Mplus computed measures of global model fit that summarized how well the parameter estimates of the latent class model reproduced the 8,192 possible patterns of psychiatric comorbidity in the data. Because no single authoritative statistical method exists in determining the appropriate number of latent classes, consistent with previous approaches, we selected the optimal LCA models based upon model fit statistics and substantive interpretability of the model (Nylund et al. 2007). We specifically used the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), the sample-size adjusted BIC (aBIC), entropy, and the Vuong-Lo-Mendell-Rubin likelihood ratio test (VLMR-LRT) to assess model fit. Models with lower values on the AIC, BIC, and aBIC and higher values on the entropy statistic indicated better fit, and the VLMR-LRT indicated whether models including a greater number of latent classes were a statistically significant improvement over those including a smaller number. Substantive interpretability was evaluated by examining patterns of psychiatric disorder probabilities within and across classes to determine if the patterns appeared theoretically meaningful (e.g. consistent with prior conceptions of how psychiatric disorders relate to one another) versus
