Presentation on Case Study: Application of Model Averaging to the IRIS Chloroform Assessment

This sub-product will consist of a presentation on the sub-product “Case Study: Application of Model Averaging to the IRIS Chloroform Assessment.” In that sub-product, the utility of model averaging will be demonstrated using a case study on the non-nested dichotomous endpoints from EPA’s recent IRIS risk assessment investigating health effects related to inhalation of chloroform. The sensitivity of the posterior estimates to the choice of prior distributions will be examined by applying multiple sets of priors. The model average results from the set of priors currently in BMDS will be compared to the PODs used in the chloroform assessment. A sensitivity analysis will be conducted using parameter priors of varying variability to investigate the effect of prior variability on the BMD estimates and confidence bounds. Model averaging will be conducted on continuous endpoints as the method becomes available for use. 

Impact/Purpose

Model averaging is a generalization of single model selection that is demonstrated to be statistically superior to single model selection. A number of external organizations have recommended its use in IRIS assessments. The intent of this case study is to motivate the use of model averaging in IRIS assessments by demonstrating its utility and feasibility.  

Citation

Blessinger, T. Presentation on Case Study: Application of Model Averaging to the IRIS Chloroform Assessment. Workshop: Advancing Quantitative Analysis in Human Health Assessments through Probabilistic Methods, Research Triangle Park, NC, October 07 - 08, 2024.