The gSEM analysis includes results on EA from the GENOA meta-analysis, Million Veteran Program (MVP), Psychiatric Genetics Consortium Substance Use Disorders Group (PGC-SUD), and the Partners Health cohorts. The MVP results are based on a meta-analysis including MVP parts 1 and 2 (total of 8529 OUD cases and 71,200 opioid-exposed controls), along with Yale-Penn and Study of Addiction: Genetics and Environment (SAGE) cohorts (total of 10,544 OUD cases and 72,163 opioid-exposed controls)27. The PGC-SUD results include 4,503 opioid dependence cases and 4173 unexposed controls26. Unexposed controls are used for the PGC-SUD because the exposed controls results have negative heritability estimates. The partners health cohort includes 1,039 OUD cases and 10,743 exposed controls28. Note that the GENOA GWAS, Yale-Penn, and SAGE parts of the MVP, and the PGC-SUD results include overlapping samples. However, accounting for this sample overlap is a feature of the gSEM approach applied in this study. A total of 2,434,903 variants were present in all cohorts and tested for association with the OA latent variable in the gSEM analysis using a total sample size of 403,915 (23,367 cases and 384,619 controls; effective sample size of 88,114).
