Meta-Analysis Protocol for Statistical Analysis

Statistical analysis was performed using STATA 12.0 software. First, we transformed ORs, RRs, or HRs and their CIs to their natural logarithms and SEs. We directly considered HR as RR7 (link)19 (link) and computed the combined RRs and 95% CIs from the estimates reported in each study20 (link)21 (link). Heterogeneity was quantified using I² values and chi-square test22 (link); when both I² ≤ 50% and p > 1.0 indicated no or acceptable heterogeneity23 (link), we used the fixed-effects model; otherwise, we used the random-effects model. In addition, we performed subgroup analyses on the basis of stratified ORs, RRs, and HRs, given that these pooled may result in the overestimation of OR variance24 (link). We also conducted a dose-response meta-analysis using STATA 12.0 software with restricted cubic spline function by the method of Orini25 (link) for those studies reported sufficient data, including RRs, serving size, and the sample size in each categories. Furthermore, we performed subgroups analyses on the basis of the study design, and type of cancer, adjustment, and definition of reference group. Publication bias was assessed by visual inspection of the funnel plots26 (link).