clinically ascribed in healthcare settings (i.e., ICD codes for AUD, requiring one inpatient or two outpatient ICD9/10 codes (Kranzler et al., 2019)) and another via self-report on a questionnaire (i.e., AUDIT) (Sanchez-Roige et al., 2019). Furthermore, the study cohorts differed -while the MVP includes mostly older male veterans with higher likelihood of AUD than the general population, the UK Biobank is a volunteer cohort of older individuals, both female and male, although not socio-economically representative of the UK. These differences could contribute to study-specific signals that may relate to different aspects of drinking, the extent of enrichment for AUD in the study cohort and to study-specific confounding (e.g., via socio-economic factors). Therefore, to minimize study-specific bias, we only retained variants that had the same directions of effects in both GWAS (2,757,680 variants) to exclude study-specific findings and findings due to random variations. These variants explained 23% (SE=0.0042) of variation by using LDSC (LD score regression) (Finucane et al., 2015). On the contrary, using all variants, the variation explained was only 5% (SE=0.002) by using LDSC (Finucane et al., 2015), indicating that many variants with lower P-values were actually study-specific and including them in calculating PRS would introduce noise therefore lower
