2.3 Selection of instrumental variables
In order to obtain enough SNPs, we relaxed the association threshold ofP < 1×10-5 [13]. LD r2 < 0.1 within a window of 500 kb was used to ensure the independence of the selected instrumental variables [14]. We calculated the F-value of each SNP and discarded those with an F-value < 10 to avoid bias from weak instrumental variables [15] (Supplementary Table 2). We looked for proxy SNPs that shared a high level of linkage disequilibrium (R2> 0.8) with the target SNPs if some SNPs were not in the outcome SNPs. In addition, we extracted the exposure SNPs from the outcome data and discarded those with a P < 0.05. Harmonization was performed to align the alleles of exposure- and outcome-SNPs and removed palindromic SNPs with intermediate effect allele frequencies or SNPs with incompatible alleles (Supplementary Table 3).