Two-Sample MR Analysis

Before statistical analysis of causal relationships, we ensured that GWAS exposure and outcome data matched well. First, information about exposure-associated SNPs was extracted from each outcome dataset. Any exposure-associated SNP absent from the outcome dataset was substituted with a proxy SNP; that is, if it existed, and it was in LD (r² > 0.8, MAF for palindromes <0.3) with the requested one. Second, the exposure and outcome data were harmonized to ensure that the effect of the SNP was on the same allele. Otherwise, the SNP was deleted.

Third, the statistical significance of the matched exposure-outcome summary data was analyzed by several methods. Inverse variance weighted (IVW) analysis assumed no or balanced pleiotropy. We used the random-effects model to avoid heterogeneity bias, which was measured using Cochran's Q test. The MR-Egger method not only allowed for, but also detected, horizontal pleiotropy based on its intercept with a y-axis. When the intercept was not zero, there was horizontal pleiotropy. The MR-Egger method was based on the “instrument strength independent of the direct effects” (INSIDE) and “no measurement error in the SNP exposure effects” (NOME) assumptions, and it also used the random-effects model. In addition, when the regression dilution I² statistic was <90%, indicating violation with the NOME assumption, the simulation extrapolation (SIMEX) correction was performed (Bowden et al., 2016). The WM method was used to generate unbiased results when ≥50% SNPs were valid variants.

Fourth, SNP pleiotropy and sensitivity were assessed using pleiotropy tests, forest plots, funnel plots, and leave-one-out plots. The forest plot estimated the causal effect of each SNP on the outcome by using Wald ratio analysis. The Funnel plot was used to assess heterogeneity by depicting the reciprocal of the se of the SNP against SNP effects on the outcome. The leave-one-out plot ascertained whether an association was disproportionately influenced by a single SNP. In such a plot, each black point represented the IVW analysis after exclusion of that particular SNP.