Mixture Models for Heterogeneity in Biomedical Disorders

Donna K. Pauler, Michael D. Escobar, and John A. Sweeney

Abstract:

Finite mixture analysis is an important tool in the biomedical sciences for detecting subgroups within populations. For clinical syndromes that may be composed of discrete subgroups with distinct pathophysiologies, this technique can be useful for addressing important questions about heterogeneity. A potential problem with the application of finite mixture analyses in practice is that these models may drastically overestimate the number of component densities. In such situations a continuous mixture model provides a viable alternative. In this paper, we apply a mixture analysis to a specific oculomotor component of eye-tracking dysfunction in schizophrenia. In the context of this example, we illustrate some difficulties inherent in the applied use of mixture analyses and some techniques to overcome these difficulties. We highlight the importance of type of heterogeneity present in the data and the role of continuous and finite mixture densities in modeling each type of heterogeneity.

Keywords: Finite mixture models, negative binomial models, parametric bootstrap, EM-algorithm, schizophrenia.