Nicky Best, Wally Gilks, Keith Tan
MRC Biostastistics Unit, Institute Of Public Health, University, Forvie Site, Robinson Way, Cambridge CB2 2SR, UK
The Gibbs sampler is a computer-based algorithm for exploring posterior probability distributions in complex modelling situations. Observations are simulated from the full conditional distributions of each parameter in the model given all the other parameters and the data. The samples of simulated values are then used to reconstruct the joint and marginal distributions for the parameters of interest.
We and others (e.g. Wakefield et al. 1991) have applied the Gibbs sampler to the estimation of pharmacokinetic parameters using routine clinical data. The method permits a flexible approach by allowing to population pharmacokinetic modeling. For example by allowing specification of t-distributed measurement errors which are robust to outlying observations common in routinely collected data. Inclusion of covariate information is also straight-forward, and the methodology can be extended to incorporate pharmacodynamic models. In our presentation, we will discuss an analysis of gentacimin pharmacokinetics in neonates using the Gibbs sampler. This dataset has also been analysed by Thomson et al.(1987) using the NONMEM software, and a brief comparison of the 2 methodologies will be made.
Reference: PAGE 2 (1993) Abstr 904 [www.page-meeting.org/?abstract=904]
Poster: oral presentation