Air Pollution and Morbidity: A Comprehensive Assessment

Abstract

As developing countries continue to industrialize and the effects of climate change begin to manifest, air pollution continues to be the greatest environmental risk factor to health and research in this area continues to be essential. However, accurate measurement of air pollution on an individual level is difficult. Recent developments in computer science have enabled us to utilize machine learning algorithms in creating spatio-temporal models which estimate pollution levels on a fine scale. In this work, we incorporated three different machine learners and an ensemble approach to estimate PM2.5 levels in Greater London from 2005 to 2013 on a one km2 scale. Our model showed strong overall and temporal predictive ability, and moderate spatial predictive ability. This model can be further utilized in epidemiology studies, risk assessments, and in the development of other spatio-temporal models. Studies looking at air pollution epidemiology have been prolific over the past two decades. However, this research has tended to focus on fatal outcomes and on the effects of short-term exposure. In this dissertation, we look at the non-fatal outcomes of long-term exposure to air pollution using causal methodology. Both of our studies found pollutants such as fine particulate matter (PM2.5), nitrogen dioxide (NO2), and ozone (O3) to be harmful for cardiovascular and respiratory health, with the greatest increase in risk occurring at lower concentrations. This dissertation demonstrates the need for continued research into the health risks of air pollution and the need for greater investment in accurate exposure measurement.