High throughput occupational exposure estimation using workplace compliance data and Bayesian hierarchical modeling
On this page:
New approaches are needed to rapidly screen the potential exposure and health hazard posed by tens of thousands of chemicals approved for production and use in the United States and elsewhere. High-throughput computational methods have been used to screen substances based on general population exposure potential from sources such as ambient air, consumer products and drinking water. However, worker exposures have significant uncertainties due to high workplace variability in chemical production and use, consequently these occupational exposures remain relatively uncharacterized for many substances. We present a high-throughput, data-driven approach that leverages a database of over 1.2 million observations of chemical concentrations in U.S. workplace air samples (spanning 1984-2018) to aid in estimating occupational exposure in the U.S. . We fit a two-stage Bayesian hierarchical model that uses the North American Industrial Classification System’s (NAICS) industry sector and subsector classifications and the physicochemical properties of each substance to predict workplace air concentration distributions. This model greatly outperforms a null model when predicting whether a substance will be detected in an air sample (75.7% detect/non-detect classification accuracy) and, if detected, at what concentration (root-mean-square error of 1.00 log10 mg m-3) when applied to a held-out test set of substances that were not used for training the model. For 57% of the held-out air samples, our model was able to correctly predict both detection or non-detection, and for detects, the measured air concentration within 1 order of magnitude. A null model that did not consider workplace type or physicochemical properties achieved this accuracy on only 35% of samples. We also found that workplace air exposure patterns varied strongly across industry types and physicochemical properties. This modeling framework can be used to predict chemical air exposure distributions for novel or data-poor substances and provide an improved estimate of occupational exposures within the context of high-throughput, risk-based chemical prioritization efforts.