Investigation of statistical methods for determination of benchmark dose limits for retinoic acid-induced fetal forelimb malformation in mice

The benchmark dose (BMD) is an estimated dose that induces an adverse response above the background level after a toxic exposure. In the present study, different statistical methods, especially those proposed for developmental toxicity studies, were evaluated by using selected toxicological endpoints (fetal malformation probability, fetal ulna length, and roundness) from a study in which pregnant mice were exposed to retinoic acid. The intra-litter correlation was accommodated either by assuming a beta-binomial distribution or by using generalized estimating equation (GEE) methods. The lower confidence limits (BMDLs) were estimated by the delta method, the likelihood-ratio based method, and the bootstrap method. The results indicated that the malformation probability was the most sensitive endpoint. The likelihood-ratio based method and bootstrap method tend to give tighter confidence limits compared to the delta method. While the likelihood-ratio based method could not be used in GEE models, the bootstrap method did not have this limitation.