"Model-based Bayesian Dose Response Modeling"
Neal Thomas, Pfizer
December 2, 2008 @ 3:30 pm - 4:30 pm
Location: BRB - II/III - Room 253
Biostatistics
TITLE: Model-based Bayesian Dose Response Modeling
Evaluating dose response is a basic activity in drug development. There has been renewed interest in improved statistical designs and analyses for dose response in clinical studies because of the recognition that dose response is not well estimated for most drugs despite past attempts at improvement, and because improved logistics (internet-based randomization systems with drug supply tracking) have made more complex designs feasible to implement. Plots from several drugs will display common features of clinical dose response curves. These common features will be used to suggest potential prior distributions for parameters of the Emax curve, which is ubiquitous in pharmacology. A heuristic hierarchy of prior information is proposed, with some prior information based on generic information obtained from compounds across many therapeutic areas, and other information very specific to the compound being developed. An on-going Bayesian dose adaptive study utilizing an informative prior distribution will be described. Large sample theory is routine in this setting, but several practical problems that arise in real implementation, which are not addressed by the theory, will be discussed.
