Air Pollution: From Exposure to Health

Abstract

This work is intended to cover the stages of research in environmental epidemiology: beginning with exposure assessment, to determining health effects, to seeing how information can be aggregated from individual studies in order to inform policy through a meta-analysis. In Chapter 1, we describe a model to predict outdoor exposure to ambient Black Carbon (a marker of diesel exhaust) in Massachusetts, Rhode Island and Southern New Hampshire from 2000 to 2012. We built this model using Nu-Support Vector Regression – a machine learning technique which incorporates nonlinear terms and higher order interactions, with appropriate regularization of parameter estimates. Our data consisted of 24,301 measurements from 368 monitors in addition to both spatial and temporal predictors. The 10 fold cross validated R2 of the model was good in both cold (10-fold CV R2 = 0.73) and warm seasons (CV R2 = 0.75). In Chapter 2, we implement a novel approach to estimate the marginal risk of all-cause mortality associated with exposure to PM2.5. We carried out our analyses among approximately 12 million movers enrolled in Medicare between 2000 and 2012 and in a subset of about 10 million movers exposed to less than 12 µg/m3 of PM2.5. In our first analysis we assumed that exposure was randomized and we used a cox model in order to estimate the risk. In our second analysis we did not make the assumption of randomization, but rather used inverse probability weights (IPW) in order to control for confounders. We then ran a cox model in using these weights. We found an increased risk of all-cause mortality of 1.16 (95% Confidence Interval (CI): 1.16 – 1.17) per 10 µg/m3 of PM2.5 in both the first and second analysis. Among movers exposed to less than 12 µg/m3 of PM2.5 the risk was 1.27 (95% CI: 1.26 – 1.27) and 1.26 (95% CI: 1.25 – 1.26) per 10 µg/m3 of PM2.5 in the non-IPW adjusted and IPW adjusted analyses respectively. In Chapter 3, we carried out a meta-analysis for the effect of air pollution on cognitive outcomes in 3 age groups: children aged 5 and younger, children aged 6 and older and among adults aged 55 and older. In the last age group we also looked at the risk of Dementia. We combined results from studies using different pollutants and different tests by adjusting the reported estimated. Among children aged 5 and younger we found a significant pooled estimate for psychomotor development (-4.71, per 10 µg/m3 of PM2.5, 95% CI: -7.25 to -1.09) but not mental development. We did not find a significant pooled estimate for children aged 6 and older. Among older adults we found a significant effect for global cognition (-0.16 per 10 µg/m3 of PM2.5, 95% CI: -0.29 to -0.03) but not for memory. Finally, we found a pooled hazard ratio of 1.48 for Alzheimer’s Disease (95% CI: 0.9 to 2.42) of PM2.5 and a pooled hazard ratio of 1.87 for Dementia (95% CI: 1.22 to 2.87) both per µg/m3 of PM2.5.