A growing number of epidemiological studies have provided strong evidence for the adverse health effects of air pollution. However, the association at levels below the U.S. Environmental Protection Agency (EPA) standards (12 µg/m3 of annual average PM2.5) is unclear. In addition, a traditional regression framework does not have a causal interpretation, due to sensitivity to model choice. Our goal was to recreate an experimental design starting from observational data, in order to strengthen the causal interpretation of the link between low levels of PM2.5 and hospital admissions. One of the major issues in environmental epidemiology is confounding. Using aggregate exposure to PM2.5 (two-years prior annual average levels at zip codes level) and all-cause hospitalization rate, we compared the standard regression-based approach, in which the confounders at the zip code level were treated as covariates, to an approach in which zip codes were initially matched according to the confounders and then the effect of PM2.5 on health was estimated with regression models restricted to matched zip codes. We showed that observed confounders widely differed depending on the PM2.5 levels. We estimated that, even in very low levels of PM2.5, increasing long-term exposure to PM2.5 by 1 μg/m3 causally increased all-cause admissions by 6.2% (95% CI = 3.8%, 8.7%) when the range of PM2.5 was 3.50-7.83 μg/m3 ,9.2% (95% CI = 1.9%, 6.9%) with a range of 7.84-8.65 μg/ m3 and 12% (95% CI = 4.7%, 19.8%) when the exposure range was 9.37-10.29 μg/ m3 using nearest-neighbor matching. With Mahalanobis distance matching method we estimated that increasing long-term exposure to PM2.5 by 1 μg/ m3 causally increased all-cause admissions by 4.7% (95%CI = 2.3%, 7.1%), 10.2% (95%CI = 2.2%, 18.9%) and 16.1% (95%CI = 8.7%, 23.9%) in the same restricted range of PM2.5 respectively. In addition, also the analysis with all variables as covariates, showed that increasing long-term exposure to PM2.5 by 1 μg/m3, even in very low levels, causally increases all-cause admissions. Our study was rooted in potential outcomes methods for causal inference that consisted of a design phase that sought using observational data to approximate the design of randomized experiments, where “unexposed” (T = 1) and “exposed” (T = 0) units were balanced with respect to observed confounders; and an outcome analysis phase where the causal effects of adverse health effects to air pollution exposure were estimated. We provided strong evidence of different confounders at each shift in exposure to PM2.5 and the developed method was robust to model misspecifications. Last but not least, we showed that long-term exposure to PM2.5 were causally associated with all-cause hospitalizations, even for exposure levels not exceeding the U.S. EPA standards, suggesting that adverse health effects occur at low levels of fine particles.
|Titolo:||Estimating the Causal Effect of Low Levels of Fine Particulate Matter on Hospitalization|
|Data di pubblicazione:||5-feb-2018|
|Appare nelle tipologie:||8.01 Tesi di dottorato|
File in questo prodotto:
|Estimating the Causal Effect of Low Levels of Fine Particulate Matter on Hospitalization_RIGGI EMILIA.pdf||tesi di dottorato||N/A||Open Access Visualizza/Apri|