Comparative effectiveness and safety of non‑vitamin K antagonists for atrial fibrillation in clinical practice: GLORIA‑AF Registry

Background and purpose Prospectively collected data comparing the safety and effectiveness of individual non-vitamin K antagonists (NOACs) are lacking. Our objective was to directly compare the effectiveness and safety of NOACs in patients with newly diagnosed atrial fibrillation (AF). Methods In GLORIA-AF, a large, prospective, global registry program, consecutive patients with newly diagnosed AF were followed for 3 years. The comparative analyses for (1) dabigatran vs rivaroxaban or apixaban and (2) rivaroxaban vs apixaban were performed on propensity score (PS)-matched patient sets. Proportional hazards regression was used to estimate hazard ratios (HRs) for outcomes of interest. Results The GLORIA-AF Phase III registry enrolled 21,300 patients between January 2014 and December 2016. Of these, 3839 were prescribed dabigatran, 4015 rivaroxaban and 4505 apixaban, with median ages of 71.0, 71.0, and 73.0 years, respectively. In the PS-matched set, the adjusted HRs and 95% confidence intervals (CIs) for dabigatran vs rivaroxaban were, for stroke: 1.27 (0.79–2.03), major bleeding 0.59 (0.40–0.88), (0.79–1.29). For the comparison of rivaroxaban vs apixaban, in the PS-matched set, the adjusted HRs were, for stroke 0.78 (0.52–1.19), myocardial infarction 0.96 (0.63–1.45), major bleeding 1.54 (1.14–2.08), and all-cause death 0.97 (0.80–1.19). Conclusions Patients treated with dabigatran had a 41% lower risk of major bleeding compared with rivaroxaban, but similar risks of stroke, MI, and death. Relative to apixaban, patients treated with dabigatran had similar risks of stroke, major bleeding, MI, and death. Rivaroxaban relative to apixaban had increased risk for major bleeding, but similar risks for stroke, MI, and death.

Currently, four NOACs are available for clinical use in patients with AF, including the direct thrombin inhibitor dabigatran and the factor Xa inhibitors rivaroxaban, apixaban, and edoxaban [3][4][5][6], but RCTs directly comparing these agents are lacking and available comparisons of individual NOACs are retrospective [13][14][15][16][17]. Studies based on claims databases have limitations in terms of data quality and follow-up duration. Prospective registries and cohort studies can provide more complete and accurate data, longer follow-up [18], and capture variables not included in claims databases [18,19].
The Global Registry on Long-Term Oral Antithrombotic Treatment in Patients with Atrial Fibrillation (GLORIA-AF) was a large, global, prospective registry providing comparative data on the use of NOACs in clinical practice. For this report, we performed a head-to-head comparison of NOACs for the outcomes of interest using the Phase III, final 3-year follow-up period of the GLORIA-AF Registry.

Study design and setting
The 3-phase design of the GLORIA-AF Registry Program has been described [20]. Consecutive patients ≥ 18 years old with recently identified AF and CHA 2 DS 2 -VASc scores ≥ 1, meeting the inclusion criteria (Online Resource: Methods 1) were enrolled and managed according to local clinical practice at the discretion of treating physicians. Patients in Phase III were followed for 3 years, regardless of antithrombotic therapy. The study was governed by Good Clinical Practice and the Declaration of Helsinki. The protocol was approved by the European Medicines Agency and institutional review boards at each participating site. Patients provided written informed consent. An independent academic steering committee oversaw the design, execution, study conduct, and manuscript development (Online Resource: Data Sharing Statement).

Clinical outcomes
The outcomes of interest were stroke (hemorrhagic, ischemic, and uncertain classification), major bleeding (International Society on Thrombosis and Haemostasis criteria), myocardial infarction, and all-cause death. Furthermore, the composite outcome of stroke, systemic embolism, myocardial infarction, vascular death, and life-threatening bleeding events was also analyzed (life-threatening bleeding events defined in Methods 2; Online Resource).

Statistical methods
Demographics and baseline characteristics were summarized descriptively and compared for dabigatran vs rivaroxaban, dabigatran vs apixaban, and rivaroxaban vs apixaban within different patient sets using standardized differences. Categorical variables were summarized by frequencies and percentages, and continuous variables as means and standard deviations (SD). For analyses comparing dabigatran vs rivaroxaban, dabigatran vs apixaban, and rivaroxaban vs apixaban, missing data for baseline covariates and cause of death were handled using multiple imputation (Online Resource: Methods 3). Descriptive analyses for the propensity score (PS)-trimmed and PS-matched sets are based on the PS calculated using the first of the multiple imputation patient sets, i.e., the first trimmed and matched sets. Outcome analyses were performed separately for each imputed patient set, and results were combined to provide estimates under the missing-at-random assumption. The PS was calculated separately for dabigatran vs rivaroxaban, dabigatran vs apixaban, and rivaroxaban vs apixaban, following which restricted sets and matched sets for the pairs compared were then derived. For each of the pairwise comparisons, outcome analyses were performed separately for the patient sets described below. Data were analyzed using SAS® software version 9.4 or later (SAS Institute, Inc., Cary, NC).

Patient sets
Post hoc comparisons of dabigatran vs rivaroxaban, dabigatran vs apixaban, and rivaroxaban vs apixaban were performed following the same methodology as defined for the analyses of the relative effectiveness and safety of dabigatran vs VKA among the PS-trimmed and PS-matched patient sets. The PS-trimmed set consisted of the cohort obtained after excluding those in the nonoverlapping tails of the PS distribution (PS-trimming) within each geographical region (Online Resource: Fig. 1, Fig. 2, and Fig. 3). Excluding patients from the tails of the PS distribution addresses channeling bias and improves the validity of comparisons. The PS-matched sets were generated from the PS-trimmed patient sets by 1:1 greedy nearest-neighbor matching of patients on dabigatran to those on rivaroxaban, dabigatran to apixaban, and rivaroxaban to apixaban, with a predefined caliper within the region (Online Resource: Methods 4). Descriptive analyses for the PS-trimmed and PS-matched sets were based on the PS calculated using the first trimmed and matched sets.

Clinical outcome analyses
Incidence rates with 95% confidence intervals (CIs) for key outcome events were calculated for dabigatran vs rivaroxaban, dabigatran vs apixaban, and rivaroxaban vs apixaban within the trimmed and matched patient sets. The initial analysis comparing effects of NOACs was conducted using a multivariable Cox regression model within the PS-trimmed patient set. The model included core variables (e.g., treatment, age, sex, and risk factors for stroke and bleeding). Further variables were included based on covariate selection procedures (Methods 5; Online Resource). Hazard ratios (HRs) with 95% CIs were presented for outcomes considered. The comparative analyses for dabigatran vs rivaroxaban, dabigatran vs apixaban, and rivaroxaban vs apixaban were also conducted in the PS-matched patient set by Cox regression with a shared frailty factor to adjust matching [21]. Among matched patients, the balance between the treatment groups was compared for individual, prespecified covariates (Online Resource: Table 1), and covariates with a standardized difference > 10% were considered unbalanced and included as a separate variable in the final regression model. Kaplan-Meier curves were plotted based on the matched patients for graphical comparison. Additionally, 1 3 we conducted a PS stratification analysis, based on strata formed by deciles of an extended PS and geographic region (Online Resource: Methods 6). Longitudinal outcomes were analyzed on an as-treated basis, censoring patients after permanent discontinuation of initial treatment or study termination.

Results
The GLORIA-AF Phase III registry included 21,591 patients enrolled at 935 sites in 38 countries, of whom 21,300 were eligible for analysis. Approximately 48% were enrolled in Europe, 24% in North America, 20% in Asia, and 8% in Latin America. The eligible patient population included 19,718 patients who received at least 1 dose of prescribed antithrombotic treatment and 1142 who did not initiate the prescribed antithrombotic treatment at baseline. Of the treated patients, 12,577 (60.3%) received a NOAC, either dabigatran (n = 3839), rivaroxaban (n = 4015), apixaban (n = 4505), or edoxaban (n = 332). Because of the small number of patients taking edoxaban, it was not further assessed. Baseline characteristics of patients treated with dabigatran, rivaroxaban, and apixaban are provided in the Online Resource (Table 2). A total of 17,140 (80.5%) patients completed the full 3 years of observation. When possible, information on vital status was collected for patients who did not complete the planned observation period, and at the end of the study was available for all but 997 (4.7%) patients.

PS-matched cohorts
The PS-matched set consisted of 2918 patients in each of the two treated groups, whose baseline characteristics are shown in Table 1B. Most were enrolled in Europe (65.2%). Of the dabigatran group, 58.2% of patients were prescribed 150 mg BID, and 39.4% received 110 mg BID; of the rivaroxaban group, 72.7% of patients received 20 mg OD. The incidence rate for the key outcomes within the PS-matched cohort are shown in Table 2B.

PS-trimmed cohorts
The PS-trimmed set included 3580 patients treated with dabigatran and 4154 patients treated with apixaban; their baseline characteristics are shown in Table 4A. Nearly half (49.2%) of patients treated with apixaban were enrolled in Europe, and 39% were in North America. Apixaban-treated patients had more comorbid conditions, including hypertension (77.1% apixaban vs 74.7% dabigatran), diabetes mellitus (22.7 vs 21.0%), coronary artery disease (18.2 vs 12.8%), chronic kidney disease (creatinine clearance < 50 mL/min: 14.2 vs 9.5%), and prior bleeding (4.4 vs 3.2%) compared with the dabigatran group. Concomitant antiplatelet therapy was more common among apixaban-treated patients (19.7 vs 12.8%). These differences are reflected in the PS Table 1 Baseline characteristics of dabigatran-and rivaroxaban-treated patients within the PS-trimmed cohort (A) and the PS-matched cohort (B) PS propensity score; AF atrial fibrillation; HAS-BLED hypertension, abnormal liver/renal function, stroke history, bleeding history or predisposition, labile international normalized ratio (INR), elderly (age > 65 years), drug/alcohol usage; OAC oral anticoagulant; OD once daily; BID twice daily; mg milligram *Standardized difference > 10% (in absolute value) is considered unbalanced between the two treatment groups # Concomitant use of drugs associated with higher bleeding risk (i.e., antiplatelet agent, Cox-2 inhibitor, or other non-steroidal, anti-inflammatory drug)  Fig. 2). Half of dabigatran-treated patients received 150 mg BID and 45.2% received 110 mg BID; over 80% of patients treated with apixaban received 5 mg BID. Incidence rates for the outcomes of interest in patients treated with dabigatran vs apixaban are shown in Table 5A.

PS-matched cohorts
After PS-matching, the study consisted of 2694 patients in each group. In the apixaban group, 79.6% of patients were treated with 5 mg BID, while only 57.6% of patients in the dabigatran group received 150 mg BID. Baseline characteristics are shown in Table 4B, and incidence rates for the outcomes of interest in Table 5B.

PS-trimmed cohorts
The PS-trimmed set included 3789 patients treated with rivaroxaban and 4227 treated with apixaban (Table 7A). The majority of patients came from Europe (50.1% in the rivaroxaban group and 49.2% in the apixaban group), but the proportion of patients prescribed apixaban was higher in North America (38.8% apixaban vs 33.7% rivaroxaban), while in Latin America, the proportion of patients receiving rivaroxaban was higher (7.1% rivaroxaban vs 3.6% apixaban). In terms of other characteristics, the

PS-matched cohorts
The PS-matched set consisted of 3559 patients in each of the two treated groups, whose baseline characteristics are shown in Table 7B. Most were enrolled in Europe (51.9%). Of the rivaroxaban group, 76.2% of patients were prescribed 20 mg OD, and 20.7% received 15 mg OD; of the apixaban group, 81.8% of patients received 5 mg BID and 18.0% received 2.5 mg BID. The incidence rates for the key outcomes within the PS-matched cohort are shown in Table 8B.
Cox regression analysis of the PS-matched patient set, adjusted for unbalanced variables, revealed that treatment with rivaroxaban was associated with increased risk of major bleeding (HR: 1.54; 95% CI: 1.14-2.08;   Table 7 Baseline characteristics of rivaroxaban-and apixaban-treated patients within the PS-trimmed cohort (A) and the PS-matched cohort (B) PS propensity score; AF atrial fibrillation; HAS-BLED hypertension, abnormal liver/renal function, stroke history, bleeding history or predisposition, labile international normalized ratio (INR), elderly (age > 65 years), drug/alcohol usage; OAC oral anticoagulant; OD once daily; BID twice daily; mg milligram * Standardized difference > 10% (in absolute value) is considered unbalanced between the two treatment groups # Concomitant use of drugs associated with higher bleeding risk (i.e., antiplatelet agent, Cox-2 inhibitor, or other non-steroidal anti-inflammatory drug) 95% CI: 0.82-1.25) were similar for rivaroxaban and apixaban. Post hoc sensitivity analyses using an extended set of covariates in the propensity score confirmed the PS-matched analysis (Table 9).

Discussion
In this large prospective study comparing various NOAC anticoagulants among patients with AF, the principal findings were that: (1) use of NOACs varied across world regions; (2) patients treated with dabigatran had a lower risk of major bleeding compared with rivaroxaban-treated patients, but risks of stroke, MI, mortality, and the composite outcome were similar with the two drugs; (3) there were similar risks of these events between patients treated with dabigatran or apixaban, and (4) patients treated with apixaban had a lower risk of major bleeding compared with rivaroxaban, but risks of stroke, MI, mortality, and the composite outcome were similar with the two drugs. Geographic differences in prescribing patterns included less frequent use of apixaban in Asia and Latin America than in Europe and North America. Dabigatran was more often prescribed in Asian countries, while apixaban and rivaroxaban were more frequently employed in European patients. Some of these differences may reflect site selection, the timing of regulatory approvals in various markets, study enrollment timelines, economic/reimbursement conditions, or other variations in healthcare settings [21,22]. While the study found broadly similar event rates with the individual NOACs, patients treated with rivaroxaban had a higher incidence rate of major bleeding compared with dabigatran. While in randomized trials all NOACs exhibited efficacy compared with warfarin, dosing was carried out differently. In the ARISTOTLE trial of apixaban [5] and the ROCKET-AF trial [4] of rivaroxaban, patients received standard doses unless reduced based on defined patient characteristics associated with increased drug exposure. In the RE-LY trial of dabigatran [6], two doses were administered in a randomized fashion without adjustment based on patient characteristics.
Observational studies have noted that NOAC dosing in clinical practice frequently does not align with labeled recommendations, with over 20% of patients prescribed the lower doses of rivaroxaban or apixaban in our analysis. We cannot ascertain whether dose selection for patients enrolled in GLORIA-AF was based on label-adherent criteria. A meta-analysis of 12 observational case-control and cohort studies using multivariable or propensity score adjustment to estimate relative effects found comparable risks of stroke or systemic embolism with rivaroxaban vs apixaban and apixaban vs dabigatran in adults with AF [13]. In that analysis, apixaban had the most favorable safety profile, based on a lower rate of major bleeding than dabigatran and rivaroxaban [13].
A Danish nationwide study of 31,522 patients with AF comparing the effectiveness and safety of standard and reduced doses of NOACs found similar rates of stroke with standard and reduced NOAC doses. Rivaroxaban was associated with a higher risk of major bleeding than dabigatran or apixaban and dabigatran was associated with a lower risk of intracranial bleeding [14]. Among patients receiving reduced doses, apixaban was associated with a lower risk of gastrointestinal bleeding than the other NOACs [14]. In Norwegian and Korean cohorts, rivaroxaban was associated with a higher risk of major bleeding than dabigatran or apixaban [16,17]. Whereas in the Global Anticoagulant Registry in the FIELD-Atrial Fibrillation (GARFIELD-AF) cohort of 25,551 anticoagulated patients with AF, the mortality rate was lower with NOACs than with VKA [23]. In that registry, therapy with dabigatran was associated with a lower risk of major bleeding than VKA (HR: 0.68; 95% CI: 0.47-0.98), while no substantial differences in the risk of bleeding were observed with FXa inhibitors compared with VKA (HR: 0.84; 95% CI: 0.63-1.12); but no separate analysis was performed for rivaroxaban and apixaban [23]. Data from these observational studies, suggesting an increased risk of major bleeding in rivaroxaban-treated patients compared with dabigatran and apixaban, and similar effectiveness with the various NOACs, are consistent with findings in GLORIA-AF.

Limitations and strengths
This study is subject to potential confounding by factors not adjusted for in the analysis. The use of multiple imputation to address missing data for comparative analysis may result in bias if the missing-at-random assumption is violated. Despite the broad variety of sites and physician specialties in Asia, Europe, Latin America, and North America, over 50% of the study group was enrolled in Europe. Of note, dosing recommendations and criteria for dose reduction for NOACs differ across the world, and dabigatran 110 mg BID is not registered in the USA or Japan. Despite the large size of the GLORIA-AF cohort, the study size was not sufficient to permit comparison of NOACs based on dosage. Furthermore, the small number of patients prescribed edoxaban precluded its inclusion in comparative analyses.
Despite these limitations, GLORIA-AF is the first prospective global study of consecutive AF patients receiving anticoagulants in routine clinical practice for over 3 years. Regular follow-up visits with physicians, on-site monitoring, and data quality assurance standards, and the low proportion of patients for whom vital status was unavailable (4.7%) yielded high-quality, reliable data.

Conclusions
In a routine clinical practice setting over 3 years, patients treated with dabigatran had a 41% lower risk of major bleeding compared with rivaroxaban and similar risks of stroke, MI and death. Relative to apixaban, dabigatran was associated with similar risks of stroke, major bleeding, MI, and all-cause death. Rivaroxaban relative to apixaban had an increased risk for major bleeding, however, similar risks for stroke, MI, and death.
Ethics approval GLORIA-AF was performed in accordance with the provisions of the Declaration of Helsinki, and the protocol and procedures were approved by the European Medicines Agency, as well as relevant institutional review boards and ethics committees where required.

Consent for publication Not applicable.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.