Data synthesis and analysis
We obtained pooled and adjusted ORs with 95% CI estimates of GDM for the exposure and control groups using the DerSimonian and Laird random-effects method. We calculated the absolute risk difference for GDM between the exposure and control groups in low- (control group: GDM risk was assumed to be 2.0%), moderate- (10%), and high- (20%) prevalence groups, using the pooled odds ratios (OR) and 95% confidence intervals (CIs). This assumption was made based on a previous report and our clinical expertise 40.
Publication bias was assessed qualitatively by visual inspection of the funnel plot and quantitatively by Egger’s test 41. Where asymmetry was observed in the funnel plot, we investigated the likely source of this asymmetry using the contour-enhanced funnel plot.
We evaluated between-study heterogeneity visually, using forest plots, and quantitatively, using I 2 and τ2 statistics. We used the Cochrane chi-square test to calculate I 2 and τ2statistics. We performed a pre-specified subgroup analysis based on types of exposure (preterm birth, LBW, or SGA). In pre-specified sensitivity analyses, we used crude ORs instead of adjusted ORs and excluded studies using non-standard definitions of GDM. Some studies assessed the risk of GDM among women born with a weight >4000 g (macrosomia); these studies were excluded from post-hoc sensitivity analysis, as a previous review has shown a U-shaped association between mother’s birth weight and GDM risk20.
All analyses were performed using STATA 14.2 (StataCorp LP, Texas) and RevMan 5.4 (Cochrane Collaboration, UK). Two-sided p- values <0.05 were considered indicative of statistical significance.