Science Inventory

Systematic Empirical Evaluation of Models to Inform Risk Prioritization

Citation:

Wambaugh, J., K. Isaacs, K. Phillips, C. Ring, J. Arnot, D. Bennett, P. Egeghy, P. Fantke, L. Huang, O. Jolliet, H. Shin, J. Westgate, AND Woodrow Setzer. Systematic Empirical Evaluation of Models to Inform Risk Prioritization. Presented at International Society of Exposure Science annual meeting, Ottawa, ON, CANADA, August 26 - 30, 2018. https://doi.org/10.23645/epacomptox.7075172

Impact/Purpose:

This is abstract is for a presentation that is part of a symposium entitled "Consensus Modeling of Chemical Exposure". The symposium will be part of the International Society of Exposure Science annual meeting in Ottawa, Canada.

Description:

Prioritizing the potential risk posed to human health by the thousands of chemicals that occur in commerce and the environment requires tools that can estimate exposure from limited information. High-throughput exposure (HTE) models exist to predict exposure via specific pathways. We consider an “exposure pathway” to include a chemical source, interaction with the environment, and a receptor (such as a person). Data for identifying the relevant pathways for chemicals in a high throughput manner are limited. Both expert opinion and conservative assumptions (all chemicals via all pathways) have drawbacks. Here we present a consensus HTE model based on 15 models and databases drawn from a collaboration involving five universities, a research consultancy, and the U.S. EPA. Machine learning predicts the probability that a chemical is associated with four assumed pathways: near-field (residential), dietary, far-field industrial, and far-field pesticide. On a pathway basis, we examine inferred chemical intake rates for the median U.S. population, from the National Health and Nutrition Examination Survey (NHANES). The intake rate for each pathway is either higher (residential, dietary) or lower (pesticidal, industrial) than the average. We use multivariate linear regression to evaluate the predictive ability of each model and database to raise or lower exposure with respect to the pathway averages. We can explain ~70% of the chemical-to-chemical variance in NHANES using the consensus model. We extrapolate model predictions to other chemicals by assuming that chemicals without biomonitoring data will have similar intake rates to those within NHANES. Predictions with quantified confidence intervals may allow risk-based prioritization using the margins between putative bioactive doses and intake rate predictions. This abstract may not reflect U.S. EPA policy.