As shown in Supplement S1, improvement in the AIC, BIC, and aBIC steadily increased until the four-class model, whereas any further improvements in statistical fit were minor for the five-class model, and models with six- to eight-classes did not converge. The four-class model also had good substantive interpretability (i.e. subtypes defined largely by distinctive probabilities of internalizing and externalizing of psychiatric disorders, as described in the following paragraphs). We deemed that the four-class model was superior based on the combination of substantive interpretability and the pattern of model fit.
