A systematic review of RWE studies enrolling patients with NVAF was the basis for this meta-analysis. The methodology of the review adhered to the guidance from the Centre for Reviews and Dissemination (CRD) from the University of York [10] and the Cochrane Handbook for Systematic Reviews of Interventions [11]. Detailed results of the SLR were published separately [12].
The population of interest was adults (aged ≥18 years) with NVAF receiving an oral anticoagulant. Both studies reporting on incident (i.e., beginning anticoagulant treatment) and prevalent (i.e., continuing treatment) patients were included. The interventions of interest were the Factor Xa inhibitors apixaban and rivaroxaban and the direct thrombin inhibitor dabigatran.
The following databases were searched on 1 December 2016: Medline and Embase (accessed using the Ovid platform), and the Cochrane Library (accessed via Wiley Interscience), including the Cochrane Database of Systematic Reviews (CDSR), Database of Abstracts of Reviews of Effects (DARE), the Cochrane Central Register of Controlled Trials (CENTRAL), the Health Technology Assessment (HTA) database, and the NHS Economic Evaluation Database (NHS EED). No restrictions were applied in terms of publication date, language, or geographical scope. Details of the search strategy are presented in the Supplementary Material.
Two independent reviewers performed the study selection, and any differences were resolved by a third reviewer. Extracted data were those on citation characteristics, study details, patient characteristics, results, and study limitations; all extracted data were quality-checked by a second reviewer.
The outcomes of interest relating to drug efficacy were: IS, all-cause mortality, myocardial infarction (MI), venous thromboembolism (VTE), a composite of ischemic stroke or systemic embolism (IS/SE), and a composite of IS/SE/all-cause mortality. Outcomes of interest relating to drug safety were: HS, ICH, major bleeding, gastrointestinal (GI) bleeding, and any bleeding. A final outcome of interest was persistence/non-persistence, defined as a break in treatment of at least 60 days.
The following three comparisons were made: 1) rivaroxaban vs. VKAs, 2) dabigatran vs. VKAs, and 3) apixaban vs. VKAs using the inverse variance-weighted method to pool hazard ratios (HRs) and their 95% confidence intervals (CIs) in a meta-analysis. The inverse variance-weighted method was used based on a common assumption that the ln(HR) followed a normal distribution. The analysis was conducted on HRs to take into account adjustments on baseline characteristics made in each study. In line with previously published methodology [13], when no HR was available the incidence rate ratio was used instead. If there were no events in one arm of a study, a continuity correction was applied, while studies with no events in either treatment arm were excluded from the analyses [11]. Details of input calculations are provided in the Supplementary Material.
If results at different follow-up times were available in a study, the longest follow-up was used. Additionally, if more than one study used the same database, only the study with the highest level of precision was used (i.e., only the study for which the standard error of the log of HR was the smallest was included). For example, 21 studies assessing dabigatran vs. VKAs based on the MarketScan® database were identified. Inclusion of studies using the same database and investigating similar outcomes could lead to the same patients being repeatedly included in the analysis, which could bias the results. Analyses did not account for different doses (i.e., data for 15 mg and 20 mg rivaroxaban doses, and 110 mg and 150 mg dabigatran doses were pooled together).
Heterogeneity between studies was assessed using the p-value of the Cochrane Q test and the I-squared, as recommended by the Cochrane Handbook for Systematic Reviews of Interventions [11]. Both the fixed- and random-effects models were fitted. Given the heterogeneity between study designs, results of the random-effects model are presented. Analyses were conducted using SAS 9.3®. Meta-analysis results are presented in the text as the number of studies included in the comparison (n), and HR with its [95% CI], unless otherwise indicated.