2.4. Bias assessment and statistical analysis

Two investigators (EC, WC) had also independently assessed the risk of bias and quality of included studies by using the Cochrane Collaboration Assessment Tool for randomized trials^[16] and the Newcastle-Ottawa Scale (NOS) for observational studies. The measure size of the included studies was chosen as risk ratio of treatment effect and publication bias was assessed by Egger test of intercept^[17] and Begg test^[18] (statistically significant with P ≤.1).

Continuous variables were expressed as mean ± SD and discontinuous variables were reported as number (percentage) and relative risk (RR) with 95% confidence interval (95% CI). The heterogeneity was assessed by Cochrane Q chi-square statistics and I² statistics.^[19] When I²≥50% and Cochrane Q chi² test P≤.1, we considered them as lacking of homogeneity and the pooled RR was evaluated by the DerSimonian–Laird method for random effects.^[20] When I² < 50%, a fixed effects model was used. Each of endpoints was corresponded to a classic Forest plot, including point estimates and 95% CI.

A sensitivity analysis was performed by omitting each study in turn when heterogeneity was found. The Revman 5.3 free package program and the statistical software package (Stata 14.0) were used for analysis. A 2-tailed P value < .05 was considered statistical significance.