Exploration on Explanations for Observed Long-Term Temporal Trend of the Short-Term Association between Fine Particulate Matter Concentration and Hospital Admissions in the U.S.

摘要

Background In earlier work, we examined whether the PM2.5-hospital admissions relative risk varies over time, using hospital admission data of U.S. Medicare beneficiaries and U.S. Environmental Protection Agency PM2.5 monitoring data. When assuming a linear temporal trend, the change in respiratory hospital admissions per 10 μg/m3 increase in same day PM2.5 decreased by 1.77% (95% CI: 0.92 to 2.63%) from the year 1999 to the year 2013, while no statistical significant temporal trend was observed for cardiovascular hospital admissions. To better understand drivers behind the observed temporal trends, this study provides details in our model specifications, as well as exploration on explanations to these observations. Methods To ensure internal validity of the modified Bayesian hierarchical model used, we conducted extensive sensitivity analyses with hypotheses tested stated a priori. Based on previous literatures and biological plausibility, we proposed and evaluated four possible explanations for the decreasing temporal trend for respiratory admissions as well as the lack of temporal trend for cardiovascular admissions: 1) the increase in relative measurement error at lower level of PM2.5 concentration; 2) flattening of concentration-response function at lower level of PM2.5 concentration; 3) change in chemical composition of PM2.5 total mass; 4) change in population susceptibility towards acute exposure to PM2.5 on respiratory adverse health outcomes. Results and conclusions Based on qualitative evidence, changes in population susceptibility towards acute exposure to PM2.5 on respiratory adverse health outcomes, and combination of changes in multiple chemical components may partially explain the observed temporal trend. Future research should consider distributed lag model and non-linear concentration response functions, which will require additional exposure data.