Data extraction, strength of evidence assessment, and bias assessment
Two reviewers (Mingjie Lin and Wenqiang Han) independently extracted the designed variables and data, including first author, year, country, study design, number of patients, age, sex, follow-up time, AF detection, dementia or cognitive deficit definition, CHADS2 or CHA2DS2-VASc Score,history of stroke, observational window, maximum adjusted covariates, defined OAC use, drugs of OACs, percentage of OACs use, comparison group, adjusted OR/RR/HR with 95% CIs. The correspondent authors were contacted by e-mail when overall adjusted OR/RR/HR were missing. The study was excluded if the authors did not reply. Another reviewer (Jingquan Zhong) checked the extracted data. Disagreements regarding any extracted data were resolved by discussion among the three reviewers.
Two reviewers (Mingjie Lin and Wenqiang Han) independently assessed the strength of evidence (SOE) and risk of bias. Any disagreements were resolved through discussion. The SOE was evaluated based on five domains: study limitations, directness, consistency, precision, and reporting bias; it was defined as four levels: high, moderate, low, insufficient.24 Risk of bias assessment was evaluated with the Cochrane risk of bias tool for RCTs20 and the Newcastle-Ottawa Scale (NOS) for observational studies.25 The RCT was classified as low, high, or unclear risk. The observational study with NOS score ≥ 8 was defined as high quality.