Latent class analysis (LCA) was conducted to identify subgroups of individuals based on the pattern of endorsement of the 7 DSM-IV AD criteria (Muthen and Muthen, 1998-2011). LCA may be viewed as a version of non-parametric cluster analysis, and makes the assumption of conditional independence (i.e. endorsement of one criterion within a class is uncorrelated with other criteria). There are two important estimates recovered from LCA: (a) the prevalence of individual classes of individuals with similar item endorsement profiles and (b) the likelihood of endorsing an item conditional on class membership. The Bayesian Information Criteria (BIC) and entropy (reflecting level of misclassification) were utilized to determine which model (i.e., number of latent classes) provided the best fit to the data.
