of risk (e.g., whether trauma exposure is a predictor at certain ages). Trauma exposures, PTSD, MDD, tobacco and alcohol involvement were modeled as time-varying covariates coded as negative in each year prior to the age at first occurrence and positive for that year and each subsequent year, so that only covariates preceding the outcomes were considered informative. Analyses were conducted in Stata (StataCorp, 2007) using Huber-White robust standard errors to adjust for the non-independence of observations in twins. Violations of the pH assumption were resolved by estimating piecewise hazard ratios (HRs) for each age ranges of ≤14, 15–18, and ≥19. All analyses were adjusted for age at most recent interview, maternal nicotine dependence, and parental educational attainment. Additionally, the CUD symptom onset analyses included early cannabis use status as a covariate.
