Statistical analyses
The included studies with adjusted OR/RR/HR and 95% CIs were meta-analyses. The RRs were used as the general effect estimates of studies; HRs were considered comparable to RRs;9 ORs were transferred to RRs using the published method.26The RRs and corresponding 95% CIs were calculated using the DerSimonian-Laird random-effects model. Statistical heterogeneity was tested using Cochran Q test (significance level: 0.1) and I2 Statistics (>50%: considerable heterogeneity).27 To explore possible sources of heterogeneity, subgroup analyses were performed based on study designs (prospective vs. retrospective studies), a history of stroke (patients with a prior stroke only vs. patients with and without prior stroke), observational window (with vs. without); sensitivity analysis was carried out by the leave-one-out method. Publication bias was explored with a funnel plot, and statistically evaluated by the Begg’s and Egger’s tests, and the trim-and-fill method. All analyses were performed using the STATA 16.0 (StataCorp, College Station, TX, USA), and ap -value of <0.05 indicated a statistically significant difference.