Ambient Air Pollution and Atherosclerosis: The Framingham Heart Study

Abstract

This dissertation assesses the associations of residential proximity to a major roadway and residential ambient fine particulate matter exposure with artery calcification, a correlate of atherosclerosis, among Framingham Heart Study participants. It also explores using quantile regression to model outcomes with distributions similar to that of artery calcification, which is zero-inflated and right-skewed, an excess of zeros followed by a long tail to the right. Chapter I examines associations of residential distance to a major roadway and residential fine particulate matter (PM2.5) exposure with coronary artery calcium Agatston score (CAC) among Framingham Heart Study participants living in the Northeastern U.S. CAC is a quantitative estimate of total coronary atheroma and a strong risk factor for coronary heart disease. We observed no associations of these exposures with the presence, extent or progression of CAC. Chapter II examines associations of residential distance to a major roadway and residential PM2.5 exposure with thoracic aortic calcium Agatston score (TAC) and abdominal aortic calcium Agatston score (AAC), correlates of systemic atherosclerosis. There were no associations of these exposures with the presence or extent of TAC or AAC or with AAC progression. The direction of some estimates ran counter to expectation. Overall, these exposures were not strongly related to artery calcification among participants living in a region with relatively low levels of and little variation in PM2.5. Chapter III assesses quantile regression performance when modeling outcomes with zero-inflated or right-skewed distributions, such as CAC. Quantile regression has the potential to identify associations of exposures with shifts in parts of the distribution of an outcome that linear regression may miss. When analyzing data in Chapters I and II, we found quantile regression performed poorly when modeling CAC. In an effort to better understand this performance, we examined quantile regression for simulated outcomes that had varying amounts of zero-inflation and skew. Quantile regression performed poorly for some scenarios in which there was zero-inflation, suggesting traditional multivariable quantile regression may not work well for zero-inflated outcomes. When using quantile regression in practice, users should check model performance by assessing shapes of covariate-outcome relationships and looking for unreasonable model predictions.