Harry Mager
Bayer AG, Centre of Drug Research, Institute of Clinical Pharmacology, Wuppertal, Germany
An important step in PK/PD modelling is the identification of descriptors accounting for the intersubject variability of pharmacokinetic parameter estimates. Reasonably well founded PK/PD models may be of utmost importance for explaining and predicting concentration-time courses in various subpopulations characterised by different covariate settings. The information accessible through pharmacometric models may thus contribute to patient-adjusted dose regimens, clinical trial design and power analysis.
The influence of several design parameters as well as of the multivariate correlation structure of the descriptor variables on the probability of chance correlations within the general framework of PK/PD models with regard to covariate selection is investigated. The performance of the most common subset selection procedures and criteria are compared and some general rules derived. Singular value decomposition of the descriptor space is suggested as an additional tool to reduce the probability of chance correlations in PK/PD modelling.
Reference: PAGE 8 (1999) Abstr 160 [www.page-meeting.org/?abstract=160]
Poster: poster