Bayesian Model Averaging for Benchmark Dose Analysis in Developmental Toxicology

Abstract

To reduce uncertainty due to model selection when a large num ber of potential candidate models is available, the use of Bayesian Model Averaging (BMA) has emerged as an important t ool. As known, the BMA methodology is a coherent approach sin ce we can express the desired quantities as a weighted average o f model specific quantities with the weights determined base d on how much the data supports each model. In toxicological studies , a wide range of statistical models have been utilized for do se-response modeling and risk assessment with no particular model recei ving a universal acceptance. Here, we consider the applicat ion of BMA for benchmark dose estimation in developmental toxicity ex periments. In such experiments, as in all noncancer studies , th choice of the model can play a crucial role in the final benchmark dose es timates. A Bayesian approach along with the MCMC method is us ed to fit each individual model used as a component in model avera ging and to derive the posterior weights. A simulation study of a developmental toxicity experiment is used to illustrate th methodology.