Dr Stephens
Imperial College London, Dept. of Mathematics Huxley Building, 180 Queen's Gate, London SW7 RBZ, England
Popkan is a PC package designed to implement Bayesian techniques for the analysis of population pharmacokinetics data based on a three-stage hierarchical model. The first stage, representing individual profiles via established PK models has a non-linear structure with multiplicative (lognormal) observation error structure as default. The second stage, representing population characteristics, is presumed to be Student-t in order to ensure inferences are robust to population outliers. The third stage represents the variability in population parameters. Bayesian analysis in such a framework is not straightforward to implement via conventional techniques. However, using sampling based methods (Gibbs Sampler, Markov chain Monte Carlo), full Bayesian inference may be performed.
The Popkan software has three sections specification, computation, and results analysis. The user interface is written using the SAS statistical package, and the bulk of computation performed using Microsoft Fortran. In the specification section, the user is required to input information relating to the data source, the appropriate PK model for the data, parameters of interest in the model, quantitative and qualitative prior opinions about the parameters, information about relevant covariates, and an observation error model. It is also possible to restrict analysis to a specific individual with population parameters fixed (and the third stage redundant). In the results analysis section, numerical and graphical summaries or the posterior distributions of individual model profile and population parameters, outlier detection, diagnostic statistics, Bayesian prediction, and covariate selection are available options.
Reference: PAGE 2 () Abstr 901 [www.page-meeting.org/?abstract=901]
Poster: oral presentation