PNNL

Abstract

PM2.5 air quality index (AQI) data and smoke density data from the Hazard Mapping System (HMS) are often used as proxies for personal chemical exposure. However, it is not well understood how these measurements compare to personal chemical exposure measured by silicone wristbands. We hypothesized that predictive models for personal chemical exposure would be significantly improved by expanding beyond PM2.5 AQI or HMS data to also include environmental and behavioral factor information. In Eugene, Oregon, participants wore daily wristbands, carried a phone that recorded GPS locations, and answered daily questionnaires for a seven-day period in multiple seasons. We also gathered publicly available daily PM2.5 AQI data from stationary air monitors and HMS data from satellites. Some summer sampling dates coincided with heavy wildfire smoke in the region. We analyzed wristbands for 94 organic chemicals, including 53 polycyclic aromatic hydrocarbons (PAHs). We used machine learning models to predict the level of chemical exposure using PM2.5 AQI only, HMS only, and a multivariate feature set including PM2.5 AQI, HMS, and other environmental and behavioral information. Wristband chemical detections and concentrations, behavioral variables (e.g., time spent indoors), and environmental conditions (e.g., PM2.5 AQI) significantly differed between seasons. Multivariate models using the additional environmental and behavioral variables performed significantly better than the PM2.5 AQI model or HMS model for all chemicals modeled. The multivariate model’s predictive accuracy was 70.1% and 70.2% times greater, on average, compared to the AQI model or HMS model, respectively. This study provides evidence to better understand key predictors of personal chemical exposure and shows that PM2.5 AQI data from stationary monitors alone or HMS data alone is insufficient to explain personal chemical exposures.