Multinomial logistic regressions within the RMLCA model were estimated to assess the association between baseline patient characteristics and drinking patterns. Specifically, baseline characteristics were used to predict expected class membership in each of the latent classes of the RMLCA model. All models were estimated using Mplus version 7.3[40]. The Bayesian Information Criteria (BIC) and sample-size adjusted BIC (aBIC) were examined to select the RMLCA model with the best overall model fit, where lower BIC and aBIC indicate a better fitting model[41]. Classification precision (defined by relative entropy) was used to evaluate how well the final latent class solution classified individuals into latent classes[42].Considering the complex sampling design in each of the studies (i.e., recruitment from multiple sites), all parameters were estimated using a weighted maximum likelihood function and all standard errors were computed using a sandwich estimator (i.e., MLR in Mplus[43]). The robust maximum likelihood estimator provides the estimated variance-covariance matrix for the available drinking data and, therefore, all available drinking data during treatment were included in the models. Mplus could not accommodate missing data in the predictors (n=262; 6.8%
