The Bayesian Approach To Categorical Responses

Jon Wakefield, Dave Lunn, Nicky Best, David Spiegelhalter

Department of Epidemiology and Public Health,Imperial College School of Medicine, St Mary's Hospital, London, W2 1PG, UK

There are many situations in pharmacodynamic modelling when a categorical response variable is obtained. Examples include pain scores, number of attacks of hay fever and the presence/absence of side-effects. In this talk we will discuss the way in which categorical variables may be modelled using Bayesian methods. In particular binary and

polychotomous data with random effects will be discussed. We will also focus on models that have been fitted within the pharmacokinetic literature. The advantage of the Bayesian approach that we suggest is that there is no need to rely on asymptotic arguments for inference, though there is an increase in the computation time. We will show how the computations may be carried out via the addition of latent variables using the BUGS software.

Reference: PAGE 6 () Abstr 591 [www.page-meeting.org/?abstract=591]

Poster: oral presentation