Development of a Spatio-Temporal Model of Air Pollution Source Apportioned Factors in a Multi-Ethnic Study of Atherosclerosis (MESA) Region

摘要

BACKGROUND. The National Particle Component Toxicity (NPACT) Study aims to investigate the association between participants' exposure to PM2.5 air pollutant components and health effects. Spatio-temporal models were developed for NPACT to characterize PM2.5 components. Here we extend this analysis using positive matrix factorization (PMF) to derive multivariate pollutant features in each MESA area. AIM. To develop a spatio-temporal model based on PMF factor scores derived from fine particle species measurements. METHODS. PMF models were developed from sampling observations collected from multiple sites for each MESA clinic area. Here we present data from the Baltimore area. The PMF model had 8 multivariate features. We chose 2 of these features for further analysis: diesel exhaust/brake wear and secondary sulfate. We then built spatio-temporal models using a large suite of geographic covariates and geostatistical smoothing methods to predict these features. RESULTS. Within PM2.5, 7% was attributed to diesel/brake wear and 36% to secondary sulfate. The secondary sulfate factor showed a stronger temporal pattern whereas the diesel/brake wear factor gave a stronger spatial pattern. The estimated temporal trend of secondary sulfate was higher in summer, lower in winter and generally homogeneous across the area. For the diesel/brake wear factor, 72% of variability of the computed long-term means was explained by geographic variables related to road network, elevation, and land use. CONCLUSIONS. These results are consistent with our hypotheses of expected spatial and temporal trends of diesel exhaust and secondary sulfate features. This suggests that further application of factorization approaches may enhance our understanding of temporal and spatial characteristics of sources in epidemiological research. This serves as the first step in developing factor-based spatio-temporal model predictions of exposure to air pollution sources for the entire MESA cohort.