Cox proportional hazards models with mixed effects (Cox, 1972; Therneau, 2022) were used to generate hazard ratios (HR) for each predictor/outcome combination while including fixed effect covariates (above) and nuclear family as a random intercept to account for data non-independence. Models were run predicting: alcohol transitions from earlier EXT, INT, and EF phenotypes; EXT and INT symptom onsets from earlier alcohol milestones; alcohol transitions from EXT, INT, and EF PGS; and EXT and INT onsets from PGS for drinks/week and problematic alcohol use. Hazard ratio estimates required for 80% power ranged from 1.16 to 1.93 given sample sizes for models estimating survival to later stage-based symptom onsets excluding ADHD symptoms (see online Supplementary Methods; Supplementary Tables S14–S15). Due to insufficient numbers of events following alcohol milestones for agoraphobia and ADHD symptoms, these outcomes were excluded from analyses. Proportional hazards violations were investigated and resolved (Supplemental Methods).
