James Bennett and Jon Wakefield
Imperial College, London, UK
Explaining and quantifying the variability in response to drugs is an important step in producing efficacious and safe dosage regimens. In particular the identification of subpopulations (for example the elderly) who treat the drug differently is required. The identification of covariates or factors influencing the kinetics of the drug is therefore an important step in population pharmacokinetic modelling. Our aim is to be able to design an a priori dosage regimen for a new individual based only on his or her covariate measurements. The dose can then be refined following observations of the concentration of the drug gained from the a priori dose.
The number of covariates measured can be large and traditional methods of stepwise regression are both time consuming and inappropriate to the nature of the problem. We incorporate in a Bayesian hierarchical framework a procedure which aims to identify promising subsets of covariates. This method highlights the most promising models which can then be further evaluated by a full analysis.
An example is given for the antibiotic drug vancomycin when administered by multiple infusion to neonates. Nine covariates including both categorical and continuous measurements were available.
Reference: PAGE 3 () Abstr 850 [www.page-meeting.org/?abstract=850]
Poster: poster