MR Sensitivity analysis
To assess the robustness of MR effect estimates to potentially invalid genetic variants, we conducted a sensitivity analysis using the weighted median and MR-Egger methods when at least five genetic instruments were available for sex hormones. The weighted median method produced a consistent causal effect estimate even when up to half of the genetic variants included in the MR analysis were invalid instrumental variables26. When a weaker assumption (i.e., Instrument Strength Independent of Direct Effect, INSIDE) was met, the MR-Egger method was able to estimate causal effects while allowing all SNPs to be pleiotropic 27.
Previous studies have suggested that obesity may be a potential cause of COVID-19 susceptibility and severity 28,29. To minimize the risk of bias resulting from violating the exclusion restriction assumption (i.e., IVs can only affect the outcome through exposure rather than alternative pathways), we identified genome-wide significant SNPs associated with obesity traits such as BMI and waist-hip ratio based on published GWASs in the PhenoScanner V2 database (http://www.phenoscanner.medschl.cam.ac.uk/), and removed these SNPs from the IVW analysis 30.