Mixture distributions in psychiatric research.

This paper describes the application of Gaussian mixture distributions to biological marker research in psychiatry. Mixtures of univariate and multivariate normal distributions can be used to determine if diagnostically similar psychiatric patients belong to biologically distinct subpopulations. The resulting biological subtypes may be important in understanding the etiology of psychiatric disorders. The general model and estimation procedure are described (EM algorithm; Dempster, Laird and Rubin 1977). The method is illustrated using two examples of biological data: (1) red cell membranes and monoamine oxidase activity data in normal individuals having no family history of psychiatric illness, the first-degree relatives of bipolar depressed patients and a heterogeneous patient population; and (2) smooth pursuit eye movements that classify relatives of schizophrenics, nonschizophrenics and normal controls into biologically distinct populations.