Consistent associations between air pollution and adverse health outcomes from epidemiological studies have played a major role in setting regulatory standards and protecting public health. For exposure assessment, population-based studies routinely utilize air quality measurements from outdoor monitoring network due to its public availability. However the monitoring network has limited spatial coverage and ambient concentrations may not reflect human exposure to air pollution from outdoor sources since individuals spend the majority of their time indoors. Therefore exposure uncertainty can arise from unobserved spatial variation in air pollution concentration, as well as spatial variations in population characteristics that contribute to differential exposure (e.g. age and residential housing type). The overarching goal of this project is to develop and apply innovative statistical methods for improving exposure assessment and quantifying exposure uncertainties in air pollution and health studies. By incorporating additional data sources to supplement ambient monitor measurements, we will (1) increase the spatial coverage and resolution of air quality data; (2) examine the spatial relationship between ambient concentration and human exposure; and (3) systematically evaluate the impacts of exposure measurement errors. Approach:
In Aim 1, by combining monitoring measurements and remotely sensed satellite image data, we will develop a spatio-temporal data fusion approach to predict daily concentrations of particulate matter less than 2.5 m in aerodynamic diameter (PM2.5). Our model allows for distinct relationships between monitoring and remotely sensed data at different spatial resolutions, while addressing the missing data problem associated with satellite images.
In Aim 2, we will develop a Bayesian spatial hierarchical model to estimate daily population exposure to ambient PM2.5 as a function of ambient concentrations and human daily activity patterns. This is accomplished by utilizing data from stochastic exposure simulators that reflect state-of-the-art human exposure science. The statistical model overcomes the computational effort associated with traditional human exposure simulation experiments and serves as an emulator for imputing spatially-resolved exposures that can be readily used in health studies.
In Aim 3, we will conduct an epidemiological time series analysis to estimate short-term associations between daily PM2.5 exposure and emergency department visits in the 20-county Atlanta metropolitan. Area: We will examine the robustness of the exposure-response function using different exposure metrics, and assess the impacts of exposure measurement error and ecological bias. Expected Outcomes: The research addresses three well-recognized sources of exposure error in air pollution epidemiology that arise from: (1) spatial variation in ambient concentration, (2) spatial variation in exposure-concentration relationship, and (3) spatial aggregation of health outcomes. Improved exposure assessment is a crucial step towards increasing the accuracy and relevance of health study results. The proposed approaches can be applied to different air pollutants and study designs, which may be extended to our other ongoing health effect studies and health impacts analyses.
This research furthers our understanding of the magnitude and impact of exposure measurement error in air pollution and health studies. Development of statistical approaches to account for exposure uncertainty will contribute to the reliability, relevance, and reproducibility of large population-based health studies, which have played an important role in risk assessment and guiding policies towards protecting public health.
Showing the most recent 10 out of 11 publications
