Healthcare costs of NVAF patients treated with rivaroxaban and apixaban in the US

Abstract

Objectives

To provide the most current assessment of real-world healthcare resource utilization (HRU) and costs among patients with non-valvular atrial fibrillation (NVAF) who newly initiated rivaroxaban and apixaban using a large US database.

Material and methods

A retrospective weighted cohort design was used with healthcare insurance claims from the Optum Clinformatics Data Mart databases (January 2012–December 2018). The index date was defined as the first dispensing of rivaroxaban or apixaban. Adult NVAF patients with an index date on or after 1 January 2016, ≥ 12 months of continuous eligibility before the index date and ≥ 1 month after, and without prior use of oral anticoagulant were included. The observation period spanned from the index date to the earliest of the end of data availability, end of insurance coverage, or death. Inverse probability of treatment weighting (IPTW) was used to adjust for differences in baseline characteristics between cohorts. All-cause healthcare resource utilization (HRU), including hospitalization, emergency room, and outpatient visits, and healthcare costs, including medical and pharmacy costs, were evaluated from the payer’s perspective during the observation period up to 18 and 24 months, separately.

Results

In total, 23,822 rivaroxaban and 53,666 apixaban users were included. After weighting, all baseline characteristics were well balanced between cohorts (mean age: 73.8 years, female: 46.6% in both cohorts). Up to 18 months of follow-up, rivaroxaban users incurred significantly lower total healthcare costs compared to apixaban users (cost difference = −$1,121; p = 0.020), driven by significantly lower rates of outpatient hospital visits and associated costs (cost difference = −$1,579; p < 0.001). Similar results were found in the analysis conducted for up to 24 months of follow-up (total cost difference = ‒$1,111; p = 0.020).

Conclusions

In this large retrospective analysis, patients with NVAF initiated on rivaroxaban incurred significantly lower healthcare costs compared to those initiated on apixaban, which were primarily driven by significantly lower outpatient visits and costs during the 18- and 24-month follow-up periods.

Keywords:

• Non-valvular atrial fibrillation
• healthcare costs
• rivaroxaban
• apixaban
• healthcare resource utilization

JEL CODES:

• I10 – General
• I13 – State and Local Government: Other Expenditure Categories

Background

Atrial fibrillation (AF) is the most common type of cardiac arrhythmic disorder, which often occurs with no signs or symptoms^1,2. In 2010, AF was estimated to affect between 2.7 and 6.1 million people in the US³, a figure projected to increase to more than 12 million people by 2030⁴. This disorder is associated with a considerable increase in mortality^5–8 and a risk of stroke^2,9. Consequently, AF results in a significant economic burden^10,11, a considerable portion of which is attributable to cardiovascular complications such as stroke and systemic embolism^12,13.

Anticoagulants are routinely used as prophylactic antithrombotic therapy in patients with AF, with direct oral anticoagulants (DOACs) now recommended by guidelines as the preferred treatment option for non-valvular atrial fibrillation (NVAF) over vitamin K antagonists (VKA), such as warfarin^14,15. Although the efficacy of DOACs to prevent stroke and systemic embolism in patients with NVAF is well-recognized^14–16, they are also associated with an increased risk of bleeding^17,18. DOACs are preferred to VKAs for their rapid onset of action, fixed dosing regimens, fewer food and drug interactions, and limited need for laboratory monitoring¹⁵. The factor Xa inhibitors, rivaroxaban and apixaban, are two DOACs that were approved by the US Food and Drug Administration (FDA) in 2011 and 2012, respectively, for the reduction of stroke and systemic embolism in patients with NVAF^19,20, and are the most preferred treatment choices in the US.

Several studies showed that patients initiated on DOACs incur lower healthcare costs than those initiated on warfarin^21–24. However, studies that directly compared healthcare costs between patients treated with rivaroxaban and apixaban report mixed results^25–28. For example, a retrospective study that included commercially-insured patients who received rivaroxaban or apixaban prior to 2015 found no significant difference in all-cause healthcare costs between patients initiated on apixaban versus rivaroxaban²⁵. Since these studies evaluated a period shortly following FDA approval of both medications (i.e. data from 2012 to 2015), such analyses may also not be representative of current cost trends in the US. Furthermore, some studies evaluated specific subpopulations (e.g. using US Department of Defense data)²⁶, thereby reaching conclusions that may not be generalizable. Finally, studies comparing healthcare resource utilization (HRU) in patients with NVAF treated with rivaroxaban or apixaban often restrict their analyses to hospitalization events^25,28, overlooking important factors such as emergency room (ER) and outpatient visits. The present study thus aims to fill a knowledge gap by providing an updated comprehensive assessment of the real-world HRU and costs among patients with NVAF who were treatment-naïve to oral anticoagulants and newly-initiated on rivaroxaban or apixaban treatments using a large claims US database.

Methods

Data source

Healthcare insurance claims from the Optum Clinformatics Data Mart database were used, with data ranging from 1 January 2012 to 31 December 2018. Optum Clinformatics Data Mart covers 13 million annual lives of UnitedHealth Group members in all census regions in the US. It contains historical data on patient demographics, insurance coverage (i.e. commercial and Medicare), dates of eligibility and death, claims for inpatient and outpatient visits, costs of services, and laboratory tests and results. Data are de-identified and comply with the patient confidentiality requirements of the Health Insurance Portability and Accountability Act (HIPAA).

Study design

A retrospective weighted cohort design was used. The index date was defined as the date of the first dispensing of rivaroxaban or apixaban (i.e. the index medication). The baseline period was defined as the 12 months prior to the index date. The observation period (follow-up period) spanned from the index date to the end of data availability, end of insurance coverage, or death, whichever occurred first. Study outcomes were evaluated over two distinct follow-up periods, censored at 18 months and 24 months of observation, separately.

Study population

In order to evaluate the most recent trends in HRU and costs, only patients with an index date on or after 1 January 2016 were included in the study population. Additional inclusion criteria were ≥ 1 diagnosis of AF during the baseline period or on the index date, ≥ 18 years of age at the index date, and ≥ 12 months of continuous eligibility prior to the index date and ≥ 1 month of continuous eligibility after the index date.

Patients were also required to be newly-initiated on oral anticoagulants (i.e. treatment-naïve to oral anticoagulants); therefore, patients with ≥ 1 pharmacy claim for an oral anticoagulant other than the index medication on the index date or during the baseline period were excluded. Patients were also excluded if they had ≥ 1 claim for mitral stenosis, mechanical heart-valve procedure, or organ/tissue transplant during the baseline period, ≥ 1 diagnosis of VTE during the baseline period, or ≥ 1 diagnosis for pregnancy at any time during the study period.

Included patients were classified into two mutually exclusive cohorts based on the index anticoagulant received.

Study outcomes

Study outcomes included all-cause HRU and costs. HRU was stratified into hospitalizations, ER visits, outpatient visits, and other visits (e.g. home services and hospice). Outpatient visits were further broken down into outpatient hospital visits and office/physician visits. Healthcare costs were stratified into medical and pharmacy costs, and medical costs were further broken down into hospitalization, ER visit, outpatient visit, and other visit costs. Healthcare costs were evaluated from the payer’s perspective (i.e. excluding patient-paid costs) during the observation period, and were inflation-adjusted to 2019 US dollars based on the medical care component of the Consumer Price Index.

Statistical analysis

To adjust for differences in baseline characteristics between the rivaroxaban and apixaban cohorts, inverse probability of treatment weighting (IPTW) based on the propensity score was used. The propensity score was derived from a multivariable logistic regression model conditional on baseline covariates, including age, sex, year of index date, US region, type of insurance plan, Quan-Charlson comorbidity index score, CHA₂DS₂-VASc score, HAS-BLED score, selected NVAF-related comorbidities (i.e. those with a prevalence of ≥ 5%), baseline risk factors for stroke/systemic embolism and bleeding events (i.e. those with a prevalence of ≥ 5%), all-cause baseline HRU, and all-cause baseline healthcare costs (hospitalization, ER visit, outpatient hospital visit, office/physician visit, other visit, and pharmacy costs).

Patient demographics and baseline clinical characteristics for the rivaroxaban and apixaban cohorts were summarized using frequencies and proportions for categorical variables, and means, standard deviations (SDs), and medians for continuous variables. Baseline characteristics were compared for unweighted and weighted cohorts using standardized differences (std. diff.), with a threshold of < 10% considered a negligible imbalance between covariates²⁹.

All-cause HRU for the rivaroxaban and apixaban cohorts were reported including the frequency and rate (reported per patient per year [PPPY] by dividing the number of HRU events observed over the observation period by the patient-years of observation) of visits. Weighted rate ratios were calculated from Poisson regression models, with 95% confidence intervals (CIs) and p-values generated using non-parametric bootstrap procedures. All-cause healthcare costs were also reported PPPY by dividing the costs incurred over the observation period by the patient-years of observation. Mean cost differences between cohorts were calculated, and 95% CIs and p-values were generated using non-parametric bootstrap procedures.

Results

Baseline characteristics

A total of 23,822 and 53,666 patients newly initiated on rivaroxaban and apixaban on or after 1 January 2016 were included in the study (Figure 1). Prior to applying IPTW weights, patients in the rivaroxaban cohort were younger than patients in the apixaban cohort (72.3 vs 74.5; std. diff. = 20.8%), and had lower Quan-Charlson comorbidity index (2.50 vs 3.02), CHA₂DS₂-VASc (2.43 vs 2.75), and HAS-BLED (1.48 vs 1.59) scores. The rivaroxaban cohort also had lower rates of all-cause hospitalizations (0.51 vs 0.66; std. diff. = 17.5%), and lower total healthcare costs ($26,908 vs $34,198; std. diff. = 13.7%). After weighting, all baseline characteristics were well balanced between cohorts (std. diff < 10%; Table 1). Mean age was 73.8 years and 46.6% were female in both cohorts. The Quan-Charlson comorbidity index score was 2.86, CHA₂DS₂-VASc score was 2.65, and HAS-BLED score was 1.55 in both cohorts. The all-cause total healthcare costs at baseline were well balanced ($32,022 vs $31,970; std. diff. = 0.1%). The most common selected NVAF-related comorbidities were diabetes (35%), renal disease (35%), and congestive heart failure (32%), while the most commonly observed risk factors for stroke/systemic embolism and bleeding events were hypertension (86%), hyperlipidemia (75%), and chronic kidney disease (32%; Table 2).

Figure 1. Patient disposition. AF, atrial fibrillation; VTE, venous thromboembolism. Note. 1. Patients initiated on treatment with both rivaroxaban and apixaban on the index date were classified as apixaban, and subsequently excluded (n = 16).

Table 1. Baseline demographics and clinical characteristics – unweighted and weighted rivaroxaban and apixaban cohorts.

Characteristics	Unweighted Cohorts			Weighted Cohorts^a
	Rivaroxaban	Apixaban	Std. diff.^b,c	Rivaroxaban	Apixaban	Std. diff.^b,c
	(N = 23,822)	(N = 53,666)	(%)	(N = 23,822)	(N = 53,666)	(%)
Demographics^d
Age, years, mean ± SD [median]	72.3 ± 10.6 [73]	74.5 ± 10.0 [75]	20.8	73.8 ± 10.1 [75]	73.8 ± 10.3 [75]	0.3
Sex, female, n (%)	10,332 (43.4)	25,774 (48.0)	9.3	11,102 (46.6)	25,003 (46.6)	0.0
Year of index date,^d n (%)
2016	5,743 (24.1)	11,104 (20.7)	8.2	5,193 (21.8)	11,673 (21.8)	0.1
2017	9,132 (38.3)	17,603 (32.8)	11.6	8,189 (34.4)	18,507 (34.5)	0.2
2018	8,947 (37.6)	24,959 (46.5)	18.1	10,439 (43.8)	23,486 (43.8)	0.1
Region,^d n (%)
South	9,253 (38.8)	23,703 (44.2)	10.8	10,139 (42.6)	22,850 (42.6)	0.0
West	5,900 (24.8)	12,851 (23.9)	1.9	5,785 (24.3)	12,998 (24.2)	0.2
Midwest	5,902 (24.8)	11,347 (21.1)	8.6	5,276 (22.1)	11,939 (22.2)	0.2
Northeast	2,685 (11.3)	5,626 (10.5)	2.5	2,533 (10.6)	5,746 (10.7)	0.2
Unknown	82 (0.3)	139 (0.3)	1.6	89 (0.4)	133 (0.2)	2.3
Insurance plan type,^d n (%)
Medicare	18,073 (75.9)	43,921 (81.8)	14.6	19,017 (79.8)	42,898 (79.9)	0.3
Commercial insurance	5,749 (24.1)	9,745 (18.2)	14.6	4,805 (20.2)	10,768 (20.1)	0.3
Quan-Charlson comorbidity index,^e mean ± SD [median]	2.50 ± 2.35 [2]	3.02 ± 2.56 [2]	21.3	2.86 ± 2.51 [2]	2.86 ± 2.51 [2]	0.1
CHA2DS2-VASc score,^e mean ± SD [median]	2.43 ± 1.45 [2]	2.75 ± 1.50 [3]	21.2	2.65 ± 1.51 [2]	2.65 ± 1.49 [2]	0.0
HAS-BLED score,^e mean ± SD [median]	1.48 ± 0.90 [1]	1.59 ± 0.92 [1]	12.7	1.55 ± 0.92 [1]	1.55 ± 0.92 [1]	0.0
Baseline use of injectable anticoagulants,^e n (%)
Unfractionated heparins	1,145 (4.8)	3,220 (6.0)	5.3	1,332 (5.6)	3,040 (5.7)	0.3
Low molecular weight heparins	943 (4.0)	2,731 (5.1)	5.4	1,063 (4.5)	2,634 (4.9)	2.1
Fondaparinux	1 (0.0)	6 (0.0)	0.8	1 (0.0)	6 (0.0)	0.8
All-cause healthcare resource utilization,^e mean ± SD [median]
Hospitalizations	0.51 ± 0.82 [0]	0.66 ± 0.95 [0]	17.5	0.62 ± 0.90 [0]	0.61 ± 0.92 [0]	0.1
ER visits	0.95 ± 2.40 [0]	1.05 ± 2.77 [0]	3.9	1.04 ± 2.81 [0]	1.02 ± 2.62 [0]	0.6
Outpatient visits	14.6 ± 16.2 [11]	16.1 ± 19.7 [12]	8.7	15.7 ± 18.0 [11]	15.7 ± 18.8 [11]	0.0
Outpatient hospital visits	4.8 ± 11.7 [2]	6.1 ± 15.9 [2]	8.7	5.7 ± 13.8 [2]	5.7 ± 14.8 [2]	0.2
Office/physician visits	9.7 ± 10.0 [7]	10.1 ± 10.3 [8]	3.5	10.0 ± 10.2 [8]	10.0 ± 10.2 [8]	0.2
Other visits^f	5.1 ± 10.0 [1]	5.8 ± 11.7 [1]	6.6	5.6 ± 10.9 [1]	5.6 ± 11.3 [1]	0.2
All-cause healthcare costs,^e,g $US 2019, mean ± SD
Total costs (medical + pharmacy)	$26,908 ± 46,636	$34,198 ± 58,949	13.7	$32,022 ± 57,214	$31,970 ± 54,463	0.1
Total medical costs	$24,331 ± 44,999	$31,346 ± 57,224	13.6	$29,258 ± 55,537	$29,205 ± 52,748	0.1
Hospitalization costs	$9,895 ± 29,938	$13,811 ± 30,706	12.9	$12,577 ± 34,439	$12,609 ± 29,326	0.1
ER visit costs	$3,661 ± 11,706	$4,514 ± 19,407	5.3	$4,347 ± 15,158	$4,265 ± 17,275	0.5
Outpatient visit costs	$9,124 ± 26,121	$10,869 ± 38,019	5.3	$10,318 ± 32,044	$10,330 ± 34,606	0.0
Outpatient hospital visit costs	$6,691 ± 21,976	$8,402 ± 36,388	5.7	$7,858 ± 29,814	$7,883 ± 32,660	0.1
Office/physician visit costs	$2,433 ± 13,028	$2,467 ± 9,492	0.3	$2,460 ± 10,465	$2,447 ± 10,070	0.1
Other visit costs^f	$1,651 ± 7,686	$2,152 ± 13,283	4.6	$2,017 ± 11,896	$2,000 ± 11,740	0.1
Pharmacy costs	$2,577 ± 8,990	$2,852 ± 9,275	3.0	$2,763 ± 9,540	$2,765 ± 9,035	0.0

ER, emergency room; HAS-BLED, hypertension, abnormal renal and liver function, stroke, bleeding, labile international normalized ratio, elderly, drugs or alcohol; NVAF, non-valvular atrial fibrillation; SD, standard deviation; Std. diff., standardized difference.

^aRivaroxaban and apixaban patients were weighted using the inverse probability of treatment weighting approach based on the propensity score. Variables used in the propensity score calculation included the following: age, sex, year of index date, US region, type of insurance plan, Quan-Charlson comorbidity index score, CHA₂DS₂-VASc score, HAS-BLED score, selected NVAF-related comorbidities (those with prevalence ≥ 5%), baseline risk factors for stroke/systemic embolism and bleeding events (those with prevalence ≥ 5%), all-cause baseline HRU and costs.

^bFor continuous variables, the standardized difference was calculated by dividing the absolute difference in means of the control and the case by the pooled standard deviation of both groups. The pooled standard deviation is the square root of the average of the squared standard deviations.

^cFor dichotomous variables, the standardized difference was calculated using the following equation where P is the respective proportion of participants in each group: |(P_case−P_control)|/√[(P_case(1 − P_case)+P_control(1 − P_control))/2].

^dEvaluated at the index date.

^eEvaluated during the 12 months prior to the index date, excluding the index date.

^f Includes visits such as home services and hospice.

^gCosts were inflation-adjusted to US$2019 using the US Medical Care consumer price index from the Bureau of Labor Statistics from the US Department of Labor.

Table 2. NVAF-related comorbidities and risk factors for stroke/systemic embolism and major bleeding – unweighted and weighted rivaroxaban and apixaban cohorts.

Characteristics	Unweighted Cohorts			Weighted Cohorts^a
	Rivaroxaban	Apixaban	Std. diff.^b,c	Rivaroxaban	Apixaban	Std. diff.^b,c
	(N = 23,822)	(N = 53,666)	(%)	(N = 23,822)	(N = 53,666)	(%)
Selected NVAF-related comorbidities,^d n (%)
Diabetes	7,851 (33.0)	19,154 (35.7)	5.8	8,317 (34.9)	18,707 (34.9)	0.1
Renal disease	6,941 (29.1)	19,909 (37.1)	16.9	8,242 (34.6)	18,588 (34.6)	0.1
Congestive heart failure	6,769 (28.4)	18,103 (33.7)	11.5	7,661 (32.2)	17,227 (32.1)	0.1
Myocardial infarction	2,666 (11.2)	7,965 (14.8)	10.8	3,269 (13.7)	7,359 (13.7)	0.0
Diverticulosis	1,970 (8.3)	4,763 (8.9)	2.2	2,073 (8.7)	4,664 (8.7)	0.0
Percutaneous coronary intervention	522 (2.2)	1,515 (2.8)	4.0	638 (2.7)	1,416 (2.6)	0.2
History of previous GI bleed
Lower GI bleed	400 (1.7)	1,070 (2.0)	2.3	435 (1.8)	1,038 (1.9)	0.8
Upper GI bleed	37 (0.2)	137 (0.3)	2.2	51 (0.2)	126 (0.2)	0.4
Coronary bypass graft	195 (0.8)	771 (1.4)	5.9	246 (1.0)	714 (1.3)	2.8
Helicobacter pylori infection	78 (0.3)	179 (0.3)	0.1	84 (0.4)	171 (0.3)	0.6
Baseline risk factors for stroke/systemic embolism and bleeding events,^d n (%)
Hypertension	19,991 (83.9)	46,982 (87.5)	10.4	20,551 (86.3)	46,358 (86.4)	0.3
Hyperlipidemia	17,095 (71.8)	40,725 (75.9)	9.4	17,749 (74.5)	40,019 (74.6)	0.1
Chronic kidney disease	6,483 (27.2)	17,965 (33.5)	13.6	7,493 (31.5)	16,924 (31.5)	0.2
Obesity	6,184 (26.0)	13,309 (24.8)	2.7	6,030 (25.3)	13,514 (25.2)	0.3
Heart failure	5,896 (24.8)	16,232 (30.2)	12.3	6,809 (28.6)	15,322 (28.6)	0.1
Excessive fall risk (i.e. Parkinson’s disease, dementia, and other mental disorders)	5,176 (21.7)	12,948 (24.1)	5.7	5,591 (23.5)	12,557 (23.4)	0.2
Anemia	4,897 (20.6)	13,667 (25.5)	11.7	5,714 (24.0)	12,854 (24.0)	0.1
Chronic obstructive pulmonary disease	4,438 (18.6)	11,546 (21.5)	7.2	4,923 (20.7)	11,070 (20.6)	0.1
Depression	3,941 (16.5)	9,870 (18.4)	4.9	4,251 (17.8)	9,564 (17.8)	0.1
Alcohol abuse	3,239 (13.6)	7,665 (14.3)	2.0	3,355 (14.1)	7,558 (14.1)	0.0
Family history of cardiovascular disease	1,856 (7.8)	4,956 (9.2)	5.2	2,114 (8.9)	4,724 (8.8)	0.3
Hepatic disease	1,355 (5.7)	3,337 (6.2)	2.2	1,437 (6.0)	3,249 (6.1)	0.1
Thrombocytopenia (i.e. low platelet count)	866 (3.6)	2,517 (4.7)	5.3	1,023 (4.3)	2,374 (4.4)	0.6
Left ventricular dysfunction	768 (3.2)	2,025 (3.8)	3.0	839 (3.5)	1,947 (3.6)	0.6
Peptic ulcer	379 (1.6)	1,056 (2.0)	2.8	441 (1.9)	994 (1.9)	0.0
Bleeding diathesis	10 (0.0)	47 (0.1)	1.8	11 (0.0)	44 (0.1)	1.5

NVAF, non-valvular atrial fibrillation; Std. diff., standardized difference.

^aRivaroxaban and apixaban patients were weighted using the inverse probability of treatment weighting approach based on the propensity score. Variables used in the propensity score calculation included the following: age, sex, year of index date, US region, type of insurance plan, Quan-Charlson comorbidity index score, CHA₂DS₂-VASc score, HAS-BLED score, selected NVAF-related comorbidities (those with prevalence ≥ 5%), baseline risk factors for stroke/systemic embolism and bleeding events (those with prevalence ≥ 5%), all-cause baseline HRU and costs.

^bFor continuous variables, the standardized difference was calculated by dividing the absolute difference in means of the control and the case by the pooled standard deviation of both groups. The pooled standard deviation is the square root of the average of the squared standard deviations.

^cFor dichotomous variables, the standardized difference was calculated using the following equation, where P is the respective proportion of participants in each group: |(P_case−P_control)|/√[(P_case(1 − P_case)+P_control(1 − P_control))/2].

^dEvaluated during the 12 months prior to the index date, excluding the index date.

All-cause HRU

Unweighted analysis

Patients were observed for a mean duration of 368 and 340 days among the unweighted rivaroxaban and apixaban cohorts, respectively, when censoring follow-up at 18 months (Table 3). Patients initiated on rivaroxaban had significantly lower rates of hospitalizations PPPY (rate ratio [95% CI] = 0.87 [0.84–0.90]; p < 0.001), ER visits (rate ratio [95% CI] = 0.91 [0.88–0.94]; p < 0.001), and outpatient visits (rate ratio [95% CI] = 0.90 [0.88–0.91]; p < 0.001) compared to patients initiated on apixaban, with similar results observed over 24 months of follow-up.

Table 3. All-cause healthcare resource utilization of the rivaroxaban and apixaban cohorts – up to 18 and 24 months follow-up.

HRU measures	Frequency of HRU		Rate (PPPY)		Rate Ratio^a (95% CI)^b [A]/[B]	p-value^b
	Rivaroxaban	Apixaban	Rivaroxaban [A]	Apixaban [B]
	(N = 23,822)	(N = 53,666)	(N = 23,822)	(N = 53,666)
Up to 18 Months
Unweighted Cohort
Observation period,^c days, mean ± SD [median]	368 ± 176 [397]	340 ± 179 [340]
Total person-years	24,047	50,045
Hospitalizations	11,213	26,779	0.47	0.54	0.87 (0.84, 0.90)	<0.001
ER visits	24,072	55,052	1.00	1.10	0.91 (0.88, 0.94)	<0.001
Outpatient visits	465,222	1,077,633	19.3	21.5	0.90 (0.88, 0.91)	<0.001
Outpatient hospital visits	180,876	477,999	7.5	9.6	0.79 (0.76, 0.81)	<0.001
Office/physician visits	284,346	599,634	11.8	12.0	0.99 (0.97, 1.00)	0.112
Other visits^d	165,197	388,312	6.9	7.8	0.89 (0.86, 0.92)	<0.001
Weighted Cohort
Observation period,^c days, mean ± SD [median]	351 ± 177 [355]	348 ± 179 [354]
Total person-years	27,654	61,696
Hospitalizations	13,042	27,904	0.47	0.45	1.04 (1.00, 1.09)	0.072
ER visits	27,622	60,532	1.00	0.98	1.02 (0.98, 1.05)	0.377
Outpatient visits	539,750	1,232,203	19.5	20.0	0.98 (0.96, 0.99)	0.008
Outpatient hospital visits	214,828	512,514	7.8	8.3	0.94 (0.90, 0.97)	<0.001
Office/physician visits	324,922	719,689	11.7	11.7	1.01 (0.99, 1.02)	0.441
Other visits^d	203,728	455,353	7.4	7.4	1.00 (0.97, 1.03)	0.850
Up to 24 Months
Unweighted Cohort
Observation period,^e days, mean ± SD [median]	419 ± 234 [397]	382 ± 233 [340]
Total person-years	27,330	56,160
Hospitalizations	12,400	29,240	0.45	0.52	0.87 (0.84, 0.90)	<0.001
ER visits	26,745	60,286	0.98	1.07	0.91 (0.88, 0.94)	<0.001
Outpatient visits	516,550	1,183,156	18.9	21.1	0.90 (0.88, 0.91)	<0.001
Outpatient hospital visits	198,499	519,457	7.3	9.2	0.79 (0.76, 0.81)	<0.001
Office/physician visits	318,051	663,699	11.6	11.8	0.98 (0.97, 1.00)	0.076
Other visits^d	185,461	429,926	6.8	7.7	0.89 (0.86, 0.92)	<0.001
Weighted Cohort
Observation period,^e days, mean ± SD [median]	396 ± 233 [355]	392 ± 234 [354]
Total person-years	37,750	83,816
Hospitalizations	16,945	36,327	0.45	0.43	1.04 (0.99, 1.08)	0.104
ER visits	36,313	79,302	0.96	0.95	1.02 (0.98, 1.06)	0.429
Outpatient visits	711,456	1,623,067	18.8	19.4	0.97 (0.96, 0.99)	<0.001
Outpatient hospital visits	276,528	661,875	7.3	7.9	0.93 (0.89, 0.96)	<0.001
Office/physician visits	434,928	961,192	11.5	11.5	1.00 (0.99, 1.02)	0.669
Other visits^d	272,899	606,080	7.2	7.2	1.00 (0.97, 1.03)	0.962

CI, confidence interval; ER, emergency room; HRU, healthcare resource utilization; LOS, length of stay; PPPY, per patient per year.

^aRate ratios were obtained from Poisson regression models with log-link.

^bConfidence intervals and p-values were calculated using non-parametric bootstrap procedures with 499 replications.

^cThe observation period spanned from the index date to the earliest of: 18 months, end of continuous enrollment, death, or end of data availability.

^dIncludes visits such as home services and hospice.

^eThe observation period spanned from the index date to the earliest of: 24 months, end of continuous enrollment, death, or end of data availability.

Weighted analysis

Following IPTW, rivaroxaban and apixaban patients had mean observation periods of 351 days and 348 days, respectively, when censoring follow-up at 18 months (Table 3). Compared to apixaban, patients initiated on rivaroxaban had significantly lower rates of outpatient visits PPPY (rate ratio [95% CI] = 0.98 [0.96–0.99]; p = 0.008), which were mainly driven by significantly lower rates of outpatient hospital visits (rate ratio [95% CI] = 0.94 [0.90–0.97]; p < 0.001), with no significant differences in rates of hospitalizations, ER visits, office/physician visits, and other visits (Table 3). Similar results were also observed up to 24 months of follow-up.

To further describe outpatient hospital visits between weighted rivaroxaban and apixaban cohorts during follow-up, the distribution of physician specialties associated with outpatient hospital visits was compared and found to be similar between cohorts (i.e. all std. diff. < 10%; Supplementary Table S1). The distribution of primary diagnoses associated with outpatient hospital visits was also similar between cohorts with the exception of chronic kidney disease, which was more frequent in the apixaban cohort than the rivaroxaban cohort (5.2% vs 1.6%; std. diff. = 20.0%; Supplementary Table S2).

All-cause healthcare costs

Unweighted analysis

In the analysis conducted up to 18 months of follow-up, patients initiated on rivaroxaban incurred significantly lower total (medical and pharmacy) healthcare costs PPPY compared to patients initiated on apixaban (cost difference: −$4,607; p < 0.001; Table 4). The healthcare cost difference was mainly driven by significantly lower medical costs (cost difference: −$4,118; p < 0.001), which accounted for 89% of the total cost difference. The primary driver of the medical cost difference was outpatient hospital visit costs (cost difference: −$1,861; p < 0.001).

Table 4. Healthcare cost components of the rivaroxaban and apixaban cohorts – up to 18 and 24 months follow-up.

Healthcare costs,^a $US 2019	Rivaroxaban [A]	Apixaban [B]	Cost difference	p-value^b
			(95% CI)^b
	(N = 23,822)	(N = 53,666)	[A] - [B]
Up to 18 months
Unweighted Cohorts
Observation period,^c days, mean ± SD [median]	368 ± 176 [397]	340 ± 179 [340]
All-cause total healthcare costs, PPPY, mean ± SD	$36,251 ± 58,451	$40,858 ± 73,162	−4,607 (−5,579; −3,734)	<0.001
Total medical costs	$30,053 ± 56,375	$34,171 ± 71,206	−4,118 (−5,076; −3,275)	<0.001
Hospitalization costs	$9,534 ± 32,239	$10,737 ± 35,716	−1,203 (−1,712; −772)	<0.001
ER visit costs	$3,551 ± 13,671	$4,104 ± 19,234	−554 (−779; −330)	<0.001
Outpatient visit costs	$14,804 ± 37,327	$16,717 ± 49,788	−1,913 (−2,546; −1,271)	<0.001
Outpatient hospital visit costs	$11,839 ± 33,055	$13,700 ± 46,443	−1,861 (−2,416; −1,341)	<0.001
Office/physician visit costs	$2,965 ± 14,723	$3,017 ± 15,847	−52 (−269; 201)	0.725
Other visit costs^d	$2,164 ± 8,968	$2,613 ± 13,553	−449 (−614; −265)	<0.001
Pharmacy costs	$6,197 ± 11,313	$6,687 ± 11,241	−489 (−675; −288)	<0.001
Weighted Cohorts
Observation period,^c days, mean ± SD [median]	351 ± 177 [355]	348 ± 179 [354]
All-cause total healthcare costs, PPPY, mean ± SD	$38,464 ± 62,881	$39,585 ± 69,247	−1,121 (−2,027; −249)	0.020
Total medical costs	$32,045 ± 60,558	$32,994 ± 67,291	−949 (−1,860; −92)	0.036
Hospitalization costs	$10,600 ± 34,427	$10,204 ± 34,826	396 (−119; 859)	0.112
ER visit costs	$3,996 ± 15,384	$3,892 ± 17,585	104 (−133; 370)	0.421
Outpatient visit costs	$14,933 ± 38,192	$16,415 ± 46,815	−1,482 (−2,127; −885)	<0.001
Outpatient hospital visit costs	$11,874 ± 33,651	$13,453 ± 43,449	−1,579 (−2,134; −1,046)	<0.001
Office/physician visit costs	$3,059 ± 15,167	$2,962 ± 15,373	97 (−115; 342)	0.409
Other visit costs^d	$2,516 ± 13,287	$2,484 ± 12,476	33 (−235; 321)	0.794
Pharmacy costs	$6,419 ± 12,607	$6,591 ± 11,038	−172 (−350; 32)	0.108
Up to 24 months
Unweighted Cohorts
Observation period,^edays, mean ± SD [median]	419 ± 234 [397]	382 ± 233 [340]
All-cause total healthcare costs, PPPY, mean ± SD	$35,310 ± 56,073	$39,847 ± 70,512	−4,538 (−5,454; −3,672)	<0.001
Total medical costs	$29,151 ± 53,912	$33,221 ± 68,533	−4,070 (−4,938; −3,221)	<0.001
Hospitalization costs	$9,308 ± 30,719	$10,522 ± 34,270	−1,215 (−1,650; −769)	<0.001
ER visit costs	$3,478 ± 13,000	$4,017 ± 18,559	−539 (−763; −316)	<0.001
Outpatient visit costs	$14,219 ± 35,666	$16,092 ± 47,736	−1,874 (−2,498; −1,277)	<0.001
Outpatient hospital visit costs	$11,282 ± 31,210	$13,127 ± 44,493	−1,844 (−2,396; −1,356)	<0.001
Office/physician visit costs	$2,936 ± 14,729	$2,965 ± 15,213	−29 (−254; 239)	0.838
Other visit costs^d	$2,146 ± 8,633	$2,589 ± 13,258	−443 (−623; −255)	<0.001
Pharmacy costs	$6,159 ± 11,122	$6,626 ± 11,148	−467 (−648; −271)	<0.001
Weighted Cohorts
Observation period,^e days, mean ± SD [median]	396 ± 233 [355]	392 ± 234 [354]
All-cause total healthcare costs, PPPY, mean ± SD	$37,482 ± 60,515	$38,594 ± 66,681	−1,111 (−1,983; −255)	0.020
Total medical costs	$31,110 ± 58,132	$32,058 ± 64,695	−948 (−1,818; −69)	0.040
Hospitalization costs	$10,332 ± 32,890	$10,000 ± 33,363	333 (−158; 761)	0.140
ER visit costs	$3,914 ± 14,693	$3,810 ± 16,996	104 (−128; 353)	0.381
Outpatient visit costs	$14,381 ± 36,754	$15,786 ± 44,810	−1,405 (−1,990; −822)	<0.001
Outpatient hospital visit costs	$11,351 ± 31,936	$12,873 ± 41,537	−1,522 (−2,011; −1,000)	<0.001
Office/physician visit costs	$3,030 ± 15,368	$2,913 ± 14,761	117 (−108; 391)	0.321
Other visit costs^d	$2,482 ± 12,696	$2,462 ± 12,199	20 (−226; 283)	0.886
Pharmacy costs	$6,372 ± 12,324	$6,536 ± 10,979	−164 (−335; 26)	0.112

CI, confidence interval; ER, emergency room; PPPY, per patient per year; SD, standard deviation.

^aCosts were inflation-adjusted to US$2019 using the US Medical Care consumer price index from the Bureau of Labor Statistics from the US Department of Labor.

^bConfidence intervals and p-values were generated based on non-parametric bootstrap procedures with 499 replications.

^cThe observation period spanned from the index date to the earliest of: 18 months, end of continuous enrollment, death, or end of data availability.

^dIncludes visits such as home services and hospice.

^eThe observation period spanned from the index date to the earliest of: 24 months, end of continuous enrollment, death, or end of data availability.

In the analysis conducted up to 24 months of follow-up, similar trends were observed whereby patients initiated on rivaroxaban incurred significantly lower total healthcare costs PPPY compared to patients initiated on apixaban (cost difference: −$4,538; p < 0.001; Table 4). This difference was also largely attributable to the lower medical costs associated with rivaroxaban (cost difference: −$4,070; p < 0.001).

Weighted analysis

Following IPTW, the analysis conducted up to 18 months of follow-up showed that patients initiated on rivaroxaban incurred significantly lower total healthcare costs PPPY compared to patients initiated on apixaban (cost difference: −$1,121; p = 0.020). Similar to the unadjusted analysis, the difference in healthcare costs was mainly driven by significantly lower medical costs (cost difference: −$949; p = 0.036), which accounted for 85% of the total cost difference. The primary driver of the medical cost difference was outpatient hospital visit costs (cost difference: −$1,579; p < 0.001; Table 4). All other healthcare cost components (i.e. hospitalization, ER, office/physician visit, other visit, and pharmacy costs) did not significantly differ between NVAF patients initiated on rivaroxaban and apixaban (all p-values > 0.05).

In the weighted 24-month follow-up analysis, the total cost of care was slightly lower compared to that observed in the 18-month follow-up analysis, but the difference between cohorts remained similar. Patients initiated on rivaroxaban had significantly lower total healthcare costs compared to patients initiated on apixaban (cost difference: −$1,111; p = 0.020). Similar to the 18-month analysis, the difference in healthcare costs was primarily driven by significantly lower outpatient hospital visit costs (cost difference: −$1,522; p < 0.001; Table 4).

Discussion

This retrospective study used claims data from a large nationally representative database to compare recent trends in cost of care between oral anticoagulant-naïve patients with NVAF who were initiated on rivaroxaban or apixaban. Unweighted results showed that patients initiated on rivaroxaban had lower HRU and corresponding healthcare costs following treatment initiation, although rivaroxaban patients seemed to have less severe disease than apixaban patients, indicated by lower Quan-Charlson comorbidity index, CHA₂DS₂-VASc, and HAS-BLED scores. After balancing differences in baseline characteristics between the rivaroxaban and apixaban cohorts, the rivaroxaban users had significantly fewer outpatient hospital visits. Furthermore, rivaroxaban users incurred significantly lower all-cause total healthcare costs than apixaban users; the difference in costs was driven by significantly lower outpatient hospital visit costs.

A previous real-world study by Amin et al.²⁵ used the OptumInsight Research Database from 2013 to 2015 in order to compare the total all-cause healthcare costs associated with rivaroxaban and apixaban treatments in NVAF patients. The study reported similar costs between cohorts and a higher risk of all-cause hospitalization among patients with rivaroxaban use. However, given that the study evaluated outcomes over a short follow-up period soon after the approval of apixaban in December 2012, the results may not be representative of the now established direct oral anticoagulants in the market. The study also did not evaluate other measures of HRU, such as outpatient visits, which the current study identified as the main contributor of cost differences between the two cohorts. Furthermore, the current study focused on oral anticoagulant-naïve patients with NVAF who were newly initiated on rivaroxaban or apixaban, while the previous study included both oral anticoagulant-naïve and treatment-experienced patients. Moreover, differences in patients’ characteristics, such as comorbidity scores, may account for some discrepancy in results between the two studies, as the current study includes patients with descriptively higher Quan-Charlson comorbidity index scores (2.9 vs 2.3), but lower CHA₂DS₂-VASc (2.7 vs 3.8) and HAS-BLED (1.6 vs 2.8) scores compared to those of patients in the Amin et al.²⁵ study.

Other studies conducted to evaluate the difference in the economic burden associated with rivaroxaban and apixaban observed higher costs^26–28 and a higher risk of all-cause hospitalization for rivaroxaban compared with apixaban^26–28. However, several differences in study population and design could explain the differences between the results of our study and those of previous studies. First, similarly to the study conducted by Amin et al.²⁵, these studies evaluated healthcare costs over relatively short follow-up periods (mean follow-up: 4.5–9 months) compared to the current study. Second, some patient populations included in these studies may have limited generalizability (i.e. using US Department of Defense data) compared to our study²⁶. Finally, even though propensity score matching was used in previous studies, some of them only accounted for baseline hospitalizations without considering other HRU components (e.g. ER and outpatient visits), nor considering baseline healthcare costs, when balancing the patient cohorts^26,28. Not adjusting for these variables could have led to healthcare resource use and cost differences that were carried forward in the follow-up. Conversely, the weighting procedure used in the current study ensured that all study outcomes (i.e. HRU and their corresponding costs) were well-balanced between cohorts in the baseline before performing comparisons in the follow-up.

The rate of hospitalizations and their associated costs observed in this study were not significantly different between rivaroxaban and apixaban users, which may indicate that patients with NVAF initiated on rivaroxaban or apixaban incur similar costs associated with stroke and SE, the major contributors of hospitalization costs within this population^12,25. Indeed, previous studies showed that patients using these treatments had similar risk of hospitalization due to stroke and systemic embolism²⁵, and incurred similar medical costs associated with stroke and systemic embolism^25,26. This study suggests that, despite similar hospitalization costs, the rate of outpatient hospital visits and corresponding costs (which may not differ significantly shortly after treatment initiation^27,30) can in the long-term drive the healthcare costs as observed in patients initiating apixaban compared to rivaroxaban. Outpatient hospital visits, which are more costly than office visits and are typically observed among patients with more severe disease^31,32, may be driven by complications associated with AF that could be managed in an outpatient setting^33,34, such as transient ischemic attacks and clinically-relevant non-major bleeding^35,36. Together with the present results, this may indicate that patients using rivaroxaban achieve better disease control than those using apixaban, thus requiring fewer outpatient hospital visits to manage complications and subsequently incurring lower medical costs. Moreover, considering that AF affects millions of people in the US each year, the small difference in total healthcare costs associated with rivaroxaban use could translate into a significant reduction in economic burden of AF patients in the US.

A previous study assessing the cost of non-adherence to oral anticoagulants showed that total medical costs for patients who adhere to medication were significantly lower compared to those who were non-adherent³⁷, and another study suggested that higher adherence to once-daily DOAC resulted in significant cost savings compared to twice-daily regimens in the real-world^37,38. It is possible that the lower healthcare costs observed among rivaroxaban users relative to apixaban users may be attributed to improved adherence, though this was not explored in the current study. Further research is warranted to evaluate the reasons that may underlie the cost differences found in the present study.

Limitations

While this study has several design strengths, it is subject to certain limitations. First, since claims databases record diagnostic and procedure codes, clinical information, such as disease severity or reasons for change in treatment, is not available. Second, in spite of the large size of the present study population and its nationwide geographical representation, results may not be generalizable to patients with health insurance plans that substantially differ from those of the population analyzed in this study. Lastly, while IPTW is used to account for potential differences between the rivaroxaban and apixaban cohorts, the possibility of residual confounding due to unmeasured confounders cannot be excluded. As seen in the unweighted baseline characteristics, patients initiated on apixaban tended to have more severe disease states, as indicated by the higher baseline risk scores, rates of hospitalization, and healthcare costs. Although observed differences have been balanced, there may be unobserved confounders that influence the final results and could not be accounted for. Despite this, retrospective, observational studies that control for confounding factors through weighting techniques provide quality results generalizable to real-life scenarios.

Conclusions

In this large retrospective claims analysis of NVAF patients, those newly initiated on rivaroxaban incurred significantly lower total all-cause healthcare costs compared to patients initiated on apixaban up to 18 and 24 months post-treatment initiation. The majority of the total cost difference was driven by a significantly lower rate of outpatient hospital visits and their associated costs for patients treated with rivaroxaban. Considering that AF affects millions of people in the US each year, the total healthcare cost difference reported in the current study may translate into substantial savings at the national level.

Transparency

Declaration of funding

This study was funded by Janssen Scientific Affairs, LLC.

Declaration of interest

GG, FL, SDM, JT, and PL are employees of Groupe d’analyse, Ltée, a consulting company that provided paid consulting services to Janssen Scientific Affairs, LLC. for the conduct of the present study. DM and BKB are employees of Janssen Scientific Affairs and shareholders of Johnson & Johnson.

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

Author contributions

All authors were responsible for the study design and the interpretation of the study results. GG, FL, SDM, and JT were responsible for the data analysis. All authors reviewed the manuscript for intellectual content and approved the final draft.

Acknowledgements

Medical writing assistance was provided by Loraine Georgy, PhD, an employee of Groupe d’analyse, Ltée., a consulting company that has provided paid consulting services to Janssen Scientific Affairs, LLC, which funded the development and conduct of this study and manuscript.