Statistical Analysis
We used the inverse variance-weighted (IVW) model as the main statistical method [33]. In sensitivity analyses, the weighted median method [34], MR-Egger method [35], Maximum likelihood, and Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) method [36] were performed to examine the consistency of associations and detect possible pleiotropy. The weighted median method provides a causal estimate when more than 50% of the weight in the analysis comes from valid instrumental variables [34]. The MR-Egger regression mode is able to detect pleiotropy by its intercept, but it compromises statistical power after pleiotropy is corrected [35]. The MR-PRESSO method, can detect and adjust for horizontal pleiotropy by outlier removal [36]. Cochran’s Q statistics were calculated to test for heterogeneity produced by different genetic variants in the IVW analyses.
Multivariable MR analysis was further applied to assess whether any association of domain-specific lifestyle behaviors with OC subtypes could be affected by potential confounders, including BMI and smoking. We also conducted a 2-step multivariable MR analysis in the analysis to adjust for BMI and smoking initiation. The associations were deemed significant associations at a strict Bonferroni corrected P -value below 0.00048 (correcting for 13 exposures and 8 outcomes), and associations with P -value > 0.00048 and < 0.05 were regarded as suggestive associations. All statistical tests were 2-tailed and performed using the TwoSampleMR [32], MR-PRESSO [36], and Mendelian randomization [37] packages in the R software (version 4.0.2; R Foundation for Statistical Computing, Vienna, Austria).