3.2 Mendelian randomization analysis
To make sure the causal relationship in MR analysis is reasonable, the instrumental variable (IV) assumptions must meet the following three conditions: (1) It must closely relate to BW. (2) It must not be relevant to other confounding factors. (3) It only affects cancers through BW. Before the MR analysis, we calculated the power value and F statistic of each cancer GWAS studies we chose to test whether the IV was strong enough to explain the exposure under the existing sample size50. (power>80%, F>100) It is based on simulations and specific parameters for two-stage least squares (2SLS) MR analyses to make sure that the degree of deviation in estimating causal correlation was within an acceptable range. The main statistical test we used to estimate BW for different cancers is a random-effects inverse-variance weighted (IVW) meta-analysis of the Wald ratio for individual SNPs. Besides, we also applied other methods including the weighted median, weighted mode and MR-Egger regression methods to test the third assumption. Then we analyzed the accuracy of MR results in three aspects. First, we conducted a heterogeneity test to identify the differences between each IVs. Furthermore, the intercept of MR-Egger and MR-PRESSO were used to check the gene pleiotropy ensuring the feasibility of the second assumption. The MR-PRESSO was a recently published method for testing gene-level pleiotropy which could assess it more accurately 51. At last, we employed a leave-one-out sensitivity analysis to assess the sensitivity of each IVs to MR results. Several palindromic SNPs were moved to decreased the bias of our MR analysis. (Table S5) Our MR analysis was conducted using the package “Two Sample MR (version 0.5.5)”.