Rapidly Parameterized Consumer Use Modeling for High-Throughput Exposure Predictions

Abstract

Background: High-throughput, typically in vitro, biological activity assays allow the ToxCast™ and Tox21 projects to compare the biological activity of chemicals with known in vivo toxicity to those with little or no in vivo data. Further in vitro assays characterize key aspects of pharmacokinetics and allow in vitro-in vivo extrapolation to predict human uptake (mg/kg/day) that might be sufficient to cause bioactivity in vivo. Aims: Without a similar capability to make quantitative, albeit uncertain, forecasts of exposure, the putative risk due to an arbitrary chemical cannot be rapidly evaluated. Methods: Using physico-chemical properties and provisional chemical use categories, most of the ~10,000 Tox21 chemicals have been evaluated with respect to exposure potential from near-field sources. A mapping of chemicals to products, and products to uses, has been used to categorize chemicals with respect to potential uses within the home. A Bayesian methodology is used to infer ranges of exposures consistent with biomarkers measured in urine samples and reported by the National Health and Nutrition Examination Survey (NHANES) in 2012. For each demographic group reported by NHANES we consider permutations of linear regression models, including as few as one and as many as all physico-chemical and use factors. Results: The linear regression provides a predictor, calibrated to the NHANES data, that can be applied to the remainder of the Tox21 list. The variance of this calibration serves as an empirical determination of uncertainty. These exposure predictions are then directly compared to the doses predicted to cause bioactivity for ~250 ToxCast chemicals. Conclusions: For chemicals with no other source of information, this approach allows prediction of confidence interval within which the average human exposure due to near field sources is likely. This abstract does not necessarily reflect U.S. EPA policy.