Estimation of Monthly Average Ambient Air Pollutant Concentrations Using Geographic Information Systems

Abstract

P-057 Introduction: The epidemiologic literature provides evidence that air pollutant exposures may increase risks of several pregnancy-related health outcomes. The literature also suggests that short-term exposures before or during pregnancy may be more relevant than long-term (e.g.[r], annual) exposures. Thus, we sought a method by which individual short-term air pollutant exposures could be estimated among 1,699 participants of a cohort study of pregnancy. Methods: We developed models to estimate ambient concentrations of carbon monoxide (CO), particulate matter ≤ 10 um in diameter (PM10), and particulate matter ≤ 2.5 um in diameter (PM2.5) within an urban United States region. Using geographic information systems software and publicly available data, we measured traffic, weather, and population characteristics at regional air pollutant monitoring stations. We used multivariable linear models to relate these local characteristics to monthly CO, PM10, and PM2.5 concentrations measured at the stations between 1996 and 2002. By applying the coefficients obtained from these models to local characteristics measured at participants' homes, we estimated residential monthly CO, PM10, and PM2.5 concentrations. We used a split-sample reliability assessment to quantify the agreement between the concentrations measured at the monitoring stations and the estimates obtained by the exposure models. Results: Estimates from the CO and PM2.5 models correlated well with concentrations measured at monitoring stations. The Pearson correlation coefficients (rho) were 0.74 and 0.56, respectively. Root mean square errors for the CO and PM2.5 models were 0.27 ppm and 2.9 ug/m3, respectively. We were unable to develop a satisfactorily predictive PM10 model (rho=0.06). The mean (standard deviation, range) estimated CO and PM2.5 concentrations at participants' residences during the the calendar month of conception were 0.95 ppm (0.30, 0.23–1.97 ppm) and 10.9 ug/m3 (3.5, 2.0–22.2 ug/m3), respectively. Discussion: Using easily accessible data sources and information from a regional air pollutant monitoring network, we were able to estimate monthly CO and PM2.5 concentrations at residential locations within the region. The strong correlations between estimates and measured concentrations suggest that this method can be used to estimate monthly ambient air pollutant exposures in epidemiologic studies.