A Cluster Analysis of Diagnoses and Symptoms

While there have been many applications of cluster analysis in psychiatric classification research, there are no studies in which cluster analysis is used to discover the taxonomic structure implicit in the DSM-III itself. In order to do so, the symptom index in the DSMIII- R manual was summarized in a two-way matrix of disorders by symptoms and then analyzed using a hierarchical classes model and companion algorithm (HICLAS) that permits overlap among classes. A novel feature of this model is that superordinate-subordinate relationships among diagnostic and symptom classes are explicitly represented. The HICLAS analysis revealed that there are several discrete symptom classes in DSM-III-R and that many psychiatric disorders can be modeled as combinations of one or more of these classes. The disorders associated with these symptom classes tend to fit the hierarchical classes model relatively well, particularly the mood disorders and the psychotic disorders. However, disorders such as adjustment, personality, and sexual disorder fit the model poorly or not at all. The results are in line with the conjecture that the taxonomic model implicit in DSM-III-R is a hybrid of discrete symptom classes and some other structure, perhaps a dimensional one.