The present study used the “National Health Insurance Research Database (NHIRD)” released by the Taiwan National Health Research Institutes. The National Health Insurance (NHI) system is a mandatory universal health insurance program that provides comprehensive medical care coverage to all Taiwanese residents. The NHIRD consists of detailed healthcare data from $>$23 million enrollees, representing $>$99% of Taiwan’s population. The cohort dataset has encrypted the patients’ original identification numbers to protect their privacy, and the encrypting procedure was consistent so that a linkage of the claims belonging to the same patient was feasible within the NHI database and could be continuously followed.

The study protocol is similar to our previous studies which have been published [19, 20]. Patients aged $\ge$20 years with AF identified from the NHIRD from 2012 to 2018 constituted the main study population. Fig. 1 shows the flowchart of the patient enrollment and study design.

Fig. 1.

Patient enrollment flowchart. A total of 203,775 patients aged $\ge$20 years with AF were identified from the nationwide cohort and 67,426 of them were taking either warfarin or NOACs. Further comparisons between males and females in terms of types of oral anticoagulants were performed. AF, atrial fibrillation; NHIRD, National Health Insurance Research Database; NOACs, non-vitamin K antagonist oral anticoagulants.

The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes were used to confirm the diagnosis. We defined patients with a certain disease only when it was a discharge diagnosis or confirmed more than twice in the outpatient department to ensure the accuracy of diagnosis [19, 20, 21]. The CHA${}_2$DS${}_2$-VASc score was calculated for each patient by assigning 1 point each for ages 65–74 years and a history of hypertension (HTN), diabetes mellitus (DM), heart failure (HF), vascular disease (myocardial infarction or peripheral artery disease), and female gender and 2 points each for a history of a stroke, transient ischemic attack (TIA), and age of $\ge$75 years [22]. The HAS-BLED score was calculated by assigning 1 point each for hypertension, abnormal renal, or liver function, stroke, bleeding history, age $\ge$65 years, and antiplatelet drug or alcohol use [23]. The information on the international normalized ratio (INR) of warfarin was unavailable in the Taiwan registry database, so the scoring in the present study excluded the component of “labile INR”, consistent with prior registry studies. Additionally, abnormal renal or liver function was defined by the ICD-9-CM codes rather than laboratory data. The diagnostic accuracy using this definition in NHIRD has been validated previously [24, 25, 26].

The clinical endpoints included the occurrence of death, ischemic stroke, intracerebral hemorrhage (ICH), major bleeding, and composite adverse events (death or ischemic stroke or ICH or major bleeding). Ischemic stroke and ICH were diagnosed with concomitant brain imaging studies, including computed tomography or magnetic resonance imaging. Major bleeding was ICH or bleeding originating from the gastrointestinal, genitourinary, or respiratory tract that requires hospitalization. Each endpoint was independently analyzed of the others without being censored. The accuracy of diagnosis of ischemic stroke in Taiwan’s NHIRD was approximately 94% [27]. Another validation study demonstrated that the diagnostic accuracy of ischemic stroke in NHIRD was high, with a positive predictive value and sensitivity of 88.4% and 97.3%, respectively [28].

Data were presented as the mean value and standard deviation for normally distributed continuous variables and proportions for categorical variables. The unpaired two-tailed t-test was used to assess differences between continuous values. Nominal variables were compared by chi-square test or Fisher’s exact test. Incidence rates of events were calculated by dividing the number of events by person-year at risk. The Kaplan-Meier method was used to plot the cumulative incidence of clinical events with statistical significance examined by the log-rank test. Multivariate Cox proportional hazards models were used for risk prediction adjusting for significant baseline variables. All statistical significances were set at a p-value of 0.05 using the Statistical Package for the Social Sciences version 17.0 statistical software (SPSS, Chicago, IL, USA).

Our study population consisted of 203,775 patients aged $\ge$20 years with AF from 2012 to 2018 and 112,836 (55.4%) of them were males. Compared to males, females were older, had more comorbidities of HF, HTN, DM, previous stroke/TIA, hyperlipidemia, autoimmune diseases, anemia, history of bleeding and less vascular disease, chronic obstructive pulmonary disease, cancer, abnormal liver function, and alcohol excess/abuse (Table 1). Therefore, females demonstrated higher CHA${}_2$DS${}_2$-VASc and HAS-BLED scores. The CHA${}_2$DS${}_2$-VASc scores remained higher in females after excluding the one point contributed by the female gender. Males were more likely to use anti-platelet drugs while slightly but significantly more females were taking NOACs, but warfarin use demonstrated no difference between sexes. Males were apt to use beta-blockers for rate control while females tended to take calcium channel blockers and digoxin. A higher percentage of dronedarone and propafenone use was found in females regarding rhythm control agents (Table 1).

Among all patients with AF, 36,820 males and 30,606 females taking OACs were further analyzed (Table 2). The distribution of baseline characteristics between warfarin and NOAC users was very similar between the sexes. Both male and female patients taking NOAC were older and had more underlying comorbidities except HF, abnormal renal function, anemia, and a history of bleeding compared to those using warfarin. NOAC users demonstrated higher CHA${}_2$DS${}_2$-VASc and HAS-BLED scores. The CHA${}_2$DS${}_2$-VASc scores remained higher in NOAC users than in warfarin users in both sexes even after excluding the point contributed by gender. Anti-platelet drug use was more common in warfarin users in both sexes, but clopidogrel use was more common in male NOAC users and female warfarin users. Female NOAC users were more likely to take beta-blockers than female warfarin users, whereas the percentage of beta-blockers used was similar between NOAC and warfarin groups in males.

The mean follow-up of 2.89 years reported 12,850 deaths, 3033 ischemic strokes, 874 ICHs, 4125 major bleeding, and 16,750 composite adverse events. Compared to warfarin use, the NOAC group had lower incidence rates of death (6.99% versus 7.32%), ischemic stroke (1.47% versus 2.07%), ICH (0.40% versus 0.58%), major bleeding (2.17% versus 2.65%), and composite adverse events (9.80% versus 10.65%). The Kaplan-Meier analysis demonstrated higher rates of clinical events in the NOAC group compared to the warfarin group for both sexes, with female patients exhibiting a more prominent decrease in cumulative incidence of ischemic stroke with NOAC use compared to warfarin (Fig. 2).

Fig. 2.

Cumulative incidence curves of ischemic stroke (A), ICH (B), and major bleeding (C) in male and female patients in relation to OAC use. The c Kaplan-Meier analysis revealed higher cumulative incidence rates in the NOAC group compared to the warfarin group for both sexes. Additionally, the reduction in the cumulative incidence of ischemic stroke with NOAC use, as opposed to warfarin, was more pronounced in female patients than in male patients. ICH, intracranial hemorrhage; NOACs, non-vitamin K antagonist oral anticoagulants; OAC, oral anticoagulants; CI, confidence interval; HR, hazard ratio.

Multivariate Cox regression analysis revealed that NOAC was associated with lower risk of death (adjusted hazard rate [aHR]: 0.726, 95% confidence interval [CI]: 0.700–0.752, p 0.001), ischemic stroke (aHR: 0.478, 95% CI: 0.444–0.515, p 0.001), ICH (aHR: 0.536, 95% CI: 0.466–0.617, p 0.001), major bleeding (aHR: 0.578, 95% CI: 0.542–0.615, p 0.001), and composite adverse events (aHR: 0.658, 95% CI: 0.628–0.679, p 0.001) (Fig. 3). Subgroup analyses between males and females were performed in terms of OAC types. NOACs demonstrated a consistent association with lower risks of death, ischemic stroke, ICH, major bleeding, and composite adverse events than warfarin in both sexes. Moreover, risk reduction of ischemic stroke with NOAC compared to warfarin was significantly greater in females than in males (interaction p = 0.040) (Fig. 3).

Fig. 3.

Incidence and risk of clinical endpoints between warfarin and NOAC use in both males and females. The whole study cohort demonstrated that NOAC was associated with lower risks of clinical endpoints compared to warfarin. Subgroup analysis revealed the consistently better outcomes associated with NOAC compared to warfarin in both sexes, whereas a greater risk reduction of ischemic stroke was observed in female patients with AF. AF, atrial fibrillation; CI, confidence interval; HR, hazard ratio; ICH, intracranial hemorrhage; NOACs, non-vitamin K antagonist oral anticoagulants.

This nationwide cohort analyzed the characteristics and long-term prognosis of male and female patients with AF in terms of OAC types and presented the following main results: (1) gender differences in baseline characteristics and medication use in patients with AF, where female patients with AF demonstrated higher CHA${}_2$DS${}_2$-VASc and HAS-BLED scores; (2) the distribution of underlying demographic characteristics between warfarin and NOAC groups was much similar between male and female patients with AF; (3) the observation that NOAC was associated with better outcomes compared to warfarin was consistent in both sexes, and female patients with AF demonstrated a greater risk reduction of ischemic stroke.

Data in terms of OAC use in both sexes differ in previous reports. The CARMEN-AF registry [29] and the Global Anticoagulant Registry in the Field (GARFIELD-AF) [30] revealed no gender differences in OAC use. Conversely, the United States PINNACLE National Cardiovascular Data Registry from 2008 to 2014 reported that women with AF were more likely to receive aspirin but not OACs [31]. The present study revealed slight but significant differences in OAC prescription because NOAC use was more common in females and the percentage of warfarin use was similar between sexes. We hypothesized that a constellation of multiple factors, such as different periods, geographic factors, and underlying demographics caused gender differences in OAC use. For example, vascular diseases were more common in males and thus more males received aspirin or clopidogrel than women did. The need for multiple blood thinners might be a crucial factor for doctors while selecting medications.

Generally, female patients with AF have a higher risk of stroke and systemic thromboembolism, and AF-related embolic stroke in women is more severe and disabling [32, 33, 34]. Warfarin was the mainstream of stroke prevention in patients with AF before the introduction of NOAC, but the Medicare administrative claims data revealed that warfarin reduced stroke less well in females. Further, female patients with AF had a slightly higher risk of hospitalizations despite warfarin use [35]. One possible explanation underlying this observation is the higher chance of poor INR control in females [36]. Until now, no RCTs have compared gender differences with OAC use. The DIRECT registry, a single-center prospective observational registry of 806 patients with AF treated with NOACs, demonstrated comparable bleeding events between men and women whereas the thromboembolic event rate was higher in women [37]. One meta-analysis, including major RCTs of NOACs versus warfarin in patients with AF (RE-LY, ROCKET-AF, ARISTOTLE, and AVERROES), revealed a higher risk of systemic thromboembolism in females compared to males when treated with warfarin, which did not occur with NOAC treatment [38]. One review article revealed that the sex disparity in stroke is no longer seen after introducing NOAC [38, 39]. Furthermore, one meta-analysis reported differential benefits of NOACs between sexes in which male patients were more protected from stroke/systemic thromboembolism and female patients from major bleeding events [40]. Our present study was partly congruent with previous studies that both sexes benefited from NOAC despite different background characteristics, and the unfavorable prognosis in females no longer existed with NOAC. However, we revealed a greater risk reduction of ischemic stroke with NOAC use in female patients despite higher CHA${}_2$DS${}_2$-VASc scores. Potential reasons underlying the different beneficial effects of NOAC between sexes in previous studies are unknown, probably due to the different study designs and cohorts, age distributions, and background characteristics. In particular, the meta-analysis included some trials that were not powerful enough to evaluate sex-specific differences. An RCT specifically designed for the evaluation of gender differences in terms of NOAC use is required for a robust conclusion.

There are some limitations in the present study. First, males and females might possess different biochemistry data and demographic information, which were lacking in the database, but this was a common limitation in the registry database. Second, the diagnosis and occurrence of events were based on the diagnostic codes registered by the physicians responsible for patient treatments, and under-diagnosis could be excluded. However, the accuracy of diagnosis in Taiwan’s NHIRD has been previously validated [24, 25, 27, 28]. Third, INR levels and time in the therapeutic range of warfarin use were not available in the database. Fourth, because this is a retrospective observational study, the reasons underlying more risk reduction of ischemic stroke with NOAC in female patients is unknown. We postulated the benefit of NOAC over warfarin might be more prominent in females because female patients with AF were more likely to have poor INR control than male patients in previous study [25]. However, this is solely an assumption because INR data is not available in the present study. Finally, the doses and types of NOACs were not analyzed in our study, thus whether or not these factors would interfere with the results remains unknown.