Weighted frequencies and their respective 95% confidence intervals (95% CI) were computed. Univariate and multivariate discrete-time time survival analyses (with person-year as the unit of analysis) [29] were implemented to assess the association between covariates and the probability of remission among individuals with lifetime substance dependence. Associations are expressed as hazard ratios measuring relative risk. The person-year variable was defined as the number of years from dependence onset to the age of remission or age at interview (for censored cases). DSM-IV mood, anxiety and comorbid substance use disorders, level of education and marital status were included as time-varying covariates. Taylor series linearization methods implemented in SUDAAN (version 9.1) (Research Triangle Institute, Research Triangle Park, N.C.) were used to estimate standard errors and significance and accommodate for complex survey design. Predictors reaching statistical significance at the 0.2 level in the univariate analyses were included in the multivariate model. Descriptive and discrete-time survival models were implemented using SUDAAN.
