The remaining analyses involving the resulting class membership variables were performed in SAS V9.1. Weighted Kappa was employed to assess agreement between the latent class variables for each pair of the craving and positive and negative affect variables. To analyze associations between the latent classes for each outcome with baseline variables and other outcome variables, we used the multinomial logistic regression model with the three-class membership variable as the outcome and baseline or other outcome variables as covariates. To analyze associations between the latent classes and time to subsequent relapse drinking, we used the Cox Proportional Hazards model.
