2.2 Genetic instrumental variables selection
To obtain robust instrumental variables that are sufficiently correlated with the exposure of interest, SNPs related to IMIDs were chosen with a genome-wide significance threshold of p < 5e-08. Only two potential instrumental variables associated with AA were identified at the threshold of p < 5e-08. Therefore, we modified a more lenient threshold of p < 5e-06 for genome-wide significance to search for additional SNPs. Then, we set parameters related to linkage disequilibrium (r2 = 0.001, distance = 10000 kb) to clump these genetic loci, in which all loci exhibit complete independence in order to avoid double-count the contribution of any particular variants(29). Next, with the help of the online tool PhenoScanner v2 (http://www.phenoscanner.medschl.cam.ac.uk/ ), we checked genotype-phenotype associations between potential IVs and outcome(30), and removed SNPs associated with confounders. Then, SNP harmonization was conducted to correct the orientation of the alleles(31). Finally, we assessed IVs strength by calculating the genetic instruments (R2) and F-statistic for each instrumental variable. R2 and F-statistic was calculated as following formula: R2=[2β2 × MAF × (1-MAF)] / [2β2 × MAF × (1-MAF)+2β2 × MAF × (1-MAF) × N × SE], and F=[R2 × (N-k-1) /k(1-R2)] (where β = effect estimate of the SNP in the exposure, MAF = minor allele frequency, N = sample size of the GWAS dataset, k = number of IVs, SE=the standard error of genetic effect) (32). Theoretically, all selected genetic IVs should have F-statistics greater than 10.