Likelihood misspecification and safe dose determination for clustered binary data

Abstract

Development toxicity experiments are designed to assess potential adverse effects of drugs and other exposures on developing foetuses. Several authors have advocated the use of dose-response models to predict safe exposure levels. Different definitions and approaches leading to a so-called virtually safe dose (VSD) have been introduced. Moreover, a range of parametric dose-response models have been developed. The aim of this paper is to compare a number of VSD estimators and to study the effect of misspecifying the probabilistic model. Attention is restricted to likelihood estimation. The methods are applied to data collected under the National Toxicology Program. The goal of these experiments was to investigate effects in mice of three different chemicals. The outcomes of interest are foetal malformations, classified as being external, visceral and skeletal.