Healthcare resource utilization and expenditures among newly-diagnosed elderly non-valvular atrial fibrillation patients initiating oral anticoagulants

Abstract

Aims: Non-valvular atrial fibrillation (NVAF) prevalence increases with age. Hence, evaluating the economic burden among older-aged patients is vital. This study aimed to compare healthcare resource utilization (HRU) and costs among newly-diagnosed older-aged NVAF patients treated with warfarin, rivaroxaban, or apixaban vs. dabigatran.

Materials and Methods: Newly-diagnosed older-aged (aged ≥65 years) NVAF patients initiating dabigatran, warfarin, rivaroxaban, or apixaban (first prescription date = index date) from 01JAN2010-31DEC2015 and with continuous enrollment for ≥12 months pre-index date were included from 100% Medicare database. Patient data were assessed until drug discontinuation/switch/dose change/death/disenrollment/study end (up to 12 months). Dabigatran initiators were 1:1 propensity score-matched (PSM) with warfarin, rivaroxaban, or apixaban initiators. Generalized linear models were used to compare all-cause HRU and costs per-patient-per-month (PPPM) between the matched cohorts.

Results: After PSM with dabigatran, 70,531 warfarin, 51,673 rivaroxaban, and 25,209 apixaban patients were identified. Dabigatran patients had significantly fewer generalized-linear-model-adjusted PPPM hospitalizations (0.114 vs. 0.123; 0.111 vs. 0.121), and outpatient visits (2.864 vs. 4.201; 2.839 vs. 2.949) than warfarin and rivaroxaban patients, respectively, but had significantly more PPPM hospitalizations (0.103 vs. 0.090) and outpatient visits (2.780 vs. 2.673) than apixaban patients (all p < .0001). Dabigatran patients incurred significantly lower adjusted total PPPM costs ($3,309 vs. $3,362; $3,285 vs. $3,474) than warfarin and rivaroxaban patients, respectively (all p < .01) but higher total PPPM costs ($3,192 vs. $2,986) than apixaban patients (all p < .0001).

Limitations: This study is subject to the inherent limitations of any claims dataset, including potential bias from coding errors and identification of medical conditions using diagnosis codes as opposed to clinical evidence. Medications filled over-the-counter or provided as samples by the physician are never captured in claims data.

Conclusions: Newly-diagnosed older-aged NVAF patients initiating dabigatran incurred significantly lower adjusted all-cause HRU and costs than warfarin and rivaroxaban patients but higher adjusted HRU and costs than apixaban patients.

Keywords:

• Healthcare costs
• novel oral anticoagulants
• older-aged patients
• non-valvular atrial fibrillation
• economic burden

Introduction

Atrial fibrillation (AF) is the most common arrhythmia¹, affecting 2% of the world’s population² and 2.7–6.1 million people in the United States^3,4 Given the aging US population and increasing risk of AF with age, the estimated prevalence of AF is expected to double by 2030 and triple by 2050^5–7. AF is associated with poor quality of life and a substantially increased risk of stroke and other thromboembolic events^1–3 In the US, about 15% of all strokes and thromboembolic events are associated with AF³. According to the American Heart Association, the total cost of stroke, including direct and indirect expenses is estimated to grow from $105.2 billion in 2012 to $240.7 billion by 2030, thus adding a substantial burden on the US economy⁸. About 96% of AF patients in developed countries have non-valvular atrial fibrillation (NVAF), defined as AF without mitral stenosis or valvular prostheses². Patients with NVAF exhibit symptoms such as chest pain, breathing difficulty, and have an increased risk of stroke or thromboembolic events compared to the non-NVAF population^9–11.

Since the 1950s, the vitamin K antagonist warfarin has been the most common oral anticoagulant (OAC) prescribed for NVAF patients^2,3,12. A meta-analysis of randomized trials showed that adjusted-dose warfarin reduced the risk of stroke by 64% when compared to placebo and by 37% compared to antiplatelet therapy such as aspirin¹³. Despite warfarin being relatively inexpensive, it requires regular monitoring and dose modification¹⁴, due to interactions with food, other drugs, alcohol, factors such as liver function, age-related alterations, and certain genetic variations, some of which may predispose patients to unsafe bioavailability^2,3.

In recent years, novel oral anticoagulants (NOACs) including dabigatran, rivaroxaban and apixaban have been approved by the US Food & Drug Administration for prevention of stroke in NVAF patients^15,16. These NOACs have been found to be safe and efficacious among NVAF patients^17–21. Dabigatran, which is available in 2 dosages (110 mg and 150 mg), is associated with lower rates of stroke and major bleeding than warfarin¹⁷. Rivaroxaban is associated with similar efficacy and lower rates of stroke and intracranial bleeding compared to warfarin¹⁹. Apixaban is associated with lower rates of major bleeding, and lower all-cause mortality rates compared to warfarin¹⁸. With NVAF’s definition and care management still evolving, current guidelines of care suggest that patients treated with NOACs may have an improved prognosis^1,20–22. Although NOACs are more expensive than warfarin, they have wide therapeutic windows that allow for predictable dosing, thus reducing the need for periodic monitoring³.

In addition to safety and effectiveness differences, real-world studies have also examined the economic burden associated with NOACs compared to warfarin^23–26. Although NOACs have higher prescription costs, they are associated with lower healthcare resource utilization (HRU) and costs compared to warfarin^5,27. It has also been suggested that dabigatran and apixaban are cost-effective NOACs²⁸. A recent retrospective study using commercial database comparing OACs to dabigatran found that dabigatran users had lower all-cause HRU and costs compared to warfarin and rivaroxaban users but had all-cause HRU and costs similar to apixaban users⁵.

Given the lack of focus on Medicare populations and the increasing prevalence of NVAF with age, it would be informative to evaluate the economic burden among Medicare beneficiaries (aged ≥65 years) treated with OACs. Expanding on findings from prior retrospective studies in different populations, the current analysis compared all-cause HRU and costs of newly-diagnosed and newly-treated NVAF patients prescribed warfarin, rivaroxaban, or apixaban, vs. dabigatran. Dabigatran was the first approved NOAC, and it was therefore chosen as the reference drug for comparison with the other OACs in the present study. Additionally, our study provides the data to the payer with direct head-to-head comparison among NOACs that are needed in their formulary decision making.

Methods

Data source

This was a retrospective cohort study using 100% Medicare database from 01 Jan 2010–31 Dec 2015. Medicare provides health insurance coverage to 46 million persons aged ≥65 years in the US^29,30. For insured beneficiaries, Medicare covers the costs of hospitalizations through Medicare Part A, the costs of physicians’ services through Medicare Part B, and the costs of prescription drugs through Medicare Part D. About a quarter of Medicare enrollees receive their insurance through Medicare Advantage (Part C). Neither patient identity nor medical records were disclosed for the purposes of this study. The study was exempt from Institutional Review Board approval as the data were de-identified and only aggregate results were reported. Compliance with all applicable laws and the Health Insurance Portability and Accountability Act regulations were maintained.

Subject selection

Patients were included in the study if they had ≥1 inpatient or outpatient claim for AF (International Classification of Diseases, 9th Revision, Clinical Modification: 427.31) between 01 Jan 2011 and 31 Dec 2014 (identification period). Patients were required to have been continuously enrolled (fee-for-service only) with medical (Parts A and B) and pharmacy (Part D) benefits for ≥6 months prior to the first observed AF diagnosis. Additionally, patients were required to have ≥1 outpatient pharmacy claim with a National Drug Code indicative of dabigatran, warfarin, rivaroxaban, or apixaban on or after the first observed AF diagnosis and between the launch date for each NOAC and 31 Dec 2014 (dabigatran vs. warfarin: 01 Jan 2011; dabigatran vs. rivaroxaban: 01 Nov 2011; dabigatran vs. apixaban: 01 Dec 2012). The first pharmacy claim for dabigatran, warfarin, rivaroxaban, or apixaban was designated as the index date. Patients were required to be ≥65 years of age on the index date and have continuous enrollment in Part A, B, and D health plans for ≥12 months prior to the index date (baseline period = 12 months prior to the index date). Patients were required to have no evidence of AF or cardiac surgery, hyperthyroidism, myocarditis, pericarditis, pregnancy, pulmonary embolism, valve replacement, chronic rheumatic heart disease, or valvular disease in the 6 months prior to the first observed AF diagnosis. Additionally, patients were excluded from the study if they had evidence of any anticoagulants use (dabigatran, rivaroxaban, warfarin, apixaban, argatroban, dalteparin, enoxaparin, fondaparinux, heparin, or tinzaparin) in the 6-month pre-index period or had evidence of >1 anticoagulant or >1 dose of the index NOAC on the index date. Patients’ data were assessed until the earliest of discontinuation of the index medication, switch to a non-index OAC, change in dosage of the index NOAC, death, end of continuous health plan enrollment, study end (31 Dec 2015), or end of the ≥12-month follow-up period (Figure 1). Each NOAC captured had multiple dosages, and physicians are likely to change these dosages depending on the creatinine clearance level and various other patient factors. To make sure that the HRU and costs were only captured during the course of the index treatment, all patients were censored at the earliest of discontinuation of the index medication, switch to a non-index OAC, change in dosage of the index NOAC, death, end of continuous health plan enrollment, study end, or end of the ≥12-month follow-up period. Discontinuation was defined as a treatment gap of >90 days from the end of the last filled index prescription. Switching was defined as the initiation of a non-index OAC within 30 days of the last days’ supply of the index medication, provided the index medication was discontinued.⁵ Patients were categorized into 3 comparison groups based on their index OAC claim in the corresponding identification period, as follows: dabigatran vs. warfarin, dabigatran vs. rivaroxaban, and dabigatran vs. apixaban.

Figure 1. Study design. Abbreviation. AF, atrial fibrillation. The identification period in the figure is shown as an example of case selection for the dabigatran vs. warfarin comparison.

Demographic and baseline clinical characteristics

Patient demographics including age, sex, race, and geographic region as of the index date were captured. Individual comorbidities were identified based on International Classification of Diseases, 9th Revision, Clinical Modification codes (Online Table 1). Comorbidity burden during the baseline period was assessed using the Deyo-modified Charlson comorbidity index score and other clinical indices measuring stroke risk (including the CHADS₂ Score [congestive heart failure, hypertension, age ≥75 years, diabetes, prior stroke] and CHA₂DS₂-VASc Score [congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/TIA/thromboembolism, vascular disease, age 65–74 years, sex category]) and bleeding risk (including HAS-BLED [hypertension, abnormal renal/liver function, stroke, bleeding history/predisposition, labile international normalized ratio, elderly, drugs/alcohol concomitantly] and ATRIA Score [anticoagulation and risk factors in atrial fibrillation]). Measures of overall health status (including the number of unique medications, the number of inpatients visits observed, the number of physician visits observed, home oxygen, and hip fracture) and concomitant medications identified by National Drug Codes were also assessed. Time to index NOAC was calculated as the number of days between the first NVAF diagnosis and the index date and the index dosage of the NOACs was reported.

Table 1. Sample size before and after propensity score matching.

	Before propensity score matching		After propensity score matching
	OAC cohort	Dabigatran cohort	OAC cohort*	Dabigatran cohort*
Warfarin	299,747 (100%)	70,608 (100%)	70,531 (23.5%)	70,531 (99.9%)
Rivaroxaban	99,998 (100%)	51,710 (100%)	51,673 (51.7%)	51,673 (99.9%)
Apixaban	45,543 (100%)	25,410 (100%)	25,209 (55.4%)	25,209 (99.2%)

Abbreviation. OAC, oral anticoagulant.

*Denotes the percentage of the sample size remaining after matching from the full cohort before matching. Covariates included in the propensity score matching were: demographics (age, gender, race, U.S region), health status (unique medications, inpatient visits per patient, physician office visits per patient, home oxygen, hip fracture), individual comorbidities (chronic kidney disease, chronic obstructive pulmonary disease, pneumonia, cirrhosis, coronary artery disease, diabetes mellitus, gastrointestinal bleed, heart failure, ischemic stroke, intracranial bleed, myocardial infarction, paraplegia, psychiatric disorders, venous thromboembolism), Deyo–Charlson comorbidity index score, CHA₂DS₂-VASc score, HAS-BLED score, concomitant medications (beta-blockers, calcium channel blockers, diuretics, other antihypertensives, antihyperlipidemics, corticosteroids, antidiabetics, antiarrhythmics, ketoconazole, non-steroidal anti-inflammatory drugs, selective serotonin reuptake inhibitors, serotonin and norepinephrine reuptake inhibitors, and antiplatelets), time to index exposure, and baseline log total healthcare costs.

Outcome measures

All-cause HRU and healthcare PPPM costs during the 12-month follow-up period were assessed. All-cause HRU was computed as the number of healthcare encounters, categorized as hospitalizations, outpatient visits, and pharmacy visits. Outpatient visits consisted of physician office, outpatient hospital, and emergency room (ER) visits. All-cause healthcare costs were categorized and assessed as hospitalization, outpatient, other (durable medical equipment, home health agency, skilled nursing facility, and hospice costs), pharmacy, total medical (hospitalization, outpatient, and other costs), and total costs (total medical and pharmacy costs). Costs included payments made by Medicare, the primary payer, patient copayments, deductibles, and coinsurance. All healthcare costs were adjusted for inflation to 2017 US dollars using the annual medical care and drug costs components of the Consumer Price Index³¹.

Statistical analysis

Descriptive statistics (means and standard deviations for continuous variables and numbers and percentages for categorical variables) were provided for all study variables, including demographics, baseline characteristics, and the outcome measures in the study cohorts. Statistical tests (chi-square for categorical and Student’s t-test for continuous variables) were conducted to assess differences between dabigatran and each of the other OACs individually.

Propensity score matching was used to achieve balance in patient characteristics. Dabigatran patients were 1:1 propensity score-matched (PSM) separately with warfarin, rivaroxaban, or apixaban patients based on a caliper width of 0.005. Logistic regression was used to calculate the propensity score for all patients, and each patient in the dabigatran cohort was matched to a patient in the respective comparison cohort with the closest propensity score. Variables used in the matching procedure included demographics, health status, comorbidity profile (Deyo–Charlson comorbidity index score, CHA₂DS₂-VASc score, HAS-BLED score, and individual comorbidities), time to index exposure, concomitant medications, and baseline log total healthcare costs. Multicollinearity of independent variables was assessed using the variance inflation factor. The adequacy of the matching procedure was assessed using standardized differences ([STDs], STD <10% is considered well-balanced) and L1 statistics³².

Generalized linear models with gamma distribution and log link function were used to estimate all-cause costs PPPM in PSM cohorts to account for the non-normal distribution of cost data³³. Additionally HRU, including the number of visits PPPM, were estimated using generalized linear models with negative binomial distribution³³. Covariates in the generalized linear models included demographics (age, sex, race, and US region), CHA2DS2-VASc and HAS-BLED score. A critical value of p < .05 was specified a priori as a threshold of statistical significance for all analyses. All analyses were conducted using SAS software (version 9.4, SAS Institute, Cary, North Carolina, USA).

Sensitivity analysis

To ensure the robustness of the results, a sensitivity analysis was performed among the standard dosage patients, which included newly diagnosed NVAF patients on dabigatran 150 mg, rivaroxaban 20 mg, and apixaban 5 mg. The above-mentioned analysis was repeated in these patients, where dabigatran patients were 1:1 PSM separately with warfarin, rivaroxaban, or apixaban patients based on a caliper width of 0.005. All-cause HRU and health care costs were estimated using GLM between these matched cohorts.

Results

After applying the selection criteria, the study included a total of 370,355 patients in the dabigatran (70,608) vs. warfarin (299,747) comparison, 151,708 patients in the dabigatran (51,710) vs. rivaroxaban (99,998) comparison, and 70,953 patients in the dabigatran (25,410) vs. apixaban (45,543) comparison groups before propensity score matching. After propensity score matching, 70,531 patients each were included in the dabigatran vs. warfarin cohorts, 51,673 patients each for dabigatran vs. rivaroxaban, and 25,209 patients each for dabigatran vs. apixaban groups (Figure 2, Table 1).

Figure 2. Patient attrition. Abbreviations. AF, atrial fibrillation; NOAC, novel oral anticoagulant; NVAF, non-valvular atrial fibrillation; OAC, oral anticoagulant.

Patient characteristics

Table 2 depicts the baseline demographic and clinical characteristics among the three comparison cohorts before propensity score matching. In the three comparison cohorts, the mean age was approximately 78 years, with a majority of patients female (about 55%) and White (about 91%). Patients in the dabigatran cohort had a significantly lower mean Deyo-Charlson comorbidity index score CHA₂DS₂-VASc score and HAS-BLED score and had lower comorbidity burden than warfarin, rivaroxaban, and apixaban respectively. Additionally, the dabigatran cohort had significantly fewer unique medications as compared to the warfarin, rivaroxaban, and apixaban cohorts, respectively, and the majority of these patients were on beta-blockers. Before PSM, most of the patients were on a higher dosage of NOACs on the index date which included approximately 80% patients on 150 mg dabigatran, 69% on 20 mg rivaroxaban and 71% on 5 mg apixaban on the index date. The HRU and costs in the baseline period are presented in Online Table 2. After propensity score matching, the variables in the matched comparison cohorts were well-balanced based on STDs <10% and a reduced L1 statistic. However, it should be noted that there were significant differences observed in the index NOAC dosage between the dabigatran and rivaroxaban or apixaban cohorts after PSM. Baseline characteristics after propensity score matching are displayed in Table 3.

Table 2. Baseline demographics and clinical characteristics (before propensity score matching).

Demographic and clinical characteristics	Dabigatran (reference) N = 70,608	Warfarin N = 299,747	p-Value	STD	Dabigatran (reference) N = 51,710	Rivaroxaban N = 99,998	p-Value	STD	Dabigatran (reference) N = 25,410	Apixaban N = 45,543	p-Value	STD
	Mean ± SD or n (%)	Mean ± SD or n (%)			Mean ± SD or n (%)	Mean ± SD or n (%)			Mean ± SD or n (%)	Mean ± SD or n (%)
Age, in years	77.70 ± 7.07	78.90 ± 7.37	<.0001	16.62	77.58 ± 7.05	77.92 ± 7.14	<.0001	4.74	77.47 ± 6.99	78.74 ± 7.37	<.0001	17.65
Age categories
65–69 years	9,602 (13.60%)	34,359 (11.46%)	<.0001	6.46	7,041 (13.62%)	12,876 (12.88%)	<.0001	2.18	3,429 (13.49%)	5,210 (11.44%)	<.0001	6.22
70–74 years	16,699 (23.65%)	59,662 (19.90%)	<.0001	9.09	12,622 (24.41%)	23,308 (23.31%)	<.0001	2.58	6,338 (24.94%)	9,597 (21.07%)	<.0001	9.21
≥75 years	44,307 (62.75%)	205,726 (68.63%)	<.0001	12.41	32,047 (61.97%)	63,814 (63.82%)	<.0001	3.81	15,643 (61.56%)	30,736 (67.49%)	<.0001	12.41
Sex
Male	32,286 (45.73%)	128,089 (42.73%)	<.0001	6.03	23,821 (46.07%)	45,390 (45.39%)	.0123	1.36	11,928 (46.94%)	20,021 (43.96%)	<.0001	5.99
Female	38,322 (54.27%)	171,658 (57.27%)	<.0001	6.03	27,889 (53.93%)	54,608 (54.61%)	.0123	1.36	13,482 (53.06%)	25,522 (56.04%)	<.0001	5.99
Race
White	64,465 (91.30%)	267,965 (89.40%)	<.0001	6.45	46,869 (90.64%)	91,212 (91.21%)	.0002	2.00	22,925 (90.22%)	41,801 (91.78%)	<.0001	5.46
Black	2,620 (3.71%)	18,019 (6.01%)	<.0001	10.71	2,065 (3.99%)	4,052 (4.05%)	.5820	0.30	1,043 (4.10%)	1,787 (3.92%)	.2377	0.92
Hispanic	1,090 (1.54%)	4,878 (1.63%)	.1123	0.67	851 (1.65%)	1,527 (1.53%)	.0777	0.95	434 (1.71%)	521 (1.14%)	<.0001	4.76
Asian	1,283 (1.82%)	4,353 (1.45%)	<.0001	2.88	1,014 (1.96%)	1,521 (1.52%)	<.0001	3.36	541 (2.13%)	663 (1.46%)	<.0001	5.08
Native American	197 (0.28%)	1,000 (0.33%)	.0214	0.99	152 (0.29%)	255 (0.26%)	.1646	0.74	84 (0.33%)	99 (0.22%)	.0044	2.17
Other/Unknown	953 (1.35)	3,532 (1.18)			759 (1.47)	1,431 (1.43)			383 (1.51)	672 (1.48)		0.00
US region
Northeast	13,518 (19.15%)	60,398 (20.15%)	<.0001	2.53	10,088 (19.51%)	18,022 (18.02%)	<.0001	3.81	5,069 (19.95%)	8,307 (18.24%)	<.0001	4.35
Northcentral	15,901 (22.52%)	84,583 (28.22%)	<.0001	13.12	11,677 (22.58%)	22,082 (22.08%)	.0267	1.20	5,853 (23.03%)	9,590 (21.06%)	<.0001	4.77
South	29,357 (41.58%)	107,001 (35.70%)	<.0001	12.10	21,039 (40.69%)	44,019 (44.02%)	<.0001	6.75	10,000 (39.35%)	20,897 (45.88%)	<.0001	13.23
West	11,746 (16.64%)	47,347 (15.80%)	<.0001	2.28	8,829 (17.07%)	15,677 (15.68%)	<.0001	3.78	4,453 (17.52%)	6,722 (14.76%)	<.0001	7.52
Other	86 (0.12%)	418 (0.14%)	.2524	0.49	77 (0.15%)	198 (0.20%)	.0331	1.18	35 (0.14%)	27 (0.06%)	.0007	2.50
Individual comorbidities
Chronic obstructive pulmonary disease	20,143 (28.53%)	106,397 (35.50%)	<.0001	14.98	15,635 (30.24%)	31,026 (31.03%)	.0016	1.72	7,723 (30.39%)	13,763 (30.22%)	.6291	0.38
Coronary artery disease	31,032 (43.95%)	150,901 (50.34%)	<.0001	12.83	23,508 (45.46%)	45,880 (45.88%)	.1199	0.84	11,304 (44.49%)	21,812 (47.89%)	<.0001	6.84
Diabetes mellitus	24,768 (35.08%)	121,715 (40.61%)	<.0001	11.42	19,005 (36.75%)	36,114 (36.11%)	.0143	1.33	9,412 (37.04%)	16,483 (36.19%)	.0244	1.76
Heart failure	21,827 (30.91%)	125,513 (41.87%)	<.0001	22.93	16,784 (32.46%)	32,385 (32.39%)	.7755	0.15	8,146 (32.06%)	15,677 (34.42%)	<.0001	5.02
Psychiatric disorders	20,771 (29.42%)	119,310 (39.80%)	<.0001	21.96	16,395 (31.71%)	34,248 (34.25%)	<.0001	5.41	8,139 (32.03%)	15,239 (33.46%)	.0001	3.05
Deyo-modified Charlson comorbidity index score	2.17 ± 2.11	2.96 ± 2.50	<.0001	33.95	2.31 ± 2.17	2.43 ± 2.27	<.0001	5.00	2.31 ± 2.20	2.50 ± 2.28	<.0001	8.47
Measures of stroke risk
CHADS2 score	2.34 ± 1.24	2.67 ± 1.27	<.0001	25.76	2.41 ± 1.24	2.41 ± 1.24	.7519	0.17	2.39 ± 1.24	2.50 ± 1.24	<.0001	8.95
CHA2DS2 VASc score	4.15 ± 1.54	4.60 ± 1.57	<.0001	29.03	4.23 ± 1.54	4.25 ± 1.53	.0009	1.79	4.19 ± 1.54	4.37 ± 1.53	<.0001	11.62
Measures of bleeding risk
HAS-BLED score	2.67 ± 1.05	2.98 ± 1.17	<.0001	27.36	2.77 ± 1.05	2.81 ± 1.07	<.0001	4.31	2.74 ± 1.05	2.88 ± 1.08	<.0001	12.56
ATRIA score	2.92 ± 1.93	3.72 ± 2.31	<.0001	37.43	3.01 ± 1.97	3.12 ± 1.98	<.0001	5.39	2.96 ± 1.94	3.29 ± 2.07	<.0001	16.54
Measures of overall health status
Number of unique medications	11.34 ± 7.04	11.87 ± 7.46	<.0001	7.27	12.50 ± 7.32	12.81 ± 7.55	<.0001	4.06	12.73 ± 7.49	13.24 ± 7.46	<.0001	6.89
Number of inpatient visits per patient	0.80 ± 1.02	1.25 ± 1.35	<.0001	37.70	0.86 ± 1.11	0.92 ± 1.16	<.0001	4.92	0.81 ± 1.11	0.87 ± 1.13	<.0001	5.33
Number of physician office visits per patient	15.69 ± 13.20	18.27 ± 15.77	<.0001	17.76	18.34 ± 13.98	19.50 ± 14.48	<.0001	8.15	18.20 ± 13.93	19.84 ± 14.31	<.0001	11.60
Home oxygen	1 (0.00%)	11 (0.00%)	.3439	0.45	1 (0.00%)	7 (0.01%)	.1977	0.76	2 (0.01%)	5 (0.01%)	.6894	0.32
Hip fracture	130 (0.18%)	2,478 (0.83%)	<.0001	9.07	109 (0.21%)	352 (0.35%)	<.0001	2.67	48 (0.19%)	148 (0.33%)	.0009	2.69
Concomitant medication
Beta blockers	44,888 (63.57%)	176,679 (58.94%)	<.0001	9.52	33,841 (65.44%)	64,349 (64.35%)	<.0001	2.29	16,703 (65.73%)	31,440 (69.03%)	<.0001	7.04
Calcium channel blockers	27,481 (38.92%)	111,363 (37.15%)	<.0001	3.64	20,932 (40.48%)	39,277 (39.28%)	<.0001	2.45	10,255 (40.36%)	18,739 (41.15%)	.0407	1.60
Diuretics	37,868 (53.63%)	166,804 (55.65%)	<.0001	4.05	28,666 (55.44%)	53,928 (53.93%)	<.0001	3.03	14,072 (55.38%)	25,623 (56.26%)	.0234	1.77
Other antihypertensives	27,985 (39.63%)	118,089 (39.40%)	.2442	0.49	21,197 (40.99%)	38,800 (38.80%)	<.0001	4.48	10,284 (40.47%)	17,892 (39.29%)	.002	2.42
Antihyperlipidemics	42,344 (59.97%)	172,325 (57.49%)	<.0001	5.04	32,072 (62.02%)	61,607 (61.61%)	.1153	0.85	15,718 (61.86%)	29,370 (64.49%)	<.0001	5.46
Corticosteroids	28,466 (40.32%)	123,285 (41.13%)	<.0001	1.66	22,992 (44.46%)	46,402 (46.40%)	<.0001	3.90	11,368 (44.74%)	21,220 (46.59%)	<.0001	3.72
Time to index exposure	109.59 ± 201.37	140.68 ± 245.18	<.0001	13.86	160.62 ± 249.36	221.95 ± 335.39	<.0001	20.75	266.83 ± 342.20	287.87 ± 395.52	<.0001	5.69
Index NOAC dosage
150 mg (dabigatran)/20 mg (rivaroxaban)/5.0 mg (apixaban)	57,301 (81.15%)				40,884 (79.06%)	68,691 (68.69%)	<.0001	23.78	19,988 (78.66%)	32,254 (70.82%)	<.0001	18.12
75 mg (dabigatran)/15 mg (rivaroxaban)/2.5 mg (apixaban)	13,307 (18.85%)				10,826 (20.94%)	31,306 (31.31%)	<.0001	23.77	5,422 (21.34%)	13,289 (29.18%)	<.0001	18.12

Abbreviations. SD, standard deviation; STD, standardized difference.

Individual comorbidities and concomitant medications with a prevalence of <25% were not reported which included chronic kidney disease, pneumonia, cirrhosis, gastrointestinal bleed, ischemic stroke, intracranial bleed, myocardial infarction, paraplegia/hemiplegia, venous thromboembolism, antidiabetics, antiarrhythmics, ketoconazole, non-steroidal anti-inflammatory drugs, selective serotonin reuptake inhibitors, serotonin and norepinephrine reuptake inhibitors, and antiplatelets.

Table 3. Baseline demographic, clinical characteristics, costs & HRU of newly-diagnosed and newly-treated NVAF patients after PSM.

Demographic and Clinical Characteristics	Dabigatran (reference) N = 70,531	Warfarin N = 70,531	p-Value	STD	Dabigatran (reference) N = 51,673	Rivaroxaban N = 51,673	p-Value	STD	Dabigatran (reference) N = 25,209	Apixaban N = 25,209	p-Value	STD
	Mean ± SD or n (%)	Mean ± SD or n (%)			Mean ± SD or n (%)	Mean ± SD or n (%)			Mean ± SD or n (%)	Mean ± SD or n (%)
Age, years	77.70 ± 7.07	77.76 ± 7.06	.0835	0.92	77.58 ± 7.05	77.66 ± 7.07	.0581	1.18	77.52 ± 6.98	77.55 ± 7.14	.5865	0.48
Age categories
65–69	9,584 (13.59%)	9,593 (13.60%)	.9443	0.04	7,032 (13.61%)	7,008 (13.56%)	.8275	0.14	3,338 (13.24%)	3,614 (14.34%)	.0004	3.18
70–74	16,679 (23.65%)	16,321 (23.14%)	.0243	1.20	12,616 (24.42%)	12,218 (23.64%)	.0038	1.80	6,278 (24.90%)	6,031 (23.92%)	.0104	2.28
≥75	44,268 (62.76%)	44,617 (63.26%)	.0543	1.02	32,025 (61.98%)	32,447 (62.79%)	.0067	1.69	15,593 (61.85%)	15,564 (61.74%)	.7904	0.24
Sex
Male	32,233 (45.70%)	32,049 (45.44%)	.3253	0.52	23,804 (46.07%)	23,729 (45.92%)	.6397	0.29	11,793 (46.78%)	11,853 (47.02%)	.5923	0.48
Female	38,298 (54.30%)	38,482 (54.56%)	.3253	0.52	27,869 (53.93%)	27,944 (54.08%)	.6397	0.29	13,416 (53.22%)	13,356 (52.98%)	.5923	0.48
Race
White	64,398 (91.30%)	63,927 (90.64%)	<.0001	2.33	46,856 (90.68%)	46,932 (90.82%)	.4145	0.51	22,842 (90.61%)	22,858 (90.67%)	.8067	0.22
Black	2,617 (3.71%)	3,277 (4.65%)	<.0001	4.68	2,062 (3.99%)	2,093 (4.05%)	.6235	0.31	1,029 (4.08%)	1,039 (4.12%)	.8223	0.20
Hispanic	1,090 (1.55%)	1,117 (1.58%)	.5624	0.31	842 (1.63%)	873 (1.69%)	.4503	0.47	389 (1.54%)	384 (1.52%)	.8562	0.16
Asian	1,279 (1.81%)	1,075 (1.52%)	<.0001	2.26	1,008 (1.95%)	866 (1.68%)	.0009	2.06	511 (2.03%)	442 (1.75%)	.0240	2.01
Native American	197 (0.28%)	228 (0.32%)	.1321	0.80	148 (0.29%)	158 (0.31%)	.5670	0.36	065 (0.26%)	69 (0.27%)	.7293	0.31
Other	762 (1.08%)	746 (1.06%)	.6787	0.22	593 (1.15%)	569 (1.10%)	.4789	0.44	265 (1.05%)	312 (1.24%)	.0491	1.75
Unknown	188 (0.27%)	161 (0.23%)	.1479	0.77	164 (0.32%)	182 (0.35%)	.3324	0.60	108 (0.43%)	105 (0.42%)	.8368	0.18
US region
Northeast	13,506 (19.15%)	11,831 (16.77%)	<.0001	6.19	10,078 (19.50%)	9,392 (18.18%)	<.0001	3.40	5,014 (19.89%)	4,843 (19.21%)	.0548	1.71
Northcentral	15,891 (22.53%)	19,780 (28.04%)	<.0001	12.71	11,675 (22.59%)	11,609 (22.47%)	.6231	0.31	5,822 (23.09%)	5,413 (21.47%)	<.0001	3.90
South	29,324 (41.58%)	26,359 (37.37%)	<.0001	8.61	21,027 (40.69%)	22,600 (43.74%)	<.0001	6.17	9,936 (39.41%)	11,441 (45.38%)	<.0001	12.10
West	11,724 (16.62%)	12,466 (17.67%)	<.0001	2.79	8,818 (17.07%)	7,974 (15.43%)	<.0001	4.43	4,403 (17.47%)	3,491 (13.85%)	<.0001	9.97
Other	86 (0.12%)	95 (0.13%)	.5032	0.36	75 (0.15%)	98 (0.19%)	.0801	1.09	34 (0.13%)	21 (0.08%)	.0795	1.56
Deyo-modified CCI score	2.17 ± 2.11	2.18 ± 2.10	.5458	0.32	2.31 ± 2.17	2.31 ± 2.17	.9010	0.08	2.32 ± 2.20	2.33 ± 2.18	.7434	0.29
Measures of overall health status
Number of unique medications	11.35 ± 7.04	11.38 ± 7.03	.3307	0.52	12.50 ± 7.32	12.52 ± 7.35	.8146	0.15	12.75 ± 7.48	12.76 ± 7.39	.8134	0.21
Number of inpatient visits per patient	0.80 ± 1.02	0.80 ± 1.00	.8313	0.11	0.86 ± 1.11	0.86 ± 1.09	.8210	0.14	0.81 ± 1.10	0.81 ± 1.09	.8771	0.14
Number of physician office visits per patient	15.70 ± 13.20	15.68 ± 13.54	.7421	0.18	18.34 ± 13.98	18.50 ± 13.61	.0737	1.11	18.28 ± 13.94	18.26 ± 13.30	.8604	0.16
Home oxygen	1 (0.00%)	1 (0.00%)	1.0000	0.00	1 (0.00%)	0 (0.00%)	.3173	0.62	2 (0.01%)	2 (0.01%)	1.0000	0.00
Hip fracture	130 (0.18%)	132 (0.19%)	.9016	0.07	109 (0.21%)	113 (0.22%)	.7881	0.17	46 (0.18%)	52 (0.21%)	.5441	0.54
Measures of stroke risk
CHADS2 score	2.34 ± 1.24	2.35 ± 1.23	.5582	0.31	2.41 ± 1.24	2.40 ± 1.24	.2324	0.74	2.39 ± 1.24	2.39 ± 1.25	.7501	0.28
CHA2DS2 -VASc score	4.15 ± 1.53	4.17 ± 1.52	.1073	0.86	4.23 ± 1.54	4.22 ± 1.52	.4799	0.44	4.20 ± 1.53	4.21 ± 1.54	.5834	0.49
Measures of bleeding risk
HAS-BLED Score	2.68 ± 1.05	2.68 ± 1.08	.9440	0.04	2.77 ± 1.05	2.77 ± 1.05	.7343	0.21	2.75 ± 1.05	2.75 ± 1.05	.7097	0.33
ATRIA Score	2.93 ± 1.93	3.05 ± 2.04	<.0001	6.11	3.01 ± 1.97	3.02 ± 1.94	.3449	0.59	2.97 ± 1.94	3.02 ± 2.00	.0047	2.52
Individual comorbidities
Chronic kidney disease	10,468 (14.84%)	10,383 (14.72%)	.5237	0.34	8,225 (15.92%)	8,297 (16.06%)	.5411	0.38	4,049 (16.06%)	4,090 (16.22%)	.6197	0.44
Chronic obstructive pulmonary disease	20,138 (28.55%)	20,319 (28.81%)	.2866	0.57	15,623 (30.23%)	15,531 (30.06%)	.5329	0.39	7,668 (30.42%)	7,605 (30.17%)	.5415	0.54
Pneumonia	8,603 (12.20%)	8,552 (12.13%)	.6778	0.22	6,829 (13.22%)	6,944 (13.44%)	.2925	0.65	3,446 (13.67%)	3,446 (13.67%)	1.0000	0.00
Cirrhosis/hepatitis	1,569 (2.22%)	1,579 (2.24%)	.8570	0.10	1,274 (2.47%)	1,268 (2.45%)	.9041	0.07	627 (2.49%)	604 (2.40%)	.5069	0.59
Coronary artery disease	31,015 (43.97%)	30,854 (43.75%)	.3877	0.46	23,491 (45.46%)	23,557 (45.59%)	.6801	0.26	11,258 (44.66%)	11,286 (44.77%)	.8020	0.22
Diabetes mellitus	24,760 (35.11%)	24,877 (35.27%)	.5142	0.35	18,988 (36.75%)	18,816 (36.41%)	.2666	0.69	9,306 (36.92%)	9,371 (37.17%)	.5489	0.53
Gastrointestinal bleed	3,594 (5.10%)	3,634 (5.15%)	.6291	0.26	3,014 (5.83%)	2,904 (5.62%)	.1408	0.92	1,415 (5.61%)	1,413 (5.61%)	.9691	0.03
Heart failure	21,825 (30.94%)	21,747 (30.83%)	.6531	0.24	16,762 (32.44%)	16,420 (31.78%)	.0227	1.42	8,096 (32.12%)	8,098 (32.12%)	.9848	0.02
Ischemic stroke/transient ischemic attack	8,730 (12.38%)	8,880 (12.59%)	.2269	0.64	6,731 (13.03%)	6,579 (12.73%)	.1581	0.88	3,158 (12.53%)	3,185 (12.63%)	.7169	0.32
Intracranial bleed	558 (0.79%)	554 (0.79%)	.9041	0.06	462 (0.89%)	449 (0.87%)	.6653	0.27	212 (0.84%)	215 (0.85%)	.8841	0.13
Myocardial infarction	7,764 (11.01%)	7,660 (10.86%)	.3749	0.47	6,041 (11.69%)	6,136 (11.87%)	.3594	0.57	2,933 (11.63%)	2,976 (11.81%)	.5516	0.53
Paraplegia/hemiplegia	2,569 (3.64%)	2,595 (3.68%)	.7124	0.20	2,046 (3.96%)	2,029 (3.93%)	.7858	0.17	971 (3.85%)	975 (3.87%)	.9263	0.08
Psychiatric disorders	20,768 (29.45%)	20,739 (29.40%)	.8655	0.09	16,386 (31.71%)	16,464 (31.86%)	.6023	0.32	8,083 (32.06%)	8,045 (31.91%)	.7167	0.32
Venous thromboembolism	1,896 (2.69%)	1,864 (2.64%)	.5968	0.28	1,569 (3.04%)	1,367 (2.65%)	.0002	2.35	735 (2.92%)	729 (2.89%)	.8736	0.14
Concomitant medication
Beta blockers	44,853 (63.59%)	44,697 (63.37%)	.3883	0.46	33,813 (65.44%)	33,611 (65.05%)	.1870	0.82	16,635 (65.99%)	16,660 (66.09%)	.8141	0.21
Calcium channel blockers	27,457 (38.93%)	27,607 (39.14%)	.4130	0.44	20,912 (40.47%)	20,792 (40.24%)	.4467	0.47	10,182 (40.39%)	10,105 (40.08%)	.4844	0.62
Diuretics	37,846 (53.66%)	38,184 (54.14%)	.0710	0.96	28,640 (55.43%)	28,203 (54.58%)	.0063	1.70	13,967 (55.40%)	13,965 (55.40%)	.9857	0.02
Other antihypertensive	27,969 (39.65%)	28,317 (40.15%)	.0585	1.01	21,176 (40.98%)	20,978 (40.60%)	.2101	0.78	10,182 (40.39%)	10,141 (40.23%)	.7097	0.33
Antihyperlipidemics	42,313 (59.99%)	42,383 (60.09%)	.7036	0.20	32,058 (62.04%)	32,043 (62.01%)	.9234	0.06	15,643 (62.05%)	15,629 (62.00%)	.8978	0.11
Corticosteroids	28,450 (40.34%)	28,397 (40.26%)	.7736	0.15	22,979 (44.47%)	23,282 (45.06%)	.0580	1.18	11,314 (44.88%)	11,289 (44.78%)	.8229	0.20
Antidiabetics	16,536 (23.45%)	16,696 (23.67%)	.3154	0.53	12,607 (24.40%)	12,386 (23.97%)	.1084	1.00	6,288 (24.94%)	6,267 (24.86%)	.8288	0.19
Antiarrhythmic	13,239 (18.77%)	12,778 (18.12%)	.0016	1.69	10,059 (19.47%)	9,602 (18.58%)	.0003	2.25	5,028 (19.95%)	5,006 (19.86%)	.8061	0.22
Ketoconazole	1,429 (2.03%)	1,414 (2.00%)	.7763	0.15	1,242 (2.40%)	1,269 (2.46%)	.5854	0.34	619 (2.46%)	637 (2.53%)	.6070	0.46
NSAIDs	13,160 (18.66%)	13,132 (18.62%)	.8482	0.10	10,642 (20.59%)	10,701 (20.71%)	.6503	0.28	5,076 (20.14%)	5,082 (20.16%)	.9469	0.06
SSRIs	10,366 (14.70%)	10,391 (14.73%)	.8510	0.10	8,146 (15.76%)	8,050 (15.58%)	.4114	0.51	3,980 (15.79%)	3,977 (15.78%)	.9708	0.03
SNRIs	2,683 (3.80%)	2,695 (3.82%)	.8675	0.09	2,151 (4.16%)	2,225 (4.31%)	.2530	0.71	1,085 (4.30%)	1,062 (4.21%)	.6119	0.45
Antiplatelets	11,629 (16.49%)	11,570 (16.40%)	.6717	0.23	8,556 (16.56%)	8,518 (16.48%)	.7503	0.20	3,741 (14.84%)	3,749 (14.87%)	.9202	0.09
Time to index exposure	109.68 ± 201.45	110.53 ± 210.64	.4376	0.41	160.67 ± 249.35	157.55 ± 272.21	.0550	1.19	267.46 ± 342.57	266.90 ± 383.50	.8623	0.15
Index NOAC dosage
150 mg (dabigatran)/20 mg (rivaroxaban)/ 5.0 mg (apixaban)	57,226 (81.14%)				40,854 (79.06%)	36,492 (70.62%)	<.0001	19.55	19,807 (78.57%)	19,066 (75.63%)	<.0001	7.00
75 mg (dabigatran)/15 mg (rivaroxaban)/ 2.5 mg (apixaban)	13,305 (18.86%)				10,819 (20.94%)	15,181 (29.38%)	<.0001	19.55	5,402 (21.43%)	6,143 (24.37%)	<.0001	7.00
All-cause health care costs per patient per month (PPPM)
Inpatient costs	$712 ± $1,330	$908 ± $1,883	<.0001	12.04	$764 ± $1,420	$788 ± $1,457	.0088	1.63	$725 ± $1,393	$720 ± $1,379	.7196	0.32
Outpatient costs	$464 ± $603	$448 ± $596	<.0001	2.59	$534 ± $657	$535 ± $681	.8280	0.14	$529 ± $648	$533 ± $661	.4830	0.62
Physician office costs	$271 ± $417	$252 ± $420	<.0001	4.65	$311 ± $460	$313 ± $481	.4213	0.50	$300 ± $440	$298 ± $419	.5555	0.53
Outpatient hospital costs	$192 ± $378	$196 ± $358	.0457	1.06	$223 ± $413	$221 ± $429	.5809	0.34	$229 ± $420	$235 ± $465	.1083	1.43
ER costs	$28 ± $42	$27 ± $43	.0186	1.25	$31 ± $46	$33 ± $50	<.0001	3.92	$32 ± $55	$35 ± $53	<.0001	5.08
Other costs	$217 ± $773	$251 ± $834	<.0001	4.26	$251 ± $830	$237 ± $797	.0069	1.68	$251 ± $831	$204 ± $724	<.0001	6.10
Pharmacy costs	$231 ± $389	$217 ± $406	<.0001	3.46	$270 ± $431	$268 ± $510	.5072	0.41	$263 ± $446	$267 ± $529	.3401	0.85
Total medical (inpatient + outpatient + other) costs	$1,393 ± $1,970	$1,608 ± $2,465	<.0001	9.64	$1,549 ± $2,124	$1,560 ± $2,141	.4197	0.50	$1,504 ± $2,104	$1,457 ± $2,009	.0090	2.33
Total (total medical + pharmacy) costs	$1,623 ± $2,057	$1,825 ± $2,542	<.0001	8.70	$1,819 ± $2,216	$1,828 ± $2,251	.5286	0.39	$1,767 ± $2,196	$1,723 ± $2,129	.0232	2.02
All-cause healthcare resource utilization
Patients with ≥1 inpatient visit	37,982 (53.85%)	37,997 (53.87%)	.9361	0.04	28,359 (54.88%)	28,728 (55.60%)	.0210	1.44	12,971 (51.45%)	13,053 (51.78%)	.4649	0.65
Patients with ≥1 outpatient visit	70,134 (99.44%)	69,837 (99.02%)	<.0001	4.81	51,568 (99.80%)	51,569 (99.80%)	.9448	0.04	25,156 (99.79%)	25,169 (99.84%)	.1772	1.20
Patients with ≥1 physician office visit	69,593 (98.67%)	68,877 (97.65%)	<.0001	7.56	51,284 (99.25%)	51,314 (99.31%)	.2709	0.68	25,005 (99.19%)	25,051 (99.37%)	.0152	2.16
Patients with ≥1 outpatient hospital visit	57,100 (80.96%)	57,835 (82.00%)	<.0001	2.68	43,976 (85.10%)	43,900 (84.96%)	.5076	0.41	21,460 (85.13%)	21,874 (86.77%)	<.0001	4.73
Patients with ≥1 ER visit	21,726 (30.80%)	22,269 (31.57%)	.0018	1.66	17,794 (34.44%)	18,135 (35.10%)	.0259	1.39	8,831 (35.03%)	9,099 (36.09%)	.0127	2.22
Patients with ≥1 prescription	69,435 (98.45%)	69,254 (98.19%)	.0002	2.00	51,103 (98.90%)	51,153 (98.99%)	.1279	0.95	24,941 (98.94%)	24,983 (99.10%)	.0576	1.69
Number of visits per patient per month (PPPM)
Inpatient visits PPPM	0.07 ± 0.09	0.07 ± 0.08	.8313	0.11	0.07 ± 0.09	0.07 ± 0.09	.8210	0.14	0.07 ± 0.09	0.07 ± 0.09	.8771	0.14
Outpatient visits PPPM	1.67 ± 1.27	1.74 ± 1.35	<.0001	5.20	1.95 ± 1.34	1.96 ± 1.31	.4046	0.52	1.96 ± 1.35	1.98 ± 1.30	.2962	0.93
Physician office visits PPPM	1.31 ± 1.10	1.31 ± 1.13	.7421	0.18	1.53 ± 1.16	1.54 ± 1.13	.0737	1.11	1.52 ± 1.16	1.52 ± 1.11	.8604	0.16
Outpatient hospital visits PPPM	0.36 ± 0.45	0.43 ± 0.54	<.0001	14.14	0.42 ± 0.50	0.42 ± 0.49	.0555	1.19	0.44 ± 0.53	0.45 ± 0.52	.0027	2.68
ER visits PPPM	0.04 ± 0.08	0.04 ± 0.09	<.0001	2.43	0.05 ± 0.09	0.05 ± 0.09	.4028	0.52	0.05 ± 0.14	0.05 ± 0.09	.7711	0.26
Prescriptions PPPM	1.78 ± 1.37	1.84 ± 1.44	<.0001	4.54	2.10 ± 1.43	2.08 ± 1.44	.0136	1.53	2.13 ± 1.45	2.12 ± 1.44	.7217	0.32
Inpatient LOS PPPM	2.68 ± 4.71	3.46 ± 5.93	<.0001	14.65	2.74 ± 4.78	2.90 ± 5.42	<.0001	3.04	2.59 ± 4.62	2.64 ± 4.75	.2132	1.11
Cumulative LOS PPM	3.83 ± 7.64	4.70 ± 8.72	<.0001	10.53	4.15 ± 8.32	4.30 ± 9.02	.0048	1.76	3.94 ± 8.21	3.93 ± 8.25	.8604	0.16

A critical value of p < 0.05 was specified a priori as a threshold of statistical significance for all analyses.

Abbreviations. CCI, Charlson comorbidity index; ER, emergency room; LOS, length of stay; NSAIDs, nonsteroidal anti-inflammatory drugs; SD, standard deviation; SNRIs, serotonin and norepinephrine reuptake inhibitors; SSRIs, selective serotonin reuptake inhibitors; STD, standardized difference.

Generalized-linear-model-adjusted HRU and healthcare costs in the 12-month follow-up period

Multivariate analysis showed that compared to the warfarin and rivaroxaban cohorts, the dabigatran cohort had significantly fewer adjusted PPPM all-cause hospitalizations (0.114 vs. 0.123; 0.111 vs. 0.121), with shorter hospital length of stay (1.462 vs. 1.861 days; 1.424 vs. 1.559 days) respectively. Similarly, dabigatran cohort had significantly fewer PPPM outpatient visits (2.864 vs. 4.201; 2.839 vs. 2.949), physician office (2.13 vs. 2.88; 2.10 vs. 2.18), outpatient hospital (0.72 vs. 1.29; 0.72 vs. 0.75), and ER visits (0.092 vs. 0.102; 0.094 vs. 0.106), as compared to warfarin and rivaroxaban cohorts respectively (all p < .0001; Figure 3). However, compared to the apixaban cohort, the dabigatran cohort had significantly more adjusted PPPM all-cause hospitalizations (0.103 vs. 0.090), outpatient visits (2.780 vs. 2.673), ER visits (0.097 vs. 0.084), and pharmacy visits (3.245 vs. 3.154) (all p < .0001; Figure 3).

Figure 3. Adjusted all-cause HRU among patients with NVAF prescribed warfarin, rivaroxaban or apixaban vs. dabigatran. Abbreviations. ER, emergency room; PPPM, per patient per month. Covariates adjusted included age, sex, race, US region, CHA₂Ds₂-VASc, and HAS-BLED score.

Similarly, the dabigatran cohort incurred significantly lower PPPM all-cause hospitalization ($1,419 vs. $1,634; $1,400 vs. $1,470), ER ($61 vs. $65; $62 vs. $74), other ($336 vs. $512; $311 vs. $328), total medical ($2,719 vs. $3,054; $2,692 vs. $2,837), and total costs ($3,309 vs. $3,362; $3,285 vs. $3,474) than the warfarin and rivaroxaban cohorts, respectively (all p < .05; Figure 4). However, the dabigatran cohort incurred significantly higher PPPM all-cause hospitalization ($1,304 vs. $1,124), outpatient ($865 vs. $844), physician office ($394 vs. $380), ER ($66 vs. $60), other ($309 vs. $294), total medical ($2,574 vs. $2,359), and total costs ($3,192 vs. $2,986) as compared to the apixaban cohort (all p < .05; Figure 4). Similar results were observed in the PSM cohorts, which are presented in Online Table 3.

Figure 4. Adjusted all-cause healthcare costs among patients with NVAF prescribed warfarin, rivaroxaban or apixaban vs. dabigatran. Abbreviations. ER, emergency room; PPPM, per patient per month. Covariates adjusted included age, sex, race, US region, CHA₂Ds₂-VASc, and HAS-BLED score.

Sensitivity analysis

After propensity score matching, 57,206 patients each were included in the dabigatran vs. warfarin cohorts, 40,815 patients each for the dabigatran vs. rivaroxaban cohorts, and 19,734 patients each for dabigatran vs. apixaban cohorts. Overall the results remained consistent with the main analysis showing that the dabigatran cohort had significantly lower total costs PPPM as compared to the warfarin ($3,102 vs. 3,193, p < .0001) and rivaroxaban ($3,045 vs. $3,225, p < .0001) cohorts but had significantly higher costs as compared to apixaban cohort ($2,964 vs. $2,758, p < .0001).

Discussion

This retrospective real-world study examined and compared the HRU and economic burden among newly-diagnosed, treatment-naïve older-aged (≥65 years) NVAF patients, who initiated treatment with either dabigatran or another OAC (warfarin, rivaroxaban, or apixaban). The current analysis is one of the largest real-world studies evaluating all-cause HRU and healthcare costs in the 100% Medicare patient population. This study demonstrated that older-aged NVAF patients treated with dabigatran had significantly lower all-cause HRU and healthcare costs on a PPPM basis, as compared to those treated with warfarin or rivaroxaban. However, HRU and healthcare costs were somewhat higher among dabigatran compared to apixaban patients.

Patients in the dabigatran cohort had significantly lower mean HRU than patients in the warfarin cohort. The dabigatran cohort had an average of 8% fewer hospitalizations, 32% fewer outpatient visits, and 26% fewer physician office visits PPPM. This may be driven by the need for frequent international normalized ratio monitoring and dose adjustments in the warfarin cohort, which is typically required every 3 days, weekly, or monthly^34,35. In addition, comorbidities including liver disease and gastrointestinal illnesses, which were more common among warfarin patients in the current study, may have influenced warfarin dosing and dose adjustments thereby contributing to the higher HRU and medical costs in this cohort³⁶. Since these may not be required for dabigatran, the results of our study showed that patients who initiated dabigatran had significantly lower all-cause PPPM healthcare costs, a finding which is consistent with previous retrospective studies among newly diagnosed NVAF patients^5,23,24,37. However, the index dose or the dose changes in warfarin were not captured in our study. Additionally, the higher HRU and costs in the warfarin cohort could also be attributed to its higher proportion of black patients. As per the Centers for Disease Control and Prevention, African Americans have a higher risk of comorbidities, including hypertension, diabetes, stroke, etc. This could be attributed to some social factors including unemployment, lower socioeconomic status, an unwillingness to see a physician because of higher costs, etc³⁸. Not surprisingly, our study showed that pharmacy costs were significantly higher for dabigatran users compared with warfarin users, which concurs with findings in prior studies^23,24. This could be attributed to the fact that dabigatran is available only as a branded medication, and has a higher unit cost relative to warfarin, which is available as a generic drug^37,39. Nevertheless, these higher pharmacy costs for dabigatran patients were offset by the significantly lower inpatient and medical costs, thus resulting in lower overall healthcare costs in patients treated with dabigatran. Due to these higher pharmacy costs, it was noticed that the difference in the total health care costs was only marginal (-$53). Although the medical costs were $355 lower for dabigatran as compared with warfarin, the magnitude of this difference is ∼50% lower than what was found in a prior study documented by Gilligan and coworkers⁵.

Albeit differences in HRU were somewhat smaller compared to those observed with warfarin, the dabigatran cohort had on average 8% fewer hospitalizations, and 4% fewer outpatient visits PPPM than rivaroxaban users. These findings are also consistent with those of Gilligan et al. and other retrospective studies among newly-diagnosed NVAF patients^5,40. Notably, using MarketScan data, Tepper et al. documented less bleeding with dabigatran as compared to rivaroxaban patients, which could explain in part the lower HRU reported in this cohort⁴¹. Further still, we found that all-cause healthcare costs were significantly lower for dabigatran-treated patients compared to rivaroxaban patients, which supports the observations of Gilligan and colleagues who noted lower inpatient, outpatient, outpatient pharmacy, and total healthcare costs for dabigatran patients compared to rivaroxaban patients⁵.

In the current study, patients treated with dabigatran had significantly higher HRU and costs when compared to apixaban-treated patients. Mixed findings have been reported in the literature among real-world studies comparing HRU and cost outcomes for dabigatran and apixaban patients^5,14,42,43. Deitelzweig et al. and Amin et al. reported that older-aged NVAF patients treated with dabigatran incurred higher all-cause healthcare costs than apixaban patients which is consistent with our study^42,43. Gilligan et al. reported an increase of 9 and 12% in outpatient visits, and outpatient office visits, respectively, among NVAF patients treated with dabigatran as compared to apixaban. However, all-cause total costs were similar between dabigatran vs. apixaban users with no statistical significance observed in spite of the marginally lower costs in dabigatran users⁵. Similar results were observed in another retrospective analysis conducted using OptumInsight research database¹⁴. Amin et al. and Deitelzweig et al. included only older-aged patients where the mean age in both studies ranged between 77 and 78 years^42,43, whereas in the Gilligan et al. study, the mean age was about 68 years, which might explain the differing findings of these studies⁵. Additionally, Amin et al. and Deitelzweig et al. suggested that the lower HRU and costs in the apixaban as compared to dabigatran patients can be attributed to the lower risk of stroke or major bleeding-related hospitalizations in the older-aged NVAF patients treated with apixaban^42,43. However, in a recently published study by Graham et al, it was observed that dabigatran and apixaban had similar risk for thromboembolic stroke and all-cause mortality, but dabigatran was associated with reduced risk for intracranial hemorrhage and increased risk for major extracranial bleeding, all of which suggested a more favorable benefit-harm profile in dabigatran as compared to apixaban⁴⁴. The diverse findings from these real-world studies have made it difficult to draw conclusions on the cost-effectiveness of dabigatran vs. apixaban in the older-aged NVAF patients. Therefore, extensive research is warranted in evaluating these findings, specifically clinical outcome-related resource utilization and costs in older-aged NVAF patients.

The economic results of our analyses provide insight into the substantial cost burden of treating NVAF patients. Particularly, inpatient costs accounted for a major portion of total healthcare costs, approximately 38% to 48%, in all treatment cohorts. Thus, comparisons of resource use and healthcare costs associated with various OACs may better inform clinicians and decision-makers regarding the optimal therapeutic options for managing NVAF patients. Our study suggests that the use of dabigatran over warfarin and rivaroxaban may reduce healthcare costs. However, as the total cost difference between dabigatran vs. the other OACs was marginal (despite costs being statistically significant), these results require additional validation. On this basis, further studies with a longer duration of follow-up are warranted to better understand the cost implications, especially for dabigatran compared with apixaban.

The potential strengths of our study include that, we evaluated a large nationally representative sample from 100% Medicare database, thus contributing to one of the largest real-world studies using data from the 100% Medicare database in the elderly patients with NVAF. Also, our study has followed the on-treatment approach, in which a patient was censored at discontinuation of the index drug or switch to a non-index OAC, thus resulting the economic outcomes to be more attributable to the index treatment. Additionally, the study used robust statistical methodologies like propensity score matching to generate comparable cohorts with similar demographic and clinical characteristics in the baseline.

Although NVAF is more prevalent in older-aged adults, findings of this study may not be generalizable to patients aged <65 years. However, it should be noted that the results from another claims-based study which did not restrict their population to those aged ≥65 years, were consistent with the results of our study⁵. Further still, it should be noted that the costs presented in our study are only crude costs and do not represent the weighted costs (i.e. cost-effectiveness analysis). Also, there may be residual confounding due to unobserved or unrecorded clinical characteristics, this would be an issue with other studies as well. Additionally, our study did not take into consideration the comparison with edoxaban due to its approval in January 2015, and it is unknown if our findings can be generalized to these patients. Finally, this study is subject to the inherent general limitations of any claims dataset, including potential bias from coding errors and identification of medical conditions using International Classification of Diseases, 9th Revision, Clinical Modification diagnosis codes as opposed to clinical evidence. While claims data are valuable for the effective examination of healthcare outcomes, treatment patterns, and costs, they are primarily collected for reimbursement purposes, and not research⁴⁵. Additionally, the presence of a claim for a filled prescription does not indicate the medication was consumed or taken as prescribed. Furthermore, medications filled over-the-counter or provided as samples by the physician are never captured in claims data.

Conclusions

The results of this study compared the HRU and costs of OACs among newly-diagnosed newly-treated older-aged patients aged ≥65 years who presented with NVAF in a real-world setting. NVAF patients initiating treatment with dabigatran incurred significantly lower all-cause HRU and healthcare costs than patients treated with warfarin or rivaroxaban but incurred higher all-cause HRU and healthcare costs than patients treated with apixaban. However, forthcoming studies evaluating the clinical-event-related burden of NVAF are needed to help facilitate our understanding of dabigatran’s cost-effectiveness in older-aged Medicare patients.

Transparency

Declaration of funding

This study was funded by Boehringer Ingelheim Pharmaceuticals.

Declaration of financial/other interests

RC had a Tufts Medical Center postdoctoral fellowship sponsored by Boehringer Ingelheim Pharmaceuticals. JFE, CW, and BOH are employees of Boehringer Ingelheim Pharmaceuticals, the study sponsor. LW is an employee of STATinMED Research, a paid consultant of the study sponsor. HY has no conflicts to disclose.

The peer reviewers on this manuscript have no relevant financial relationships or otherwise to disclose.

Author contributions

RC, JFE, CW, and BOH conceptualized and designed the study, contributed to the interpretation of the data, and substantially contributed to critical revisions of the intellectual content.

LW and HY conceptualized and designed the study, contributed to the acquisition and interpretation of the data, wrote the manuscript, and substantially contributed to critical revisions of the intellectual content.

Previous presentations

These study results have been presented at ACC.19.

Acknowledgements

The authors would like to thank Bonnie Donato, PhD, Swetha Palli, MS, and David A. Singer, PharmD, MS of Boehringer Ingelheim for their valuable contribution in preparation of this manuscript.

Data availability

The raw insurance claims data used for this study originate from Medicare data, which are available from the Centers for Medicare and Medicaid through ResDAC (https://www.resdac.org/).