To assess causality among MD, AD, AC-quantity, and AC-frequency, we used GWAS summary association data to conduct two-sample MR analyses (Davey Smith and Hemani, 2014; Burgess et al., 2015). Since different MR methods have sensitivities to different potential issues, accommodate different scenarios, and vary in their statistical efficiency (Polimanti et al., 2017, 2018; Ravera et al., 2018; Wendt et al., 2018), we considered multiple MR methods (online Supplemental Table S1). These include methods based on median (Bowden et al., 2017), mean (Bowden et al., 2016), and mode (Hartwig et al., 2017), and various adjustments, such as fixed v. random effects (Bowden et al., 2017), Rucker framework (Rucker et al., 2011), and Steiger filtering (Hemani et al., 2017). We verified the stability of the results, comparing the effect directions across the different MR-variant filtering methods (online Supplemental Table S1). MR-Egger regression intercept was considered to verify the presence of pleiotropic effects of the SNPs on the outcome (i.e. to verify whether the instrumental variable is associated with the outcome independently from its association with the exposure) (Bowden et al., 2015). In
