A Spatio-Temporal Approach for Estimating Chronic Effects of Air Pollution

Estimating the health risks associated with air pollution exposure is of great importance in public health. In air pollution epidemiology, two study designs have been used mainly. Time series studies estimate acute risk associated with shortterm exposure. They compare day-to-day variation of pollution concentrations and mortality rates, and have been criticized for potential confounding by timevarying covariates. Cohort studies estimate chronic effects associated with longterm exposure. They compare long-term average pollution concentrations and time-to-death across cities, and have been criticized for potential confounding by individual risk factors or city-level characteristics. We propose a new study design and a statistical model, which use spatiotemporal information to estimate the long-term effects of air pollution exposure on life expectancy. Our approach avoids confounding by time-varying covariates and individual or city-level risk factors. By estimating the increase in life expectancy due to decreases in long-term air pollution concentrations, it provides easily interpretable values for public policy purposes. We develop a suitable backfitting algorithm that permits efficient fitting of our model to large spatio-temporal data sets. We evaluate spatio-temporal correlation in the data and obtain appropriate standard errors. We apply our methods to the Medicare Cohort Air Pollution Funding was provided by the U.S. Environmental Protection Agency (EPA) (grants RD-83241701 and RD-83362201). Although the research described in this article has been funded wholly or in part by the U.S. EPA, it has not been subjected to the agency’s required peer and policy review and does not necessarily reflect the views of the agency, and no official endorsement should be inferred. 1 Hosted by The Berkeley Electronic Press Study, including data on fine particulate matter (PM2.5) and mortality for 18.2 million Medicare enrollees from 814 locations in the U.S. during an average of 65 months in 2000-2006. Supplemental material including R code implementing our methods is provided in a web appendix.