DOSE-RESPONSE MODELS FOR CORRELATED MULTINOMIAL DATA FROM DEVELOPMENTAL TOXICITY STUDIES

Developmental anomalies induced by toxic chemicals may be identified b y using laboratory experiments with rats and mice. This paper examines dose-response models for correlated multinomial data arising in studi es of developmental toxicity. A hierarchical probability structure is used to unify and extend previous methods of modelling correlated bina ry data arising in such studies. Generalized estimating equations in c onjunction with an extended Dirichlet-multinomial covariance function are used to estimate jointly the regression parameters in Weibull dose -response models for both embryolethality and fetal malformations. Qua dratic estimating equations are used to estimate the overdispersion (c orrelation) parameters. Maximum likelihood estimates based on a Dirich let-multinomial distribution are also computed for comparison. The mod elling procedures proposed are illustrated through the analysis of dat a from a recent large scale study of the developmental toxicity of the herbicide 2,4,5-trichlorophenoxyacetic acid.