Bayesian framework for extrapolation from multi-source preclinical data to human
Résumé
The development of a new drug requires a preclinical trial phase that includes in vitro, in vivo and in silico experiments. These experiments allowto determine a safe and effective starting dose range for the first-in-human study. Nevertheless, usually each experiment is analyzed independently, without using previous analysis results, and the final doses are often chosen based on one experience only. We propose a Bayesian framework for the extrapolation (from preclinical to clinical) of multi-source and multi-species data to predict the doses of interest (e.g. the minimum effective dose, the maximum tolerated dose, etc.) in human. A full Bayesian approach, divided in four main steps, is built to deal with the sequential estimation nature, the extrapolation, the commensurability checking and the information merging. The new framework is evaluated via an extensive simulation study, inspired by a real-life example in oncology, the inhibition of TGF-beta signaling to block tumor growth.