[51], clustering for familial relatedness, and included cross-term interactions as covariates (e.g., PRS*sex, PRS*age, etc.), as these have been shown to potentially bias effects in analyses involving PRS [52]. In addition, including cross-term interactions allowed for the primary analysis to investigate PRS*sex interaction effects. All analyses were repeated using the Million Veteran Program (MVP) PRS that were generated from summary statistics for the MVP GWAS on DSM-IV PTSD diagnosis [22] for European (EA) and African ancestry (AA) COGA participants (N: EA = 3522; AA = 1678) using PRS-csx [22]. To account for multiple comparisons while taking into consideration the correlation between our outcome variables (i.e., PTSD and substance dependence diagnoses; Supplementary Tables 4 and 5), we calculated adjusted p-values using the Benjamini–Hochberg Procedure [53] to decrease the false discovery rate in our models. Given that COGA’s sample is densely affected with alcohol use disorders, secondary exploratory analyses also included a problematic alcohol use PRS calculated from Barr et al. (2022) as a covariate to investigate any main and interaction effects of polygenic risk for problematic alcohol use in our sample [54].
