Model Selection in Toxicology: Principles and Practice

Abstract

The key assumptions underlying modern toxicology are that (1) animals can serve as accurate predictive models of toxicity in humans, (2) that selection of an appropriate model to use is the key to accurate prediction in humans, and (3) that understanding the strengths and weaknesses of any particular model is essential to understanding the relevance of specific findings to humans. It must first be made clear, however, that the term model encompasses more than a test organism. Then the basic principles underlying these assumptions and the general limitations of models (both in vivo and in vitro) can be critically examined. The actual practice of model selection and why it deviates from the generally accepted principles are overviewed, along with the issue of errors induced by the “high-class model” problem. Historical results as to the accuracy of prediction of human effects are summarized. Finally, the reasons that models fail to predict effects in humans in some cases are examined, along with methods and problems in cross-species extrapolation and the issue of species peculiarities.