂-VASc score, HAS-BLED score, follow-up period duration, baseline total healthcare cost, baseline bleeding-related medical cost, baseline stroke-related medical cost, baseline comorbidities (thrombocytopenia, congestive heart failure, diabetes, hypertension, renal disease, myocardial infarction, dyspepsia/stomach discomfort, peripheral vascular disease, transient ischemic attack, coronary artery disease), and baseline medication usage (angiotensin-converting enzyme inhibitors, amiodarone, angiotensin receptor blockers, beta blockers, H2-receptor antagonists, proton pump inhibitors, statins, anti-platelet drugs). Separate 1:1 propensity score matching (PSM) was conducted (using the nearest neighbor algorithm) among patients treated with apixaban and rivaroxaban, those treated with apixaban and dabigatran, and those treated with apixaban and warfarin. The matched apixaban vs. rivaroxaban, apixaban vs. dabigatran, and apixaban vs. warfarin cohorts were inspected to verify that cohorts were well balanced with key patient characteristics not statistically different (p > .05).

Demographics and clinical characteristics

Demographic information and clinical characteristics, including CCI score, CHADS₂ score, CHA₂DS₂-VASc score, HAS-BLED score, and prior bleeding and stroke diagnoses during the 12 month baseline period were determined for each patient in the study cohorts prior to and after PSM.

Outcome measurements

During the follow-up, S/SE and MB events were evaluated, based on hospitalization claims with the corresponding ICD-9-CM codes at the first position among the diagnosis codes associated with any of the inpatient claims. The diagnosis codes used for S/SE and MB were based on a validated administrative-claim-based algorithm as well as the International Society on Thrombosis and Haemostasis’ definition of MB as used in the ARISTOTLE trial^7,20,21. The clinical outcomes evaluated included rates of S/SE (any S/SE, with breakdown into following subcategories: ischemic stroke, hemorrhagic stroke, and SE) and MB (any MB with breakdown into the following subcategories; intracranial hemorrhage [ICH], gastrointestinal [GI] MB, and other MBs). One patient may have had more than one subcategories of S/SE or MB and the categories are not mutually exclusive.

Statistical analyses

Descriptive statistics were used to evaluate differences between study cohorts (apixaban vs. rivaroxaban, apixaban vs. dabigatran, and apixaban vs. warfarin) in demographics and clinical characteristics prior to and after PSM; t-tests and chi-square tests were used for the continuous and categorical variables, respectively. Propensity scores were generated using a multivariate logistic regression. Among the propensity score matched populations, Cox proportional hazards models were further used to assess the relationship between anticoagulation treatment (apixaban vs. rivaroxaban, apixaban vs. dabigatran, and apixaban vs. warfarin) and risks for S/SE and MB during the follow-up. No other covariates were used in such regression analyses after the PSM, since patient cohorts had balanced distribution for other covariates. Kaplan Meier analysis was used to illustrate time to events for any S/SE and any MB among matched cohorts. The threshold of p < .05 was used to determine statistical significance. Statistical software SAS version 9.4 (Cary, NC, USA) was utilized to perform all data analyses.

Results

Unmatched study cohorts

Table 1 shows the baseline demographics and clinical characteristics prior to matching of study cohorts of patients treated with apixaban and other OACs. Of the unmatched study population, 8250 received apixaban, 11,082 received rivaroxaban, 2474 received dabigatran, and 14,051 received warfarin.

Table 1. Baseline demographics and clinical characteristics of study cohorts treated with apixaban vs. other oral anticoagulants prior to matching.

	Apixaban N = 8250		Rivaroxaban N = 11,082		p-value	Dabigatran N = 2474		p-value	Warfarin N = 14,051		p-value
Age (years)
Mean (SD )	78.0 (9.0)		77.2 (8.6)		<.001	76.8 (8.3)		<.001	78.2 (9.0)		.03
Median	76		76			75			77
Gender	N	%	N	%	.17	N	%	.002	N	%	<.001
Female	4001	48.5	5265	47.5		1110	44.9		6294	44.8
Male	4249	51.5	5817	52.5		1364	55.1		7757	55.2
US geographic region	N	%	N	%	<.001	N	%	.002	N	%	<.001
South	5707	69.2	7266	65.6		1610	65.1		7807	55.6
Midwest	1582	19.2	2438	22.0		537	21.7		4168	29.7
West	775	9.4	1103	10.0		265	10.7		1640	11.7
Northeast	186	2.3	275	2.5		62	2.5		436	3.1
Race	N	%	N	%	.75	N	%	.82	N	%	<.001
White	7385	89.5	9902	89.4		2230	90.1		12,507	89.0
Black	481	5.8	631	5.7		136	5.5		977	7.0
Other	192	2.3	282	2.5		56	2.3		329	2.3
Unknown	192	2.3	267	2.4		52	2.1		238	1.7
Duration of follow-up (months)
Mean (SD)	6.3 (5.2)		7.4 (6.3)		<.001	7.6 (6.4)		<.001	8.3 (6.8)		<.001
Median	5		5			6			6
Charlson Comorbidity Index score
Mean (SD)	3.0 (2.5)		2.8 (2.4)		<.001	2.7 (2.4)		<.001	3.4 (2.6)		<.001
Median	3		2			2			3
Charlson Comorbidity Index score group	N	%	N	%	<.001	N	%	<.001	N	%	<.001
0	1197	14.5	1832	16.5		421	17.0		1618	11.5
1–2	2877	34.9	4033	36.4		950	38.4		4408	31.4
3–4	2169	26.3	2800	25.3		614	24.8		3786	26.9
≥5	2007	24.3	2417	21.8		489	19.8		4239	30.2
CHADS2 score								<.001
Mean (SD)	2.7 (1.4)		2.6 (1.3)		<.001	2.6 (1.3)			2.9 (1.4)		<.001
Median	3		2			2			3
CHADS2 score group	N	%	N	%	<.001	N	%	.003	N	%	<.001
0	220	2.7	342	3.1		78	3.2		285	2.0
1–2	3842	46.6	5496	49.6		1237	50.0		5800	41.3
3–4	3257	39.5	4163	37.6		924	37.4		6108	43.5
5–6	931	11.3	1081	9.8		235	9.5		1858	13.2
CHA2DS2-VASc score
Mean (SD)	4.6 (1.6)		4.4 (1.6)		<.001	4.3 (1.6)		<.001	4.7 (1.6)		<.001
Median	4		4			4			5
CHA2DS2-VASc score group	N	%	N	%	<.001	N	%	<.001	N	%	<.001
1–2	725	8.8	1185	10.7		281	11.4		995	7.1
3–4	3570	43.3	4916	44.4		1135	45.9		5577	39.7
5–6	2868	34.8	3677	33.2		801	32.4		5342	38.0
≥7	1087	13.2	1304	11.8		257	10.4		2137	15.2
HAS-BLED score
Mean (SD)	3.1 (1.2)		3.0 (1.1)		<.001	2.9 (1.1)		<.001	3.2 (1.2)		<.001
Median	3		3			3			3
HAS-BLED score group	N	%	N	%	.03	N	%	<.001	N	%	<.001
0–2	3023	36.6	4232	38.2		1044	42.2		4435	31.6
≥3	5227	63.4	6850	61.8		1430	57.8		9616	68.4
Baseline conditions	N	%	N	%		N	%		N	%
Prior stroke	970	11.8	1277	11.5	.62	269	10.9	.23	2216	15.8	<.001
Prior bleeding	1561	18.9	2269	20.5	.007	431	17.4	.09	3375	24.0	<.001
Index DOAC dose
Apixaban 2.5 mg	1955	23.7	–	–		–	–		–	–
Apixaban 5 mg	6295	76.3	–	–		–	–		–	–
Rivaroxaban 10 mg	–	–	987	8.9		–	–		–	–
Rivaroxaban 15 mg	–	–	2565	23.2		–	–		–	–
Rivaroxaban 20 mg	–	–	7530	68.0		–	–		–	–
Dabigatran 150 mg	–	–	–	–		2048	82.8		–	–
Dabigatran 75 mg	–	–	–	–		426	17.2		–	–

p-values are for comparisons between the apixaban cohort vs. other each oral anticoagulant cohort.

SD: standard deviation.

Apixaban vs. rivaroxaban

Relative to those treated with rivaroxaban, NVAF patients treated with apixaban had a higher mean age (78.0 vs. 77.2 years, p < .001), CCI score (3.0 vs. 2.8, p < .001), CHADS₂ score (2.7 vs. 2.6, p < .001), CHA₂DS₂-VASc score (4.6 vs. 4.4, p < .001), and HAS-BLED score (3.1 vs. 3.0, p < .001) at baseline. The proportion of patients with bleeding diagnoses during the baseline period was lower among those treated with apixaban vs. rivaroxaban (18.9% vs. 20.5%, p = .007). Additionally, the mean duration of follow-up was shorter for patients treated with apixaban in comparison to those treated with rivaroxaban (6.3 vs. 7.4 months, p < .001).

Apixaban vs. dabigatran

Relative to dabigatran, patients treated with apixaban also had a higher mean age (78.0 vs. 76.8 years, p < .001), CCI score (3.0 vs. 2.7, p < .001), CHADS₂ score (2.7 vs. 2.6, p < .001), CHA₂DS₂-VASc score (4.6 vs. 4.3, p < .001), and HAS-BLED score (3.1 vs. 2.9, p < .001). The mean duration of follow-up was shorter for patients treated with apixaban in comparison to those treated with dabigatran (6.3 vs. 7.6 months, p < .001).

Apixaban vs. warfarin

Unlike the other OAC comparisons, NVAF patients treated with apixaban vs. those treated with warfarin had a lower mean age (78.0 vs. 78.2 years, p = .03), CCI score (3.0 vs. 3.4, p < .001), CHADS₂ score (2.7 vs. 2.9, p < .001), CHA₂DS₂-VASc score (4.6 vs. 4.7, p < .001), and HAS-BLED score (3.1 vs. 3.2, p < .001). The proportions of patients with stroke (11.8% vs. 15.8%, p < .001) and bleeding (18.9% vs. 24.0%, p < .001) diagnoses during the baseline period were lower among those treated with apixaban vs. warfarin. As in other OAC comparisons, the mean duration of follow-up was shorter for patients treated with apixaban than those treated with warfarin (6.3 vs. 8.3 months, p < .001).

Propensity score matched study cohorts

Table 2 shows the baseline demographics and clinical characteristics after matching of study cohorts of patients treated with apixaban and other OACs.

Table 2. Baseline demographics and clinical characteristics of study cohorts treated with apixaban vs. other oral anticoagulants after matching.

	Apixaban N = 6810		Rivaroxaban N = 6810		Apixaban N = 2327		Dabigatran N = 2327		Apixaban N = 7107		Warfarin N = 7107
Age (years)
Mean (SD)	77.1 (8.0)		77.0 (7.8)		77.3 (9.0)		76.9 (8.3)		78.2 (9.1)		78.1 (8.8)
Median	76		76		76		75		77		77
Gender	N	%	N	%	N	%	N	%	N	%	N	%
Female	3255	47.8	3244	47.6	1053	45.3	1048	45.0	3367	47.4	3419	48.1
Male	3555	52.2	3566	52.4	1274	54.8	1279	55.0	3740	52.6	3688	51.9
US geographic region	N	%	N	%	N	%	N	%	N	%	N	%
South	4656	68.4	4665	68.5	1547	66.5	1523	65.5	4675	65.8	4717	66.4
Midwest	1329	19.5	1314	19.3	494	21.2	496	21.3	1514	21.3	1504	21.2
West	662	9.7	671	9.9	219	9.4	253	10.9	737	10.4	705	9.9
Northeast	163	2.4	160	2.4	67	2.9	55	2.4	181	2.6	181	2.6
Race	N	%	N	%	N	%	N	%	N	%	N	%
White	6109	89.7	6104	89.6	2096	90.1	2098	90.2	6373	89.7	6371	89.6
Black	396	5.8	403	5.9	136	5.8	128	5.5	440	6.2	436	6.1
Other	155	2.3	161	2.4	58	2.5	54	2.3	167	2.4	160	2.3
Duration of follow-up (months)
Mean (SD)	6.5 (5.1)		6.4 (5.1)		7.1 (5.5)		7.0 (5.5)		6.7 (5.3)		6.6 (5.4)
Median	5		5		6		5		5		5
Charlson Comorbidity Index score
Mean (SD)	2.8 (2.3)		2.7 (2.3)		2.6 (2.3)		2.6 (2.3)		3.0 (2.4)		3.0 (2.4)
Median	2		2		2		2		3		3
Charlson Comorbidity Index score group	N	%	N	%	N	%	N	%	N	%	N	%
0	1123	16.5	1089	16.0	401	17.2	409	17.6	1020	14.4	995	14.0
1–2	2538	37.3	2589	38.0	923	39.7	891	38.3	2461	34.6	2500	35.2
3–4	1745	25.6	1733	25.5	555	23.9	581	25.0	1878	26.4	1913	26.9
≥5	1404	20.6	1399	20.5	448	19.3	446	19.2	1748	24.6	1699	23.9
CHADS2 score
Mean (SD)	2.6 (1.3)		2.6 (1.3)		2.6 (1.3)		2.6 (1.3)		2.7 (1.4)		2.7 (1.3)
Median	2		2		2		2		3		3
CHADS2 score group	N	%	N	%	N	%	N	%	N	%	N	%
0	209	3.1	184	2.7	74	3.2	74	3.2	193	2.7	166	2.3
1–2	3428	50.3	3442	50.5	1169	50.2	1174	50.5	3276	46.1	3329	46.8
3–4	2571	37.8	2610	38.3	866	37.2	874	37.6	2843	40.0	2898	40.8
5–6	602	8.8	574	8.4	218	9.4	205	8.8	795	11.2	714	10.1
CHA2DS2-VASc score
Mean (SD)	4.4 (1.6)		4.4 (1.6)		4.3 (1.6)		4.3 (1.6)		4.6 (1.6)		4.6 (1.6)
Median	4		4		4		4		4		4
CHA2DS2-VASc score group	N	%	N	%	N	%	N	%	N	%	N	%
1–2	691	10.2	663	9.7	253	10.9	267	11.5	602	8.5	602	8.5
3–4	3156	46.3	3190	46.8	1076	46.2	1078	46.3	3107	43.7	3094	43.5
5–6	2287	33.6	2250	33.0	759	32.6	754	32.4	2495	35.1	2552	35.9
≥7	676	9.9	707	10.4	239	10.3	228	9.8	903	12.7	859	12.1
HAS-BLED score
Mean (SD)	2.9 (1.1)		2.9 (1.1)		2.9 (1.1)		2.9 (1.1)		3.0 (1.1)		3.1 (1.1)
Median	3		3		3		3		3		3
HAS-BLED score group	N	%	N	%	N	%	N	%	N	%	N	%
0–2	2750	40.4	2771	40.7	991	42.6	998	42.9	2608	36.7	2578	36.3
≥3	4060	59.6	4039	59.3	1336	57.4	1329	57.1	4499	63.3	4529	63.7
Baseline conditions	N	%	N	%	N	%	N	%	N	%	N	%
Prior stroke	684	10.0	670	9.8	231	9.9	238	10.2	842	11.9	834	11.7
Prior bleeding	1141	16.8	1131	16.6	382	16.4	377	16.2	1339	18.8	1350	19.0
Index DOAC dose
Apixaban 2.5 mg	1373	20.2	–	–	466	20.0	–	–	1679	23.6	–	–
Apixaban 5 mg	5437	79.8	–	–	1861	80.0	–	–	5428	76.4	–	–
Rivaroxaban 10 mg	–	–	546	8.0	–	–	–	–	–	–	–	–
Rivaroxaban 15 mg	–	–	1555	22.8	–	–	–	–	–	–	–	–
Rivaroxaban 20 mg	–	–	4709	69.2	–	–	–	–	–	–	–	–
Dabigatran 150 mg	–	–	–	–	–	–	1927	82.8	–	–	–	–
Dabigatran 75 mg	–	–	–	–	–	–	400	17.2	–	–	–	–

The matched apixaban vs. rivaroxaban, apixaban vs. dabigatran, and apixaban vs. warfarin cohorts were inspected to make certain cohorts were well balanced with key patient characteristics statistically similar (p > .05); t-tests and chi-square tests were used for the continuous and categorical variables, respectively.

SD: standard deviation.

Apixaban vs. rivaroxaban

After PSM, 6810 NVAF patients were in each cohort. Mean ages (77.1 vs. 77.0 years, p = .83), CCI scores (2.8 vs. 2.7, p = .97), and stroke and bleeding risks, according to CHADS₂ scores (2.6 vs. 2.6, p = .97), CHA₂DS₂-VASc scores (4.4 vs. 4.4, p = .86), and HAS-BLED scores (2.9 vs. 2.9, p = .90), were not statistically different. The proportions of patients with stroke (10.0% vs. 9.8%, p = .69) and bleeding (16.8% vs. 16.6%, p = .81) diagnoses during the baseline period were also similar for NVAF patients treated with apixaban vs. rivaroxaban, as were the mean durations of follow-up (6.5 vs. 6.4 months, p = .54).

Apixaban vs. dabigatran

After PSM, 2327 NVAF patients were in each cohort. Mean ages (77.3 vs. 76.9 years, p = .11), CCI scores (2.6 vs. 2.6, p = .86), and stroke and bleeding risks, according to CHADS₂ scores (2.6 vs. 2.6, p = .61), CHA₂DS₂-VASc scores (4.3 vs. 4.3, p = .58), and HAS-BLED scores (2.9 vs. 2.9, p = .77), were not statistically different. The proportions of patients with stroke (9.9% vs. 10.2%, p = .73) and bleeding (16.4% vs. 16.2%, p = .84) diagnoses during the baseline period were also similar for NVAF patients treated with apixaban vs. dabigatran, as were the mean durations of follow-up (7.1 vs. 7.0 months, p = .71).

Apixaban vs. warfarin

After PSM, 7107 NVAF patients were in each cohort. Mean ages (78.2 vs. 78.1 years, p = .54), CCI scores (3.0 vs. 3.0, p = .97), and stroke and bleeding risks, according to CHADS₂ scores (2.7 vs. 2.7, p = .37), CHA₂DS₂-VASc scores (4.6 vs. 4.6, p = .66), and HAS-BLED scores (3.0 vs. 3.1, p = .22), were not statistically different. The proportions of patients with stroke (11.9% vs. 11.7%, p = .84) and bleeding (18.8% vs. 19.0%, p = .81) diagnoses during the baseline period were also similar for NVAF patients treated with apixaban vs. warfarin, as were the mean durations of follow-up (6.7 vs. 6.6 months, p = .66).

Post propensity score matching: unadjusted stroke/systemic embolism and major bleeding event rates of matched cohorts, apixaban vs. rivaroxaban

Among the matched populations, the unadjusted annual event rates of any S/SE (2.4% vs. 3.3%), ischemic stroke (1.7% vs. 2.6%), and hemorrhagic stroke (0.6% vs. 0.7%) during the follow-up were numerically lower for NVAF patients treated with apixaban in comparison to those treated with rivaroxaban. The rate of SE was 0.2% among apixaban treated patients and 0.1% among rivaroxaban treated patients. Additionally, event rates of any MB (4.7% vs. 9.5%), GI MB (2.5% vs. 5.9%), and other MB (1.8% vs. 3.6%) during the follow-up were numerically lower for NVAF patients treated with apixaban in comparison to those treated with rivaroxaban. The rate of intracranial hemorrhage was 0.7% among both apixaban and rivaroxaban treated patients. Illustrated with Kaplan Meier curves, times to any S/SE and any MB among matched cohorts treated with apixaban vs. rivaroxaban are shown in Supplementary Figures 1 and 2, respectively.

Cox regression results: apixaban vs. rivaroxaban

Cox regression results showed that among matched study cohorts the risks for any S/SE (hazard ratio [HR]: 0.72, p = .003), ischemic stroke (HR: 0.67, p = .012), any MB (HR: 0.49, p < .001), GI MB (HR: 0.43, p < .001), and other MB (HR: 0.51, p < .001) were significantly lower during the follow-up for patients treated with apixaban vs. rivaroxaban (Table 3). There were no significant differences in the risks for hemorrhagic stroke, systemic embolism, or intracranial hemorrhage between the two groups (Table 3).

Table 3. Cox regression results among matched cohorts: comparison of risks of stroke/systemic embolism and major bleeding during the follow-up associated with apixaban treatment vs. rivaroxaban, dabigatran, and warfarin.

	Hazard Ratio	95% Confidence Limits		p-value
Outcome: Apixaban vs. Rivaroxaban
Stroke/Systemic Embolism
Any	0.72	0.55	0.95	.003
Ischemic Stroke	0.67	0.49	0.92	.012
Hemorrhagic Stroke	0.87	0.50	1.53	.63
Systemic Embolism	1.19	0.36	3.94	.78
Major Bleeding
Any	0.49	0.41	0.59	<.001
Intracranial Hemorrhage	0.99	0.58	1.69	.96
Gastrointestinal	0.43	0.33	0.54	<.001
Other	0.51	0.38	0.68	<.001
Outcome: Apixaban vs. Dabigatran
Stroke/Systemic Embolism
Any	0.78	0.51	1.21	.27
Ischemic Stroke	0.77	0.46	1.29	.32
Hemorrhagic Stroke	1.10	0.43	2.85	.84
Systemic Embolism	0.50	0.09	2.70	.42
Major Bleeding
Any	0.82	0.59	1.14	.23
Intracranial Hemorrhage	0.98	0.39	2.45	.97
Gastrointestinal	0.68	0.44	1.04	.08
Other	0.84	0.48	1.45	.52
Outcome: Apixaban vs. Warfarin
Stroke/Systemic Embolism
Any	0.65	0.51	0.83	<.001
Ischemic Stroke	0.63	0.48	0.83	<.001
Hemorrhagic Stroke	0.66	0.39	1.12	.13
Systemic Embolism	1.31	0.46	3.8	.62
Major Bleeding
Any	0.53	0.45	0.63	<.001
Intracranial Hemorrhage	0.71	0.43	1.17	.18
Gastrointestinal	0.53	0.42	0.66	<.001
Other	0.48	0.37	0.62	<.001

Post propensity score matching: unadjusted stroke/systemic embolism and major bleeding event rates of matched cohorts, apixaban vs. dabigatran

Among the matched populations, the unadjusted annual event rates of any S/SE (2.6% vs. 3.3%), ischemic stroke (2.0% vs. 2.5%), and SE (0.2% vs. 0.3%) during the follow-up were numerically lower for NVAF patients treated with apixaban in comparison to those treated with dabigatran. The rate of hemorrhagic stroke was 0.7% among apixaban treated patients and 0.6% among dabigatran treated patients. Additionally, the event rates of any MB (4.7% vs. 5.8%), GI MB (2.6% vs. 3.8%), and other MB (1.7% vs. 2.0%) during the follow-up were numerically lower for NVAF patients treated with apixaban in comparison to those treated with dabigatran. The rate of intracranial hemorrhage was 0.7% among both apixaban and dabigatran treated patients. Illustrated with Kaplan Meier curves, times to any S/SE and any MB among matched cohorts treated with apixaban vs. dabigatran are shown in Supplementary Figures 3 and 4, respectively.

Cox regression results: apixaban vs. dabigatran

Cox regression showed that the risks for any S/SE (HR: 0.78, p = .27) and any MB (HR: 0.82, p = .23) among NVAF patients treated with apixaban vs. dabigatran trended to be lower, but did not reach statistical significance (Table 3). There were no significant differences in the risks for any of the subcategories of S/SE and MB (Table 3).

Post propensity score matching: unadjusted stroke/systemic embolism and major bleeding event rates of matched cohorts: apixaban vs. warfarin

Among the matched populations, the unadjusted annual event rates of any S/SE (2.8% vs. 4.2%), ischemic stroke (2.1% vs. 3.3%), and hemorrhagic stroke (0.6% vs. 0.9%) during the follow-up were numerically lower for NVAF patients treated with apixaban in comparison to those treated with warfarin. The rate of SE was 0.2% among apixaban treated patients and 0.1% among warfarin treated patients. Additionally, event rates of any MB (5.1% vs. 9.4%), intracranial hemorrhage (0.7% vs. 0.9%), GI MB (2.7% vs. 5.1%), and other MB (2.1% vs. 4.3%) during the follow-up were numerically lower for NVAF patients treated with apixaban in comparison to those treated with warfarin. Illustrated with Kaplan Meier curves, times to any S/SE and any MB among matched cohorts treated with apixaban vs. warfarin are shown in Supplementary Figures 5 and 6, respectively.

Cox regression results: apixaban vs. warfarin

Cox regression results showed that among matched study cohorts the risks for any S/SE (HR: 0.65, p < .001), ischemic stroke (HR: 0.63, p < .001), any MB (HR: 0.53, p < .001), GI MB (HR: 0.53, p < .001), and other MB (HR: 0.48, p < .001) were significantly lower during the follow-up for patients treated with apixaban vs. warfarin (Table 3). There were no significant differences in the risks for hemorrhagic stroke, systemic embolism, or intracranial hemorrhage between the two groups (Table 3).

Discussion

Elderly NVAF patients are at increased risk for stroke and bleeding^2,18. Information on the stroke and bleeding risk in this unique patient population will help better manage these patients. This retrospective cohort study compared the risk of any S/SE and any MB between apixaban and other OACs among NVAF patients ≥65 years of age newly initiated on anticoagulant treatments in routine clinical practice. The findings from this study indicate that treatment with apixaban was associated with significantly lower risk of S/SE and MB than treatment with rivaroxaban and warfarin. The risks of S/SE and MB among elderly NVAF patients treated with apixaban vs. dabigatran trended to be lower, but did not reach statistical significance.

Elderly NVAF patients initiating different DOACs in routine clinical practice demonstrated several differences in patient characteristics. Patients treated with apixaban in comparison to those treated with rivaroxaban or dabigatran were older, more likely to have comorbidities, and have higher stroke and bleeding risks. These differences in patient characteristics are similar to those observed in an earlier evaluation of NVAF patients identified from the Premier Hospital and Cerner Health Facts databases who received apixaban, rivaroxaban, or dabigatran¹⁶. Additionally, our results are consistent with those of the study conducted by Lip and colleagues of NVAF patients identified from the MarketScan Commercial and Medicare Supplemental databases¹⁴. In this latter study patients treated with apixaban vs. rivaroxaban and dabigatran were also older, more frequently diagnosed with comorbidities, and had higher stroke and bleeding risks according to CHADS₂, CHA₂DS₂-VASc, and HAS-BLED scores¹⁴. Our study also shows some differences in patient characteristics between patients treated with warfarin vs. apixaban, with warfarin patients being older, more likely to have comorbidities, and having higher stroke and bleeding risks than apixaban patients. Consistently, Lip and colleagues reported that NVAF patients treated with warfarin were older and sicker than patients treated with other OACs¹⁴. The underlying reasons for the different patient characteristics associated with different OAC use are not completely clear and warrant further investigations.

After balancing patient characteristics with PSM, the Cox regression analyses indicated that apixaban treatment in comparison to treatment with warfarin was associated with a lower risk for both stroke (35% lower S/SE risk) and MB (47% lower MB risk). These findings are consistent with that of the ARISTOTLE trial, in which apixaban demonstrated superiority vs. warfarin for both S/SE and MB⁷. These findings are also consistent with a recent retrospective cohort study conducted by Yao and colleagues¹³. They reported that treatment with apixaban vs. warfarin was associated with a 33% lower risk for S/SE and 55% lower risk for MB among NVAF patients (median age: 73 years)¹³.

Our study showed that in the matched cohorts, apixaban treatment in comparison to treatment with rivaroxaban was associated with a 28% lower risk for S/SE (any type) and a 51% lower risk for MB (any type). Furthermore, the risks for ischemic stroke, GI MB, and other MB were also significantly lower for patients treated with apixaban vs. rivaroxaban. When compared to dabigatran, treatment with apixaban was associated with numerically lower risks of S/SE and MB, but the differences did not reach statistical significance. Although our findings are consistent with that of randomized clinical trials^8,9, the event rates are not exactly the same, which is likely attributed to the differences in patient populations, with those in real-world settings being more diverse than those included in clinical trials. Thus, the results of the current study complement those of the clinical trials with those of real-world findings. While there are no head-to-head trials to compare different DOACs, an indirect comparison based on clinical trials that compared the outcomes of patients treated with each of the different DOACs vs. warfarin suggested that treatment with apixaban would be associated with a significantly lower risk of MB vs. treatment with rivaroxaban²². Unlike the clinical trials, the current study focused on elderly NVAF patients and patients in routine clinical practice may be different from those participating in controlled clinical trials. Findings on MB from this study are generally consistent with the findings of Lip et al., who also compared the safety of DOACs in routine clinical practice after PSM¹⁵. Lip et al. reported a nearly two-fold lower risk of MB (HR: 1.82; 95% CI: 1.36–2.43) for NVAF patients who received apixaban in comparison to those who received rivaroxaban (mean ages of cohorts: 68 years)¹⁵. The risk of MB was also lower for patients who received apixaban vs. those who received dabigatran, but the difference was not statistically significant (mean ages of cohorts: 67 years)¹⁵. The results on MB from this study are also similar to the findings of an early evaluation of MB risk among NVAF patients treated with different DOACs, which found the use of rivaroxaban compared to apixaban was associated with significantly greater risk of bleeding-related hospital readmissions across two different database analyses¹⁶. Consistently, a recent study by Noseworthy et al. using the Optum database indicated that the MB risks were significantly lower for patients treated with apixaban vs. rivaroxaban (HR: 0.39, p < .001)¹⁷. This study also found a significantly lower risk of MB for apixaban vs. dabigatran (HR: 0.50, p < .001)¹⁷. Unlike the current study, Noseworthy and colleagues did not find a significant difference in the risk of S/SE among NVAF patients treated with apixaban vs. rivaroxaban (HR: 1.05, p = .85). While their study also used PSM to control for differences in patient characteristics among different DOAC cohorts, it included patients of all ages and did not exclude patients who had prior treatment with warfarin¹⁷. In contrast, the current study focused on elderly NVAF patients newly initiating on OAC treatment.

It is important to understand treatment-related clinical outcomes among NVAF patients treated with different OACs in routine clinical practice to discern the most appropriate treatment strategies and reduce the disease burden. The results of our study provide further evidence about the risk of S/SE and MB associated with different OACs among NVAF patients, especially those ≥65 years of age. Future additional studies with larger sample sizes and longer time frames may be needed to further confirm the results of this real-world comparative study.

Study limitations

Retrospective, observational analyses using claims databases have certain inherent limitations, including that no causal relationship between treatment and outcomes can be established. Second, although a claim for a filled prescription may be documented, this does not indicate that the medication was consumed or that it was taken as prescribed. Also, medications filled over the counter or provided as samples by the physician are not captured in the claims data. Lab values are not reliably captured in the Humana database and therefore how often warfarin was within the therapeutic range could not be reliably assessed. Also, drug adherence to OACs was not evaluated and could impact outcomes. Edoxaban was not included in this analysis as its FDA approval was not long before the end of the study period and future studies are warranted to also include this more recently approved DOAC in the comparison of the effectiveness and safety of OACs in the real-world setting. Furthermore, NVAF patients may receive different dosages of DOACs in the real-world setting. In this study NVAF patients receiving any dosage of DOACs were included and future additional studies may be needed to examine whether our findings are consistent among those receiving only standard dosages or lower dosages of DOACs. Since a diagnosis code may be incorrectly coded or included as rule-out criteria rather than actual disease, the presence of a diagnosis code on a medical claim may not be positive indication of a particular disease. Additionally, mortality data is not reliably captured in the data source and was not evaluated as an outcome in this study. Although PSM was used to control for multiple confounders, there is potential for residual bias in this study. PSM does not take into account other confounding factors that were not included in the covariate list of the PSM process, such as drug adherence. However, many relevant real-world factors that may potentially impact the study outcomes were included in the PSM process already. Also, in the PSM process patients who did not have good matching were removed from the study cohorts and thus the findings may not be generalizable to the entire real-world population of elderly NVAF patients. Despite these potential limitations, the PSM process generated well balanced patient cohorts and provided a reliable tool for comparison of patient outcomes. Lastly, the Humana Research Database comprises claims of persons primarily residing in the Southern and Midwestern regions of the US and the results of this study may not be generalizable to the entire US elderly population.

Conclusions

In a real-world study of elderly NVAF patients in the US, after controlling for differences in patient characteristics, treatment with apixaban was associated with significantly lower risks of S/SE and MB than treatment with rivaroxaban and warfarin, and trended towards better outcomes compared with treatment with dabigatran. Further evaluations with longer time frames and/or based on other data sources may be needed to validate the findings of this study.

Transparency

Declaration of funding

This work was supported by Pfizer and Bristol-Myers Squibb.

Author contributions: S.D., X.L., K.G., J.T., J.M., T.C., and J.L. were responsible for study design. All authors contributed to data analysis and interpretation. M.L.-S., B.M., and J.L. drafted the manuscript with critical contributions from other authors. All authors approved the final manuscript for publication.

Declaration of financial/other relationships

S.D. has disclosed that he is a consultant for Pfizer and Bristol-Myers Squibb. X.L., J.T. and J.M. have disclosed that they are employees of Pfizer and own stock in the company. K.G. and T.C. have disclosed that they are employees of Bristol-Myers Squibb and own stock in the company. M.L.-S., B.M. and J.L. have disclosed that they are employees of Novosys Health, which has received research funds from Pfizer and Bristol-Myers Squibb in connection with conducting this study and development of this manuscript.

CMRO peer reviewers on this manuscript have received an honorarium from CMRO for their review work, but have no relevant financial or other relationships to disclose.