Subtyping depression: testing algorithms and identification of a tiered model.

We seek to distinguish psychotic, melancholic, and nonmelancholic depression by clinical features and to test varying algorithm models to determine optimal criteria sets. We report a study of 269 depressed inpatients and outpatients. A latent class analysis (LCA) of 16 clinical features allowed for specificity or overrepresentation of features to be examined across the three classes. Varying algorithm models for distinguishing melancholic and nonmelancholic depression, involving endogeneity symptoms and observer-rated psychomotor disturbance (PMD) were compared. Psychotic depression was readily distinguished by the specific presence of psychotic features, and PMD was most severe in this class. Melancholic depression was most clearly distinguished from the residual nonmelancholic class by the presence of PMD. Although some endogeneity symptoms were overrepresented in the melancholic class, their specificity was unimpressive. An algorithm involving PMD components alone was highly efficient in discriminating LCA classes and, more importantly, superior to DSM-IV decision rules when examined against a range of clinical validators of melancholia. Subtyping appears assisted by a hierarchical model, based on a small set of features. The move from nonmelancholic to melancholic depression appears defined by a tier of observably rated PMD, whereas the move from melancholic to psychotic depression is determined by a tier of psychotic features and contributed to by significantly higher levels of PMD.