dependence) and nicotine persistence (among those with baseline nicotine dependence). These models controlled for demographics specified above, and were run to determine the unique effects of the varied maltreatment types. Fourth, logistic regressions were run to determine whether having more types of childhood maltreatment affected persistence risk incrementally. For these analyses, separate logistic regressions were run for alcohol and for nicotine; each included the count of childhood maltreatment types as a predictor, and demographics (and other childhood adversities) as covariates. The childhood maltreatment count variable was specified as continuous to determine the overall effect of an increasing number of maltreatment types, and then re-run as categorical to determine the effects of each level of maltreatment exposure. All analyses were performed in SUDAAN software (RTI International, Research Triangle Park, North Carolina) to adjust for the complex survey design. SUDAAN software properly adjusts standard errors for the clustered sampling designs through the use of Taylor series linearization. Logistic regression results are presented using adjusted odds ratios (AORs) and 95% confidence intervals (95% CIs).
