Sex-Specific Comparative Effectiveness of Oral Anticoagulants in Elderly Patients With Newly Diagnosed Atrial Fibrillation

Introduction

Nonvalvular atrial fibrillation (AF) affects 3 million adults in the United States.^1,2 By 2050, ≈8 million US adults will have AF.¹ AF confers a 3- to 5-fold increase in the risk of stroke.^3,4 For decades, warfarin was the only oral anticoagulant available for stroke prophylaxis in patients with AF. In recent years, direct oral anticoagulants (DOAC) dabigatran and rivaroxaban have been approved for stroke prophylaxis in this population.^5,6 Randomized controlled trial data support the efficacy of these DOACs,^7–11 and they are commonly used in clinical practice.

Women with AF have significantly higher stroke risk compared with their male counterparts, irrespective of their age and comorbid disease profile.^12,13 As a result, female sex has been incorporated into the most widely accepted risk scoring algorithms to identify AF patients who will benefit from anticoagulation.^14,15 However, stroke risk remains elevated in women compared with that in men even after initiating and sustaining similar quality warfarin therapy.¹⁶ Hence, there is a need to understand the effectiveness of DOACs in women with AF.

One shortcoming of directly extrapolating randomized controlled trial data on DOACs into clinical management of women with AF is that women constituted much smaller numbers in both the dabigatran and rivaroxaban trials^7–9 compared with contemporary clinical practice. Also, variability in treatment adherence and patient follow-up are some of the challenges in the clinical management of patients that are not reflected in randomized trials.^17,18 Finally, previous clinical trials for DOACs do not allow for head-to-head comparisons of rivaroxaban to dabigatran, and sex-specific effectiveness of rivaroxaban to dabigatran is not known.

In order to bridge this literature gap, we used a nationally representative cohort of elderly Medicare beneficiaries with newly diagnosed AF to assess the sex-specific comparative effectiveness of rivaroxaban and dabigatran to each other and to warfarin.

Methods

Using the Centers for Medicare and Medicaid Services patient records, we linked data sources including (1) Beneficiary Summary File Base and Chronic Conditions segments; (2) Inpatient (part A) and Carrier (part B) Standard Analytic Files for 2011 through 2013; and (3) Pharmacy Drug Event (part D) files for 2011 to 2013. The institutional review board of University of Iowa approved the study. This being a retrospective cohort analysis of claims data, informed consent was not required.

We identified 213 705 Medicare beneficiaries who were enrolled in Centers for Medicare and Medicaid Services part D prescription drug coverage plan, were newly diagnosed with AF between November 1, 2011, and October 31, 2013, and initiated dabigatran 150 mg BID, rivaroxaban 20 mg QD, or warfarin within 90 days after AF diagnosis. Patient selection and study cohort formation algorithms are detailed in Figure I in the Data Supplement. New AF was defined based on previously published algorithms (ie, one inpatient claim or 2 outpatient claims within 90 days with International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 427.31 as primary or first secondary diagnosis).^19,20

Outcomes

The primary outcomes were inpatient admissions for acute ischemic stroke or major bleeding, as defined by Rothendler et al²¹ and Suh et al²² based on the primary ICD-9-CM diagnosis on inpatient standard analytical files claims for acute care stays. The secondary outcomes were subdivisions of major bleeding, defined as intracranial hemorrhage (ICH; including hemorrhagic stroke), gastrointestinal hemorrhage (GIH), and other major non-GIH, based on previously published algorithms.²³

Patient Characteristics

Patient characteristics were derived from Medicare enrollment data and inpatient and carrier claims. Age, sex, and race were identified from Medicare enrollment data. Comorbid diseases^19,20 defined by Elixhauser et al²⁴ were identified by ICD-9-CM diagnoses in inpatient and outpatient claims during the 12 months preceding AF diagnosis. Previous cerebrovascular events and previous bleeding episodes were also identified using previously published algorithms.^21,22 We also identified additional comorbidities of importance to AF outcomes, including other dysrhythmias (ICD-9-CM codes 427.X, excluding 427.3), cardiomyopathy (ICD-9 codes 425.X), cardiac conduction disorder (eg, bundle branch block; ICD-9 codes 426.X), and previous implantable cardiac device (eg, pacemaker; ICD-9 codes V45.0 and V53.3). Stroke risk was assessed using the standard CHA₂DS₂-VASc (1 point each for congestive heart failure diagnosis, female sex, hypertension diagnosis, diabetes diagnosis, age 65-75 years, and vascular disease diagnosis; 2 points each for age >75 years and prior stroke or transient ischemic attack) scoring system.²⁵ The HAS-BLED (1 point each for hypertension diagnosis, renal disease, liver disease, stroke history, prior major bleeding, labile INR, age >65 years, medication usage predisposing to bleeding and alcohol or drug use history) algorithm was used to assess bleeding risk.²⁶ Finally, the comorbidity score defined by Gagne et al²⁷ was calculated to assess disease burden. This score is of proven value to improve death prediction in hospitalized patients.²⁷

Statistical Analysis

Separate male and female cohorts were constructed. Comparisons were made between 3 treatment groups: participants initiated on dabigatran 150 mg BID (dabigatran group), participants initiated on rivaroxaban 20 mg QD (rivaroxaban group), and participants who were initiated on warfarin (warfarin group). Demographic variables, comorbid diseases, medication use, CHA₂DS₂-VASc score, HAS-BLED score, and Gagne score were compared between the 3 treatment groups separately in men and women, using a χ² test or 1-way ANOVA as applicable. We then used the 3-way propensity-matching method described by Rassen et al²⁸ to create groups of patients receiving dabigatran, rivaroxaban, or warfarin that were balanced with respect to patient covariates and also had clinical equipoise—that is, patients included in the matched samples were plausible candidates for all 3 anticoagulants under study. Propensity matching was conducted separately for men and women. Success of the matching algorithm was evaluated by comparing standardized differences in demographic variables, comorbid diseases, medication use, CHA₂DS₂-VASc score, HAS-BLED score, and the Gagne score between each drug in the matched samples. In accordance with Austin,²⁹ we evaluated the success of the matching algorithm using standardized differences rather than P values because P values depend on sample sizes and may, therefore, not adequately reflect meaningful differences. Standardized differences <10% (ie, 0.10 times the SD of the difference) suggest adequate balance.²⁹ We then used the propensity-matched samples to calculate event rates/patient year of follow-up for each outcome for the 3 anticoagulant groups in men and women separately. In addition, Kaplan–Meier curves for each anticoagulant were plotted for each study outcome in men and women. Log-rank test was performed to compare the curves for the 3 anticoagulants. Finally, we used multivariable Cox proportional hazards regression on the matched samples to further control for possible differences between treatment groups within individual sex groups. In these models, the dependent variables were time (in days) from anticoagulant initiation to a given event (eg, admission for stroke or censoring), whereas candidate-independent variables included patient demographics, comorbid conditions, concurrent medication use, and previous health services utilization as described previously. Censoring events included end of observation (December 31, 2013), cessation of the initial anticoagulant (defined as the date of the last fill plus days supplied), or death. Variables were selected for inclusion in Cox models based on relationship to the outcome, using a statistical criterion on 0.05. Covariates adjusted in the Cox models for each of the outcomes are detailed in the Data Supplement. Models also included indicators for the type of anticoagulant used. (dabigatran versus warfarin [reference], rivaroxaban versus warfarin [reference], and rivaroxaban versus dabigatran [reference]). Because propensity matching created dependencies in the data, we used robust SEs for the Cox regression models. Results of the regression analyses were reported as hazard ratios (HR) and 95% confidence intervals (CI) for dabigatran versus warfarin, rivaroxaban versus warfarin, and rivaroxaban versus dabigatran. Data set creation and propensity matching were conducted using SAS; all other analyses were performed using STATA 11 software.

Results

The final study cohort included 21 979 patients in the dabigatran group, 23 177 in the rivaroxaban group, and 101 715 in the warfarin group. There were 65 734 men (44.7%), and among them, 10 740 initiated dabigatran, 11 606 initiated rivaroxaban, and 43 388 initiated warfarin. There were 81 137 women (55.3%), of which 11 239 initiated dabigatran, 11 571 initiated rivaroxaban, and 58 327 initiated warfarin.

There were significant differences in baseline characteristics across the 3 anticoagulant groups in men and women (Table I in the Data Supplement) before propensity matching. After propensity matching (Table 1; Tables II and III in the Data Supplement), there were 22 854 total men in the matched sample (7618 taking each drug) and 33 093 women (11 031 taking each drug). After propensity matching in men, there were no statistically significant differences in demographic characteristics, comorbid conditions, medications in previous 90 days, and healthcare utilization between the 3 anticoagulant groups (Table 1). Moreover, all standardized differences between drug groups for men were substantially <10%. In women, after propensity matching, statistically significant differences remained for some comorbid conditions (eg, heart failure, diabetes mellitus, CHA2DS2-VASc score). However, all standardized differences between the 3 anticoagulant groups were substantially <10%, suggesting good covariate balance.

Outcomes

Sex-specific rates of each outcome expressed as number of events and as rates/patient year of follow-up are provided in Table 2 for the propensity-matched cohorts. As expected, stroke rates were higher among women than men. There were 185 strokes experienced by men and 356 experienced by women. There were 533 major bleeding events (ICH, GIH, and non-GIH) experienced by men and 897 experienced by women, of which >80% were GIH.

Table 3 shows the hazard of each outcome in patients taking dabigatran (relative to warfarin), rivaroxaban (relative to warfarin), and rivaroxaban (relative to dabigatran), separately for men and women, based on multivariable Cox regression on propensity-matched samples. Figures 1A, 1B, 2A, and 2B show the associated survival curves (with embedded graphs showing log-transformed survival rates to provide visual separation between curves).

• Download figure
• Download PowerPoint

Among men, rivaroxaban use was associated with significantly reduced risk of stroke compared with dabigatran use (HR, 0.66; 95% CI, 0.45–0.96; P=0.029) and warfarin use (HR, 0.69; 95% CI, 0.48–0.99; P=0.048; Figure 1A). There were no significant differences between treatment groups in relative risk of stroke in women (Figure 1B).

In men, there were no significant differences between the 2 DOAC groups in any of the bleeding outcomes (Table 3; Figure 2A). There were also no significant differences between treatment groups in GIH. Men who initiated dabigatran had lower risk of any major bleeding (HR, 0.73; 95% CI, 0.59–0.90) and lower risk of ICH (HR, 0.29; 95% CI, 0.12–0.75) compared with men who initiated warfarin. Both DOAC groups had a lower risk of other non-GIH compared with warfarin users (HR, 0.53; 95% CI, 0.33–0.84 and HR, 0.47; 95% CI, 0.28–0.78) for men taking rivaroxaban versus warfarin and dabigatran versus warfarin, respectively.

• Download figure
• Download PowerPoint

Several differences across drugs with respect to bleeding outcomes were noted in women, with women rivaroxaban users having the highest relative bleeding risks (Figure 2B). Rivaroxaban use was associated with an increased risk of any major bleeding compared with both warfarin (HR, 1.20; 95% CI, 1.03–1.42) and dabigatran (HR, 1.27; 95% CI, 1.09–1.48). A significant increase in GIH risk was also observed for rivaroxaban compared with warfarin (HR, 1.43; 95% CI, 1.20–1.71) and dabigatran (HR, 1.19; 95% CI, 1.01–1.44). There was no significant difference between dabigatran and warfarin users in risk of any major bleeding or GIH. For ICH, a significant risk reduction compared with warfarin was observed for women taking dabigatran (HR, 0.40; 95% CI, 0.22–0.71) but not for women taking rivaroxaban. Finally, women taking either rivaroxaban (HR, 0.56; 95% CI, 0.38–0.81) or dabigatran (HR, 0.32; 95% CI, 0.20–0.50) had a significantly lower risk of other non-GIH compared with warfarin users, whereas the risk of other non-GIH was greater for women taking rivaroxaban than dabigatran (HR, 1.78; 95% CI, 1.13–2.87).

Discussion

In this nationally representative analysis of Medicare claims data from the United States, we report sex-specific comparative effectiveness of oral anticoagulants in patients with newly diagnosed AF. In men, rivaroxaban use decreased stroke risk when compared with dabigatran use and warfarin use and was associated with similar risk of major bleeding. In women, although stroke risk was similar in the 3 anticoagulant groups, risk of major bleeding was higher with rivaroxaban use.

In a subgroup analysis of the ROCKET-AF⁹ trial that compared sex-specific effectiveness of rivaroxaban versus warfarin, the risk of stroke and major bleeding were similar with rivaroxaban compared with warfarin in both men and women.⁹ In contrast, we observed rivaroxaban to be more effective than warfarin for stroke prevention in men and to be similarly effective as warfarin in women. Also, risk of major bleeding was higher in women (but not in men) with rivaroxaban use. Although the discrepancies between our findings and those of ROCKET AF are hard to explain, the baseline CHADS2 score in ROCKET-AF was 3.48±0.94, whereas the baseline CHADS2 scores in our men and women were 2.3±0.97 and 2.4±1.1, respectively. The lower baseline stroke risk in our study participants may explain the superiority of rivaroxaban over warfarin for stroke prevention in men, noted in our study that was not reported in ROCKET-AF. Also, ≈20% of ROCKET AF participants used 15 mg of rivaroxaban that could have decreased efficacy of rivaroxaban in ROCKET AF study, whereas all our study participants used 20 mg rivaroxaban. This could explain the better stroke prevention observed in men and increased risk of major bleeding noted in women in our study. In the survival curves comparing stroke-related hospitalizations in men (Figure 1A), most strokes seem to have happened in the first 200-day follow-up among rivaroxaban users. After 200 days, stroke rates seem to have decreased for the remaining duration of follow-up. Although hard to explain based on our data and that this is possibly an observation because of chance, it may be worthwhile for future studies to assess if there is an association between the duration of rivaroxaban use and the effectiveness of stroke protection it confers. In contrary, this observation was not seen in women (Figure 1B). Previous studies using observational data report conflicting results^30–33 with some supporting a superiority of rivaroxaban to warfarin for stroke prevention and bleeding risk,^32,33 and some reporting similar efficacy of the 2 anticoagulants.³¹ However, no previous study using observational data has assessed sex-specific effectiveness of these 2 anticoagulants.

In a sex-specific subgroup analysis of RELY,⁷ dabigatran 150 mg BID was superior to warfarin for stroke prevention in both men and women, whereas sex-specific bleeding outcomes were not reported. The baseline stroke risk of RELY trail participants (CHADS2 score of 2.1±1.1) were similar to our men (2.3±0.97) and women (2.4±1.1) using dabigatran. In spite we noted similar effectiveness of dabigatran to warfarin with stroke protection in men and women whereas RELY reported superiority of dabigatran to warfarin for stroke prevention in men and women. Observational data have both supported^23,34,35 and contradicted^36–38 the primary analysis of the RELY trial. However, sex-specific outcomes were not reported in these observational studies. A Canadian study,³⁹ using a propensity-matched analysis involving 31 786 women and 31 324 men with AF from administrative data, compared sex-specific effectiveness of dabigatran (110 and 150 mg) to warfarin. The study concluded that dabigatran use was associated with similar stroke risk compared with warfarin in both sexes but was protective against major bleeding only in men. The results of this Canadian observational study are in concordance with our findings, in spite of the fact that all our study participants used dabigatran 150 mg BID. In our study, dabigatran and warfarin were similarly effective for stroke prevention in both sexes, while dabigatran decreased risk of major bleeding in men but not in women.

Our study suggests the possibility of a higher bleeding risk in women with AF treated with DOACs; an observation noted in other clinical settings as well. The meta-analysis by Alotaibi et al⁴⁰ reported a 21% higher relative risk of bleeding in women treated with DOACs for venous thromboembolism compared with that in men. Women, by virtue of their lean body weight, especially our elderly Medicare population, have decreased creatinine clearance compared with men and hence may attain higher serum levels of DOACs predisposing them to bleed more. Furthermore, differences in sex hormones between sexes may influence variability in hemostasis and vascular reactivity,⁴¹ although it should be noted that in our study, all female subjects were post-menopausal. It is also possible that sex gaps in access to care may contribute to bleeding differences between men and women. Evidence suggesting suboptimal access to care in women with AF exists. Bhave et al,⁴² using Medicare data from 2010 to 2011, reported that women are less likely than men to be prescribed an oral anticoagulant and are often denied other evidence-based AF therapies including catheter-based ablations.

With regard to the site of bleeding, although women in our study bled more with DOACs and men did not, the bleeding was predominantly gastrointestinal and not ICH for both comparisons (rivaroxaban versus warfarin and dabigatran versus warfarin). This is in concordance with the literature behind all DOACS, and they increase risk of GIH and not ICH when compared with warfarin.⁴³

Direct, randomized head-to-head comparisons between rivaroxaban and dabigatran have not been conducted. Indirect comparisons in the form of network meta-analyses are available.^44,45 One suggests superiority of dabigatran over rivaroxaban,⁴⁴ whereas the other suggests similar stroke prevention for the 2 DOACs.⁴⁵ Both support a similar bleeding risk for rivaroxaban and dabigatran. In a direct comparison by Graham et al,⁴⁶ involving 118 891 patients with AF from an administrative claims database, rivaroxaban (20 mg QD) performed similar to dabigatran (150 mg BID) for stroke prevention, although rivaroxaban use was associated with a significantly higher risk of major bleeding compared with dabigatran use. However, sex-specific comparisons are lacking in these reports.

Limitations

Although the strengths of our study include the large sample size of patients, the use of propensity matching to address possible confounding, and inclusion of patients with only new-onset AF who initiated standard-dose DOACs thereby minimizing variability in exposure definition, there are several limitations to note. First, there is always the possibility of residual confounding in analysis of observational data. Although we did achieve successful balance in patient characteristics across the 3 drugs compared, it is still possible that unmeasured confounders could have biased our results. Second, our study included only patients ≥66 years of age; the results, therefore, may not be extrapolated to younger patients (although we note that the age range in the Medicare data is consistent with patients in the RE-LY and ROCKET-AF trials). Third, inclusion of Medicare part D (prescription benefit plan) enrollees only could impact generalizability if such patients are systematically different than Medicare beneficiaries who do not enroll in a prescription benefit plan. Moreover, beneficiaries with prescription drug coverage may also have greater ability to adhere to prescription medications compared with patients without prescription coverage. Furthermore, considering that our study sample was an administrative data set, we lacked granular details such as AF burden, international normalized ratio, and time in therapeutic range in warfarin users. Finally, our patients had a relatively short duration of follow-up (median of 14 months); hence, risk assessments may be considered to be short term.

Conclusions

Sex differences are possible in the effectiveness of DOACs. Women tended to bleed more with DOACs compared with warfarin, although the risk of bleeding in men was similar for DOACs and warfarin. Rivaroxaban may be more effective for stroke prevention compared with dabigatran and warfarin in men, but all 3 drugs seem to provide similar stroke prevention in women. Considering the observational nature of our analysis, further validation is needed to replicate these findings and to understand the mechanism behind sex-specific effects. Our study results may help clinicians tailor their choice of anticoagulants in men and women.

Acknowledgments

All authors had access to the data and participated in the design and writing of the article.

Footnotes