This study has several limitations. First, both AOU and IB are not representative of general populations of European ancestry, and the prevalence of AUD in the AOU cohort is much lower than in general populations. More datasets are needed to further test the PGS generalizability. Second, familial AUD may have different AUD-associated genes; therefore, using the COGA and OZALC datasets in the screening stage could include familial AUD genes but may miss those from nonfamilial AUD cases, resulting in decreased PGS generalizability. Third, we excluded individuals with other substance use disorders from the controls in the screening datasets. Although this increased statistical power, the estimated effect sizes may be biased and thus could also decrease the PGS generalizability. Fourth, although the same SNVs were used in calculating PGS, different studies used different genotyping platforms, which have a substantial effect on PGS distributions, making it challenging to determine uniform thresholds to define high or low risks in different studies. Fifth, we did not perform PGS analysis in populations of non-European ancestry because the sample sizes were small. Furthermore, for admixed populations
