Exposure to fine particulate matter (PM2.5) poses significant health risks and accurately determining the shape of the relationship between PM2.5 and health outcomes has crucial policy implications. Although various statistical methods exist to estimate this exposure-response curve (ERC), few studies have compared their performance under plausible data-generating scenarios.
This study compares seven commonly used ERC estimators across 72 exposure-response and confounding scenarios via simulation. Additionally, we apply these methods to estimate the ERC between long-term PM2.5 exposure and all-cause mortality using data from over 68 million Medicare beneficiaries in the United States. Our simulation indicates that regression methods not placed within a causal inference framework are unsuitable when anticipating heterogeneous exposure effects. Under the setting of a large sample size and unknown ERC functional form, we recommend utilizing causal inference methods that allow for nonlinear ERCs.
In our data application, we observe a nonlinear relationship between annual average PM2.5 and all-cause mortality in the Medicare population, with a sharp increase in relative mortality at low PM2.5 concentrations. Our findings suggest that stricter limits on PM2.5 could avert numerous premature deaths. To facilitate the utilization of our results, we provide publicly available, reproducible code on Github for every step of the analysis.
Members, non-Members, all welcome. |
Michael Cork, Harvard University, USA (presenting)
Daniel Mork, Francesca Dominici Harvard University, USA (co-authors)