Direct Oral Anticoagulants in Patients With Atrial Fibrillation (DOACs vs DOACs)

The objective of this study is to assess safety and effectiveness of direct oral anticoagulants (DOACs) for stroke prevention in patients with non-valvular atrial fibrillation (AF).

A common-protocol approach will be used to conduct retrospective cohort studies using administrative health care data from nine jurisdictions (the Canadian provinces of Alberta, British Columbia, Manitoba, Nova Scotia, Ontario, Quebec, Saskatchewan, as well as the United Kingdom (UK) Clinical Practice Research Datalink (CPRD) and the United States (US) Marketscan). Briefly, the Canadian databases include population-level data on physician billing, diagnoses and procedures from hospital discharge abstracts, and dispensations for prescription drugs. The data in Alberta, Nova Scotia, and Ontario will be restricted to patients aged 65 years and older, as prescription data are not available for younger patients. The CPRD is a clinical database that is representative of the UK population and contains the records for patients seen at over 680 general practitioner practices in the UK; these data will be linked to the Hospital Episode Statistics (HES) database, which contains in-hospital diagnosis and procedure data. US MarketScan includes individuals and their dependents covered by large US employer health insurance plans, and government and public organizations.

In each jurisdiction, the investigators will assemble a base cohort that includes all patients newly prescribed a DOAC for stroke prevention in AF. Study period will be from the date of the first DOAC approval for stroke prevention in AF at each site to the date of latest data availability at each site. All patients newly dispensed a DOAC (i.e. with no prescription for any oral anticoagulant in the prior year) with a diagnosis of AF within the 3 years prior to the date of the prescription will be eligible to be included into the study cohorts, given they present no exclusion criteria. The date of study cohort entry will be defined by the prescription (for CPRD) or dispensation (for all other sites) date of the newly prescribed DOAC. Patients will be censored at the earliest of death, end of healthcare coverage, or the end of the study period, whichever occurs first.

Exposure to a DOAC will be defined as a new prescription for a DOAC (apixaban, dabigatran, rivaroxaban) on the date of cohort entry. The investigators will use an analysis analogous to an intention-to-treat approach. The primary outcome will be defined as a hospitalization or emergency department visit for ischemic stroke or systemic embolization. The secondary outcomes will be: 1) major bleeding; 2) a composite of stroke (ischemic or hemorrhagic), systemic embolization, major bleeding or all-cause mortality; 3) myocardial infarction; 4) gastrointestinal bleeding; 5) intracranial bleeding; and 6) all-cause mortality.

The study cohort will be analyzed using a matched cohort design. In each participating site, three distinct study cohorts will be assembled, one for each of the following comparisons, nested in the same base cohort: 1) dabigatran vs. rivaroxaban, 2) dabigatran vs. apixaban, and 3) rivaroxaban vs. apixaban. DOAC users will be matched 1:1 to DOAC users on sex, age, cohort entry date, and propensity score (which will be constructed using a multivariable logistic regression model estimating the odds of being treated with DOACs, while adjusting for a number of pre-identified covariates to account for baseline differences at the time of cohort entry). Cox-proportional hazards regression models will be used to estimate adjusted hazards ratios (HRs) and corresponding 95% confidence intervals (CIs) for ischemic stroke or systemic embolization in the three cohorts. Meta-analyses of the site-specific results will be performed using random effects models. As secondary analyses, the composite outcome will be stratified by age (<85 and ≥85) and sex. In addition, an as treated analysis using inverse probability of censoring weights (IPCW) will be performed to account for non-random censoring.