A Flexible Spatio-temporal Model for Air Pollution Cohort Studies

Air pollution cohort studies rely on spatial contrasts in exposure for inference about health impacts.  Thus some cohort studies conduct special-purpose, time-limited pollutant monitoring campaigns that provide spatially rich pollutant data.  This supplements pollutant data from regulatory monitoring networks which are temporally rich but limited to a few locations.  Many cohort studies therefore face monitoring designs in which pollutant data are unbalanced in space and time.  We have developed a spatio-temporal modelling framework that combines data from several different monitoring networks with geographic covariates in order to predict ambient air pollution at subject homes.  This work is part of the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air) study, a prospective cohort study funded by the US EPA to investigate the relationship between chronic exposure to air pollution and cardiovascular disease.  I will describe the model and highlight its ability to produce high quality predictions at small spatial scales.  I will illustrate its success in predicting long-term average concentrations for NOx and several PM_2.5 components while taking full advantage of the unbalanced monitoring data available to the MESA Air study.  I will note the nuances necessary in cross-validation model assessment to quantify the spatial (rather than spatio-temporal) predictive ability of the model.  This work has been implemented in an R-package, SpatioTemporal. 

Acknowledgements:  This study was supported by funding from EPA (CR-834077101, RD831697), with additional funding from the Health Effect Institute (4749-RFA05-1A/06-10), NIEHS (P50 ES015915), and STINT (IG2005-2047).  The views expressed are those of the author and do not necessarily reflect the views or policies of the U.S. Environmental Protection Agency.