Model-based clustering of binary longitudinal atopic dermatitis disease histories by latent class mixture models.

Atopic dermatitis is a skin disease which affects mainly children, has a very strong genetical component, and manifests itself clinically as flexural excema in connection with torturing itching. The course of disease is notoriously changeable and runs in phases, therefore it is difficult to predict the future course of disease. To improve prediction it would be interesting to identify clusters of children with different disease histories because this would shed light on common genetic and environmental risk factors. We use, relying on previous work of Nagin, a Latent class mixture model to estimate, in a data-dependent and model-based fashion, a clustering of typical binary atopic dermatitis disease histories in children. The data were collected from 1990 to 1997 in the so called MAS-study, a prospective cohort study of 1314 children in five German cities. The original method of Nagin is extended in two different aspects, first we use bootstrap confidence intervals to account for uncertainty in curve fitting, and second, we propose to model covariates for cluster membership by Anderson's Stereotype regression model. We feel that the Latent class mixture model is a valuable tool for assessing the course of atopic dermatitis, yielding a wealth of communicable and graphically displayable results.