Risk of bias
Risk of bias within studies, concerning the primary outcome of interest, was assessed in duplicate with the National Institutes of Health Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies[10], modified to better fit our studies (Supplementary table 1). Disagreements were resolved by consensus. Studies with an overall score ≥80% were deemed as high quality. Publication bias was evaluated both visually – using contour-enhanced funnel plot (with the level of statistical significance set at 1%, 5% and 10%) – and with Egger’s test[11].