A Method to Estimate Health Effects Based on Error-Prone Simulated Environmental Exposure: An Application to a Multi-Country Study on Birthweight and Fine Particulate Matter

Earth System Models provide spatiotemporally continuous environmental exposure data but remain underused in environmental epidemiology because of uncertainty from measurement errors. We developed a novel latent-variable approach to correct for measurement error characterized by spatiotemporal error covariance, which was derived from comparisons between Coupled Model Intercomparison Project Phase 6 (CMIP6) monthly fine particulate matter (PM_2.5) simulations and station-based monitoring data from 5,661 global sites. To demonstrate the utility of the framework, we associated these exposures to birthweight records from 132 Demographic and Health Surveys. The results showed variable correlations between the models and the observations (r = 0.40–0.68) as well as widely varying effect estimates across Earth System Models, from a 0.01 g (95% confidence interval: −0.85–0.87) reduction to a 15.11 g (12.69–17.54) reduction in birthweight per 10 μg/m³ increase in PM_2.5. After correcting measurement error, the optimal estimate indicated a more precise and consistent reduction of 3.34 g (2.57–4.11) in birthweight per 10 μg/m³. These findings demonstrate that the negative association between PM_2.5 exposure and birthweight is robust to different levels of measurement error embedded in CMIP6-based exposures, and that correction for measurement error in environmental epidemiology can help avoid misestimating the effect by reducing bias and improving consistency.