You are here:
A MULTIMODEL APPROACH FOR CALCULATING BENCHMARK DOSE
Citation:
Garcia, R. AND R W. Setzer. A MULTIMODEL APPROACH FOR CALCULATING BENCHMARK DOSE. Presented at Society for Risk Analysis, Baltimore, MD, December 7-10, 2003.
Description:
A Multimodel Approach for Calculating Benchmark Dose
Ramon I. Garcia and R. Woodrow Setzer
In the assessment of dose response, a number of plausible dose- response models may give fits that are consistent with the data. If no dose response formulation had been specified a priori then
the analyst is left with the decision of which model to select to represent the data. Typically, in this situation the analyst uses a model selection criterion to derive which model is superior to
the others and proceeds to make inferences using this superior model. However, inferences made using the superior model will result in standard errors which are biased downward because they
do not take model uncertainty into account. This effect can become significant when one is using the model to extrapolate away from the data. This study examined whether the use of multiple plausible models provides a more robust estimate of Benchmark Dose (BMD) compared to using the one single model. First, this study established a collection of models typically
used to model quantal responses in the toxicological literature, based on the list of models in EPA's BMDS software. Next, datasets were simulated using physiologically based pharmacokinetic (PBPK) models coupled to various pharmacodynamic (PD) models, to generate data for which the "true" underlying model was known. The BMD, corresponding to an increase in incidence of 10%, was estimated (with standard error) from the data using the model with the lowest Akaike Information Criterion and the Akaike-weighted average BMD from the all the models in the collection. The 95% confidence intervals of BMD using both methods were compared to the actual BMD. This process was performed for several PBPK and PD models. Simulation results indicated that the confidence interval using the Akaike - weighted average BMD included the true BMD more often that the single model approach. The confidence interval provided using the single model was much shorter in length than the Akaike - weighted averaged BMD method but it was at the cost of not including the true BMD as often. The Akaike-weighted average BMD is relatively simple to calculate and provides a more robust measure of BMD than using one single model. (This is an abstract of a proposed presentation and does not necessarily reflect EPA policy).