To assess the adverse effect of a toxin, animal bioassay experiments are frequently designed and performed on animals. In a typical experiment, pregnant mice are exposed to a dose of a chemical during a critical period of gestation. Animals are sacrificed before term and uterine content is examined for the number of fetuses that are dead/resorbed, malformed or normal. The outcomes are used to fit a dose-response relationship for risk assessment and determination of safe dosage levels. The choice of the dose-response model can severely affect the results. Although several models can fit the data well in the experimental dose ranges, when extrapolated to low human exposure levels, the estimate of safe exposure levels can vary substantially. To reduce uncertainty in the presence of multiple outcomes from developmental toxicity experiments, we propose the application of the Bayesian Model Averaging (BMA) technique whereby a series of candidate dose-response models are used, the safe exposure level is determined based on each model, and a weighted average is used for the final estimate. Simulation studies are presented to show that the methodology works well and can reliably be applied in practice. The methodology is further illustrated using a real experimental data set.
Digital Object Identifier (DOI)
Khorsheed, Eman and Razzaghi, Mehdi
"Model Averaged Benchmark Dose Analysis for Multiple Outcomes in Developmental Toxicity Experiments,"
Applied Mathematics & Information Sciences: Vol. 14:
4, Article 22.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol14/iss4/22