A random-effects mixture model for classifying treatment response in longitudinal clinical trials.

A random-effects regression model that allows the random coefficients to have a multivariate normal mixture distribution is described for classifying treatment response in longitudinal clinical trials. The proposed model is capable of dealing with longitudinal data from unknown heterogeneous populations. As applied to longitudinal clinical trials, for example, the model can distinguish subgroups of treatment response. Use of the proposed model is illustrated by analyzing data from two psychiatric clinical trials. The first includes depressed patients assigned to drug treatment who are repeatedly measured in terms of their level of depression. The second trial examined schizophrenic patients longitudinally who were assigned to either a drug or placebo condition. For both, the random-effects mixture model allows an assessment of whether patients comprise distinct populations in terms of their treatment response. Based on parameter estimates of the mixture model, ample evidence for a mixture of response to treatment is observed for both datasets.