Next unadjusted and adjusted hazard ratios and 95% confidence intervals using Cox regression were calculated with Whites as the reference group. In adjusted models, we controlled for age and family history. Analyses for onset of dependence and progression from drinking initiation to dependence excluded lifetime abstainers and also controlled for age at drinking initiation and alcohol consumption. For the hazard models predicting onset of dependence we did not remove individuals at the time they stopped drinking. We recognize that one strategy might be to remove former drinkers from risk set at the time they stopped drinking; however, we chose to conduct the analyses without doing so because former drinkers are still at risk of relapse, and we do not have annualized data to account for brief periods of non-drinking. Supplemental analyses attempting to censor former drinkers at their age of last drink were performed and resulted in no differences in the conclusions. Because of the complex survey design employed by the NESARC, analyses were conducted with attention to variation in analytical weights and clustering of respondents within sample strata.
