Grading the evidence
Based on the strength and validity of evidence, the associations between environmental risk factors and childhood cancer were classified into strong, highly suggestive, suggestive and weak. A strong association was indicated, when the p -value of the random-effects meta-analysis was below 10−6, the number of cancer cases was greater than 1000 to significantly reduce false positive findings, the largest study in meta-analysis was nominally statistically significant (p<0.05), heterogeneity was low to moderate (I2 <50%), there was no indication of small study effect or excess significance bias, and the 95% prediction intervals excluded the null value. A highly suggestive association was claimed, if the p -value of the random-effects meta-analysis was below 10−6, the number of cancer cases was greater than 1000, and the largest study in meta-analysis was nominally statistically significant (p <0.05). The criteria for a suggestive association were fulfilled if the p -value of the random-effects meta-analysis was below 10−3, and the number of cancer cases was greater than 1000. All other nominally statistically significant associations (p <0.05) were considered to have weak evidence.
The primary analysis in this umbrella review focused of all studies included in each meta-analysis. Sensitivity analyses were conducted including only the meta-analyses with four or more cohort or nested case-control studies. Analyses were performed using Stata version 14 (College Station, TX) and all p -values were two tailed at a 5% significance level.