Abstract
IMPORTANCE: Identification of individuals at high risk of alcohol use disorder (AUD) and subsequent application of prevention and intervention programs has been reported to decrease the incidence of AUD. The polygenic score (PGS), which measures an individual's genetic liability to a disease, can potentially be used to evaluate AUD risk. OBJECTIVE: To assess the estimability and generalizability of the PGS, compared with family history and ADH1B, in evaluating the risk of AUD among populations of European ancestry. DESIGN, SETTING, AND PARTICIPANTS: This genetic association study was conducted between October 1, 2023, and May 21, 2024. A 2-stage design was used. First, the pruning and thresholding method was used to calculate PGSs in the screening stage. Second, the estimability and generalizability of the best PGS was determined using 2 independent samples in the testing stage. Three cohorts ascertained to study AUD were used in the screening stage: the Collaborative Study on the Genetics of Alcoholism (COGA), the Study of Addiction: Genetics and Environment (SAGE), and the Australian Twin-Family Study of Alcohol Use Disorder (OZALC). The All of Us Research Program (AOU), which comprises participants with diverse backgrounds and conditions, and the Indiana Biobank (IB), consisting of Indiana University Health system patients, were used to test the best PGS. For the COGA, SAGE, and OZALC cohorts, cases with AUD were determined using Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) or Fifth Edition (DSM-5) criteria; controls did not meet any criteria or did not have any other substance use disorders. For the AOU and IB cohorts, cases with AUD were identified using International Classification of Diseases, Ninth Revision (ICD-9) or International Classification of Diseases, Tenth Revision (ICD-10) codes; controls were aged 21 years or older and did not have AUD. EXPOSURE: The PGS was calculated using single-nucleotide variants with concordant effects in 3 large-scale genome-wide association studies of AUD-related phenotypes. MAIN OUTCOMES AND MEASURES: The main outcome was AUD determined with DSM-IV or DSM-5 criteria and ICD-9 or ICD-10 codes. Generalized linear mixed models and logistic regression models were used to analyze related and unrelated samples, respectively. RESULTS: The COGA, SAGE, and OZALC cohorts included a total of 8799 samples (6323 cases and 2476 controls; 50.6% were men). The AOU cohort had a total of 116 064 samples (5660 cases and 110 404 controls; 60.4% were women). The IB cohort had 6373 samples (936 cases and 5437 controls; 54.9% were women). The 5% of samples with the highest PGS in the AOU and IB cohorts were approximately 2 times more likely to develop AUD (odds ratio [OR], 1.96 [95% CI, 1.78-2.16]; P = 4.10 × 10-43; and OR, 2.07 [95% CI, 1.59-2.71]; P = 9.15 × 10-8, respectively) compared with the remaining 95% of samples; these ORs were comparable to family history of AUD. For the 5% of samples with the lowest PGS in the AOU and IB cohorts, the risk of AUD development was approximately half (OR, 0.53 [95% CI, 0.45-0.62]; P = 6.98 × 10-15; and OR, 0.57 [95% CI, 0.39-0.84]; P = 4.88 × 10-3) compared with the remaining 95% of samples; these ORs were comparable to the protective effect of ADH1B. PGS had similar estimabilities in male and female individuals. CONCLUSIONS AND RELEVANCE: In this study of AUD risk among populations of European ancestry, PGSs were calculated using concordant single-nucleotide variants and the best PGS was tested in targeted datasets. The findings suggest that the PGS may potentially be used to evaluate AUD risk. More datasets with similar AUD prevalence as in general populations are needed to further test the generalizability of PGS.