Estimating pollution-attributable mortality at the regional and global scales: challenges in uncertainty estimation and causal inference

Significance Exposure to outdoor concentrations of fine particulate matter is considered a leading global health concern, largely based on estimates of excess deaths using information integrating exposure and risk from several particle sources (outdoor and indoor air pollution and passive/active smoking). Such integration requires strong assumptions about equal toxicity per total inhaled dose. We relax these assumptions to build risk models examining exposure and risk information restricted to cohort studies of outdoor air pollution, now covering much of the global concentration range. Our estimates are severalfold larger than previous calculations, suggesting that outdoor particulate air pollution is an even more important population health risk factor than previously thought.

Although many studies have linked elevations in tropospheric ozone to adverse health outcomes, the effect of long-term exposure to ozone on air pollution-related mortality remains uncertain. We examined the potential contribution of exposure to ozone to the risk of death from cardiopulmonary causes and specifically to death from respiratory causes. Data from the study cohort of the American Cancer Society Cancer Prevention Study II were correlated with air-pollution data from 96 metropolitan statistical areas in the United States. Data were analyzed from 448,850 subjects, with 118,777 deaths in an 18-year follow-up period. Data on daily maximum ozone concentrations were obtained from April 1 to September 30 for the years 1977 through 2000. Data on concentrations of fine particulate matter (particles that are < or = 2.5 microm in aerodynamic diameter [PM(2.5)]) were obtained for the years 1999 and 2000. Associations between ozone concentrations and the risk of death were evaluated with the use of standard and multilevel Cox regression models. In single-pollutant models, increased concentrations of either PM(2.5) or ozone were significantly associated with an increased risk of death from cardiopulmonary causes. In two-pollutant models, PM(2.5) was associated with the risk of death from cardiovascular causes, whereas ozone was associated with the risk of death from respiratory causes. The estimated relative risk of death from respiratory causes that was associated with an increment in ozone concentration of 10 ppb was 1.040 (95% confidence interval, 1.010 to 1.067). The association of ozone with the risk of death from respiratory causes was insensitive to adjustment for confounders and to the type of statistical model used. In this large study, we were not able to detect an effect of ozone on the risk of death from cardiovascular causes when the concentration of PM(2.5) was taken into account. We did, however, demonstrate a significant increase in the risk of death from respiratory causes in association with an increase in ozone concentration. 2009 Massachusetts Medical Society

A number of models have been developed to estimate PM2.5 exposure, including satellite-based aerosol optical depth (AOD) models, land-use regression, or chemical transport model simulation, all with both strengths and weaknesses. Variables like normalized difference vegetation index (NDVI), surface reflectance, absorbing aerosol index, and meteoroidal fields are also informative about PM2.5 concentrations. Our objective is to establish a hybrid model which incorporates multiple approaches and input variables to improve model performance. To account for complex atmospheric mechanisms, we used a neural network for its capacity to model nonlinearity and interactions. We used convolutional layers, which aggregate neighboring information, into a neural network to account for spatial and temporal autocorrelation. We trained the neural network for the continental United States from 2000 to 2012 and tested it with left out monitors. Ten-fold cross-validation revealed a good model performance with a total R(2) of 0.84 on the left out monitors. Regional R(2) could be even higher for the Eastern and Central United States. Model performance was still good at low PM2.5 concentrations. Then, we used the trained neural network to make daily predictions of PM2.5 at 1 km × 1 km grid cells. This model allows epidemiologists to access PM2.5 exposure in both the short-term and the long-term.