While the study took advantage of latent class models to define risk groups, these groups were not pre-defined nor were the number of groups pre-selected. This method was chosen as we are unaware of any accepted a priori definitions for high craving states. Indeed the point of the methodology was to determine if such groups existed. The latent class model was robust in identifying three classes but two of the classes had small samples and clinical outcomes suggested that the mid- and high craving classes were more similar than distinct. Finally, all the data were self-reported with may account for some of the associations between affect and craving as there may be a tendency to self-report more symptoms. A replication of these results should include more subjects and consider an a priori two-class model.
