Exposures to Mixtures of Air Pollutants: Analysis of Biological, Personal and Area Monitoring.

Abstract

The objective of this dissertation is to understand the nature and significance of exposures to volatile organic compounds (VOCs) by characterizing exposure distributions, trends, mixtures, and determinants. VOC data were drawn from two datasets: the Relationship between Indoor, Outdoor and Personal Air study (RIOPA), the National Health and Nutrition Examination Survey (NHANES). The RIOPA study collected outdoor, indoor and personal VOCs twice in three U.S. cities. NHANES, nationally representative samples, collected both blood and personal VOCs. To estimate extreme exposures, generalized extreme value (GEV) distributions were fitted to the top 5 and 10% of VOCs. Simulated extreme value datasets were compared to observations. VOC trends in 1988 - 2004 were evaluated using linear quantile regression models at three quantiles. VOC mixtures were identified using positive matrix factorization (PMF) and by toxicological mode of action. Dependency structures of mixture components were examined using copulas. Results from copulas and multivariate lognormal models were compared to observations. Exposure determinants were identified using linear mixed-effect models. Extreme value exposures fitted the GEV distributions. Lognormal distributions significantly underestimated the likelihood of extrema. In NHANES, most VOCs showed decreasing trends at all quantiles. Trends varied by VOC and quantile, and were grouped into three patterns: similar decreases at all quantiles, most rapid decreases at upper quantiles, and fastest declines at central quantiles. Four VOC mixtures were identified by PMF (gasoline vapor, vehicle exhaust, chlorinated solvents and disinfection by-products, and cleaning products and odorants). Three mixtures were identified by toxicological mode of action (hematopoietic, liver and renal tumors). The dependency structures of the VOC mixtures fitted Gumbel and t copulas. The copulas accurately reproduced risk predictions, and performed better than multivariate lognormal distributions. The analysis of VOC determinants showed that exposures were affected by indoor concentrations, city, and some personal activities, household characteristics and meteorological factors. Exposure data feature extreme values, temporal changes, dependency structures, and other complex characteristics. Advanced statistical methods can improve estimates exposures and risks, and are needed to develop control and management guidelines and policies. These results extend our understanding of and ability to model VOC exposures.