Environ Sci Technol. 2025 Dec 18. doi: 10.1021/acs.est.4c13918. Online ahead of print.
ABSTRACT
There is increasing interest in addressing measurement error in gridded exposure estimates. Here, we provide a framework to account for exposure measurement error in gridded air pollution estimates used for health effects estimation, i.e., when numerical air quality models or satellite-derived data are the principal source of exposure data. We employed a two-stage Bayesian hierarchical modeling framework consisting of an exposure measurement error model and a health model, linking the unobserved true ambient exposure at the residential address for each participant in the cohort to health outcomes. In sensitivity analysis, we considered different health models, spatial smoothing parameters, and spatial resolutions (1.33 and 4 km grid cells) for observation-fused CMAQ output. In an example application, comparison of gridded observation-fused CMAQ estimates and spatially smoothed observation-fused CMAQ estimates by leave-one-out cross-validation at monitoring stations indicated that prediction accuracy at those locations is comparable between the 1.33 km resolution gridded estimates and the smoothed estimates, while the spatially smoothed estimates slightly outperform the 4 km resolution gridded estimates. Accounting for exposure measurement error also resulted in somewhat greater PM2.5 health effect estimates on continuous neuroimaging outcomes and smaller health effect estimates on binary cardiovascular outcomes, although overall conclusions remained similar.
PMID:41411036 | DOI:10.1021/acs.est.4c13918
