Vladimir K. Piotrovsky, Achiel Van Peer
Human Pharmacokinetics, Janssen Research Foundation, Beerse, Belgium
A patient population is usually heterogeneous with respect to response to drug therapy. In any clinical efficacy trial there are patients who respond and those who do not respond or even deteriorate. One can assume responders and non-responders form subpopulations, and their responses can be described by independent distributions each characterised by a separate mean and/or variance. The population modelling package NONMEM provides a tool to separate subpopulations and get estimates of means and variances of distributions. This software has a useful option, $MIX, that enables splitting a composite distribution into a mixture of two or more distributions. An example of a dose-response analysis will be presented where the response is a number of events per week (a count). The drug increases the number of evens, and individual responses are assumed to follow the Poisson distribution. In the data subject to analysis, there were placebo and active treatment groups receiving various doses, and assessments were performed weekly. A mixed-effects model was developed that included time as a predictor, however, no explicit function for response versus time profile was assumed. Also, no explicit model for dose-response relationship was implemented. Two subpopulations were assumed: responders and non-responders. The model was fitted by minimizing minus twice log(likelihood) with the use of the first-order conditional Laplacian estimation method. Finally, each patient was assigned to either responder or non-responder group, and dose-response was assessed as the change in the proportion of responders with the dose. Fitting the model without subpopulations to the same data set resulted in much worse fit (AIC, BIC and likelihood ratio test). Advantages of the proposed approach versus standard approaches to dose-response analysis will be discussed.
Reference: PAGE 9 () Abstr 109 [www.page-meeting.org/?abstract=109]
Poster: poster