Parametric approaches to the analysis of in vivo cytogenetics studies.

A number of parametric models are evaluated for the analysis of data arising from in vivo cytogenetics studies. Parallels between the structure of data arising from these studies and teratology studies are discussed. Evaluation of historical control data demonstrates the importance of recognizing non-independence of responses among cells within an animal. Models investigated that take into account this non-independence of responses include the beta-binomial, negative binomial, and correlated binomial models. Results indicate that the beta-binomial model performs best in terms of its ability to fit available control data. Suggestions for the incorporation of historical control data, based upon the two-stage structure of the candidate models, are made.