For EA participants, the discovery dataset used to calculate PGSAUD was from the meta-analysis of two large EA GWAS of AUD-related phenotypes: AUD determined using ICD codes from the Million Veteran Program (MVP-AUD; N=202,004, 7.3% are female, Mean age=63.3. GWAS results are available through dbGaP: https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs001672.v11.p1) (Kranzler et al., 2019) and scores derived from the problem subscale (questions 4–10) of the Alcohol Use Disorder Identification Test (AUDIT) from the UK Biobank (UKBB-AUDIT-P; N=121,604, 56.2% are female, Mean age=56.1. GWAS results are available from the senior authors upon request) (Sanchez-Roige et al., 2019). As described previously, since both GWAS used different instruments and had different participants, we only retained variants with the same direction of effects in both MVP-AUD and UKBB-AUDIT-P to exclude study-specific findings and false positives due to random variations, i.e., retained variants were likely AUD-associated in general populations (Lai et al., 2022a). PRS-CS (Ge et al., 2019) was used to estimate the posterior effect sizes of each variant through a Bayesian regression framework using continuous shrinkage priors. European samples from 1000 Genomes Project were used as the LD
