A Real-World Observational Study of Hospitalization and Health Care Costs Among Nonvalvular Atrial Fibrillation Patients Prescribed Oral Anticoagulants in the U.S. Medicare Population

This article has been corrected. Please see J Manag Care Spec Pharm, 2020;26(5):682 BACKGROUND: Clinical trials have shown that direct oral anticoagulants (DOACs)—including dabigatran, rivaroxaban, apixaban, and edoxaban—are at least as effective and safe as warfarin for the risk of stroke/systemic embolism (SE) and major bleeding (MB) in patients with atrial fibrillation (AF). However, few studies have compared oral anticoagulants (OACs) among elderly patients. OBJECTIVE: To compare hospitalization risks (all-cause, stroke/SE-related, and MB-related) and associated health care costs among elderly nonvalvular AF (NVAF) patients in the Medicare population who initiated warfarin, dabigatran, rivaroxaban, or apixaban. METHODS: Patients (aged ≥ 65 years) initiating warfarin or DOACs (apixaban, rivaroxaban, and dabigatran) were selected from the Centers for Medicare & Medicaid Services database from January 1, 2013, to December 31, 2014. Patients initiating each OAC were matched 1:1 to apixaban patients using propensity score matching to balance demographic and clinical characteristics. Cox proportional hazards models were used to estimate the risk of hospitalization of each OAC versus apixaban. Generalized linear models and two-part models with bootstrapping were used to compare all-cause health care costs and stroke/SE- and MB-related medical costs between matched cohorts. RESULTS: Of the 264,479 eligible patients, 77,480 warfarin-apixaban, 41,580 dabigatran-apixaban, and 77,640 rivaroxaban-apixaban patients were matched. The OACs were associated with a significantly higher risk of all-cause hospitalization compared with apixaban (warfarin: HR = 1.27, 95% CI = 1.23-1.31, P < 0.001; dabigatran: HR = 1.13, 95% CI = 1.08-1.18, P < 0.001; and rivaroxaban: HR = 1.22, 95% CI = 1.18-1.26, P < 0.001) and were associated with a significantly higher risk of hospitalization due to stroke/SE (warfarin: HR = 2.18, 95% CI = 1.80-2.64, P < 0.001; dabigatran: HR = 1.45, 95% CI = 1.12-1.88, P = 0.006; and rivaroxaban: HR = 1.40, 95% CI = 1.14-1.71, P = 0.001). Also, the OACs were associated with significantly higher risk of hospitalization due to MB-related conditions compared with apixaban (warfarin: HR = 1.76, 95% CI = 1.59-1.95, P < 0.001; dabigatran: HR = 1.44, 95% CI = 1.23-1.68, P < 0.001; and rivaroxaban: HR = 1.89, 95% CI = 1.71-2.09, P < 0.001). Compared with apixaban, warfarin ($3,577 vs. $3,183, P < 0.001); dabigatran ($3,217 vs. $3,060, P < 0.001); and rivaroxaban ($3,878 vs. $3,180, P < 0.001) had significantly higher all-cause total health care costs per patient per month. Patients initiating the OACs had significantly higher MB-related medical costs compared with apixaban: warfarin ($472 vs. $269; P < 0.001); dabigatran ($364 vs. $245, P < 0.001); and rivaroxaban ($493 vs. $270, P < 0.001). Warfarin was also associated with higher stroke/SE-related medical costs compared with apixaban ($124 vs. $62, P < 0.001). CONCLUSIONS: This real-world study showed that among elderly NVAF patients in the Medicare population, apixaban was associated with significantly lower risks of all-cause, stroke/SE-related, and MB-related hospitalizations compared with warfarin, dabigatran, and rivaroxaban. Accordingly, apixaban showed significantly lower all-cause health care costs and MB-related medical costs.


R E S E A R C H
A trial fibrillation (AF) is the most common sustained heart arrhythmia and is estimated to affect approximately 9% of the population aged ≥ 65 years in the United States. 1,2 The presence of AF increases the relative risk of stroke by 5-fold, with attributable risk increasing from 4.6% among patients aged 50-59 years to over 20% among those aged 80-89 years. 3 AF's annual national incremental costs were estimated at $26 million compared with patients without AF, and hospitalizations were the primary cost driver. 4 For Medicare beneficiaries, AF onset leads to an adjusted mean incremental treatment cost of $14,199 per patient per year. 5

Patient Selection
OAC treatment-naive patients were included in the study if they had ≥ 1 prescription claim for apixaban, dabigatran, rivaroxaban, or warfarin during the identification period (January 1, 2013-December 31, 2014. Edoxaban was approved by the U.S. Food and Drug Administration in 2015; therefore, it was not included in our study. The first OAC pharmacy claim date was designated as the index date. Patients were required to be aged ≥ 65 years on the index date, have ≥ 1 AF medical claim (ICD-9-CM code 427. 31), and have continuous health plan enrollment with medical and pharmacy benefits for 12 months before the index date (baseline period). 20 Patients were excluded if they had evidence of rheumatic mitral valvular heart disease, mitral valve stenosis, heart valve replacement or surgery; transient AF (pericarditis, hyperthyroidism, and thyrotoxicity), venous thromboembolism, or an OAC pharmacy claim during the 12-month baseline period; pregnancy during the study period; or > 1 OAC prescription claim on the index date.
Patients were followed from the index date until the earliest of the OAC prescription discontinuation date, switch date from index drug to another OAC, date of death, date of health plan disenrollment, or December 31, 2014. Discontinuation was defined as no evidence of an index prescription for 30 days from the last day of the supply of the last filled prescription (discontinuation date). Switching was defined as having a prescription for an OAC other than the index drug within 30 days before or after the discontinuation date. 21

Outcomes
The primary outcomes were likelihood of all-cause hospitalization, hospitalization due to stroke/SE, hospitalization due to MB-related conditions, and health care costs, including allcause health care, all-cause medical, all-cause pharmacy, allcause hospitalization, all-cause emergency room (ER)/outpatient, stroke/SE-related medical, and MB-related medical costs.
Stroke/SE and MB hospitalization events were identified using hospital claims that had a stroke/SE or MB code as the primary discharge diagnosis. 22 The ICD-9-CM codes used for stroke and MB were based on a validated administrative claims-based algorithm as well as the clinical trial definition of stroke and MB. 7,23,24 Stroke/SE was stratified by ischemic stroke, hemorrhagic stroke, and SE; MB was stratified by gastrointestinal bleeding, intracranial hemorrhage, and other MB.
Stroke/SE-related medical costs were defined as hospitalization costs associated with the first stroke/SE event plus all subsequent stroke/SE costs occurring in the inpatient or outpatient setting (primary or secondary diagnosis) after the first stroke/SE during the follow-up. MB-related medical costs were defined as the hospitalization costs associated with the first MB event plus all subsequent MB costs occurring in the inpatient or outpatient setting (primary or secondary diagnosis) after the Warfarin, a vitamin K antagonist in use since the 1950s, has been proven to reduce ischemic and hemorrhagic stroke by 64% compared with placebo. 6 However, the narrow therapeutic window managed by the international normalized ratio and increased risk of bleeding have hindered the proper use of warfarin, especially in the elderly population. 2 Several new direct oral anticoagulants (DOACs) targeting key coagulation factors-including dabigatran, rivaroxaban, apixaban, and edoxaban-have been approved for stroke risk reduction in nonvalvular AF (NVAF) in recent years. Additionally, DOACs have demonstrated to be at least as effective as warfarin for the risk reduction of stroke and systemic embolism (SE) and are associated with similar or lower rates of major bleeding (MB). [7][8][9][10] While there are NVAF trials of DOACs versus warfarin, there are no head-to-head clinical trials comparing DOACs to each other. A few real-world studies have examined the risk of hospitalizations due to stroke/SE and MB among OACs. However, there is a dearth of real-world data for all-cause hospitalizations and health care costs. 11 Although warfarin has a lower pharmacy cost, using data from clinical trials and a Markov decision analysis model, apixaban, dabigatran, and rivaroxaban have shown to be more cost-effective than warfarin. 12 Real-world studies comparing health care costs among NVAF patients have also shown that apixaban patients had lower hospitalization costs compared with warfarin patients. 13,14 The objective of this study was to compare the risk of hospitalizations (all-cause, stroke/SE-related, and MB-related) and associated health care costs among elderly NVAF patients who initiated warfarin, dabigatran, rivaroxaban, or apixaban in the Medicare population.

■■ Methods Data Source
This real-world retrospective database analysis used data from the Centers for Medicare & Medicaid Services from January 1, 2012, to December 31, 2014. Medicare is the federal health insurance program for people aged ≥ 65 years, certain younger people with disabilities, and people with end-stage renal disease (permanent kidney failure requiring dialysis or a transplant). The database includes around 38 million fee-for-service beneficiaries. 15 It contains medical and pharmacy claims from 100% national Medicare data, which includes hospital inpatient, outpatient, Medicare carrier, Part D pharmacy, skilled nursing facility, home health agency, and durable medical equipment files. Medical claims were obtained through the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes, as well as Health Care Common Procedure Coding System and Current Procedural Terminology codes. Pharmacy claims were obtained through National Drug Code numbers. The comparative effectiveness research methods guidance documents aided researchers in designing the study. [16][17][18][19]  score, CHADS 2 score, CHA 2 DS 2 -VASc score, HAS-BLED score, comorbid conditions, and comedication use), as well as health care resource utilization, were assessed during the baseline period. The CHA 2 DS 2 -VASc stroke risk score was calculated using ICD-9-CM codes in the claims data as the summed total of the points determined for each diagnosis or characteristic and based on the CHADS 2 score (congestive heart failure,   25 The HAS-BLED bleeding risk score was based on evidence of hypertension, abnormal kidney or liver function, stroke, bleeding, aged > 65 years, and drugs/alcohol abuse or dependence. 26

Statistical Methods
All study variables were analyzed descriptively in each cohort, using apixaban as the reference. Means and standard deviations were reported for continuous variables, and student's t-tests were used to detect differences. Percentages were reported for categorical variables, and chi-square tests were used to detect differences in these variables. A P value of 0.05 was used as the threshold for statistical significance. Propensity score matching (PSM) was conducted to balance identified baseline demographics and clinical characteristics when comparing apixaban to dabigatran, rivaroxaban, or warfarin. Patients were matched 1:1 on the propensity scores generated by multivariable logistic regressions based on age, sex, geographic region, CCI score, CHA 2 DS 2 -VASc score, HAS-BLED score, prior bleed and stroke, comorbidities, baseline comedications, and baseline hospitalization. The covariates included in the PSM were determined based on clinical rationale. Nearest neighbor without replacement with a caliper of 0.01 was used to match the patients. 27 The balance of covariates was checked based on standardized differences with a threshold of 10%. 28 The incidence rates of hospitalization (all-cause, strokerelated, and MB-related) in the matched cohorts were calculated using the number of hospitalized patients divided by total person-years of exposure and multiplied by 100. Cox proportional hazards regression models were used to assess the likelihood of all-cause hospitalization, hospitalization due to stroke/SE, and hospitalization due to MB-related conditions in patients treated with other OACs relative to apixaban. 27 Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated for each outcome of interest.
Generalized linear models with log-link and a gamma distribution were used for the analysis of health care costs among the cohorts. 29 Additionally, two-part models with bootstrapping were used in the analysis of MB-and stroke-related medical costs, given the high proportion of cost fields with 0 values. The marginal effect of costs, 95% CIs, and P values for each matched cohort were reported.

Sensitivity Analyses
Three sensitivity analyses were conducted. First, for the DOAC cohorts, standard-dose (dabigatran 150 mg, rivaroxaban 20 mg, and apixaban 5 mg) and low-dose (dabigatran 75 mg, rivaroxaban 10 mg/15 mg, and apixaban 2.5 mg) cohorts were created based on the index dosage. Each patient initiating warfarin was assigned to one of the 2 subgroups according to the dose of the matched DOAC patient (standard and low dose). The balance of baseline characteristics was tested in each subgroup; when imbalance was detected

Baseline Characteristics
In the 3 postmatching cohorts, the mean age was around 78 years. The dabigatran-apixaban patients had the lowest mean CCI score (2.5), followed by rivaroxaban-apixaban (2.7) and warfarin-apixaban (2.8 and 2.7) patients. The CHA 2 DS 2 -VASc scores ranged from 4.3 to 4.6 across (standardized difference > 10%), the variable was included in the multivariate model. Risk of hospitalization (all-cause health care, stroke-related, and MB-related) was compared between the study cohorts, and the statistical significance of the interaction between treatments and subgroups was evaluated. Second, patients were censored at 6 months to create a more balanced length of follow-up between the treatment groups. Third, only patients with ≥ 30 days of follow-up were evaluated to exclude patients with too short of a follow-up to develop any stroke/SE or MB events. The second and third analyses were to help address the more recent approval of apixaban relative to dabigatran and rivaroxaban.

■■ Discussion
Using national Medicare data, we found that NVAF patients initiating warfarin, dabigatran, or rivaroxaban had a higher risk of all-cause, stroke/SE-related, and MB-related hospitalization compared with patients initiating apixaban. In addition, patients initiating warfarin, dabigatran, and rivaroxaban had significantly higher all-cause and MB-related health care costs compared with patients initiating apixaban. The ARISTOTLE trial demonstrated a significantly lower risk of stroke/SE (HR = 0.79, 95% CI = 0.66-0.95, P = 0.01) and MB (HR = 0.69, 95% CI = 0.60-0.80, P < 0.001) for apixaban patients compared with warfarin patients, which is consistent with our results. 7,30 In addition to clinical trials, a few observational studies comparing apixaban and warfarin have added real-world evidence in different patient populations. 22,[31][32][33][34] In a study of OptumLabs data by Yao et al. (2016), apixaban users had a 33% lower risk of stroke/SE and 55% lower risk of MB compared with warfarin. 31 In a study of 4 pooled datasets by Li et al. (2017), apixaban demonstrated lower risks of stroke/ SE (HR = 0.67, 95% CI = 0.59-0.76) and MB (HR = 0.60, 95% CI = 0.54-0.65) compared with warfarin. 32 Although no head-to-head DOAC clinical trials are available, several real-world studies have compared the risks of stroke/SE and MB among dabigatran, rivaroxaban, and apixaban. 33,35 In our analysis, apixaban had a lower risk of hospitalization due to stroke/SE and MB compared with the other the cohorts. About 20% of all matched patients had baseline bleeding, and more than 10% had baseline stroke/SE (Table 1).

Hospitalization: All-Cause, Stroke/SE, and MB
Incidence of all-cause hospitalizations and hospitalizations related to MB and stroke/SE are shown in Figure 2.

Health Care Costs
Patients prescribed warfarin, dabigatran, and rivaroxaban had significantly higher all-cause total health care costs PPPM compared with apixaban patients (Table 2). Inpatient and outpatient costs were the main drivers for health care costs.
Warfarin, dabigatran, and rivaroxaban patients had significantly higher MB-related medical costs compared with apixaban patients (Figure 3). Warfarin patients had significantly higher stroke/SE-related medical costs compared with apixaban patients.

Subgroup and Sensitivity Analyses Results
Results of the subgroup and sensitivity analyses were generally consistent with those of the main analysis (Appendix B,    35 However, we found in our study that dabigatran and rivaroxaban patients had a statistically significantly higher risk of both stroke/SE and MB than apixaban, which may be due to the larger sample size and hence increased power and different study populations.
The results of the sensitivity analyses showed consistent results with the primary analysis, which showed that standarddose or low-dose apixaban was associated with a lower risk of all-cause, stroke/SE-related, and MB-related hospitalization compared with other OACs.
There are a few economic studies that have compared apixaban to warfarin, dabigatran, and rivaroxaban among NVAF patients. In studies using IMS PharMetrics Plus, Humana, and Optum claims databases, warfarin patients had significantly higher total all-cause health care costs, stroke/SE-related costs, and MB-related medical costs compared with apixaban. 22 40 These studies are generally aligned with our findings on health care costs associated with apixaban relative to other oral anticoagulants.

Limitations
This study has several limitations. Given the nature of retrospective observational studies, only associations were assessed, and no causality can be concluded. This database contains information from the Medicare population and may not be generalizable to the entire U.S. population of NVAF patients. Additionally, administrative claims data are primarily collected for billing purposes rather than research, and the analysis is constrained by codes that may contain coding errors and missing data. In addition, the cause of stroke/SE and major bleeding is not available in the claims data. Moreover, unobserved confounders such as compliance, AF duration, and over-the-counter aspirin use may exist for which the analysis did not control. Nevertheless, we used PSM to balance observed demographics and clinical characteristics. The follow-up time was short, not uniform, and was not consistent with the clinical trials. Therefore, the sensitivity analysis with patients censored at 6 months was conducted to address the issue of imbalanced follow-up times. Sensitivity analysis results for MB and stroke/SE were consistent with those in the main analysis. Finally, the interpretation of stroke/SE-related outcomes should be carefully considered because of the low number of stroke/SE events.

■■ Conclusions
This real-world observational study is one of the largest that has compared the risks of stroke/SE and MB and the associated health care costs between OACs in elderly NVAF patients. In this study, apixaban was associated with significantly lower risks of all-cause, stroke/SE-related, and MB-related hospitalizations compared with warfarin, dabigatran, and rivaroxaban. Accordingly, apixaban showed significantly lower all-cause health care costs as well as MB-related medical costs. This study may assist clinicians in determining the appropriate OAC for OAC-naive elderly NVAF patients and could be informative to decision makers managing Medicare populations.