Multiple Pollutant Methods for Air Pollution and Health Research

Investigators interested in relationships between airborne pollutants and human health have long faced challenges in estimating the effects of individual pollutants and their interactions in complex multiple-pollutant atmospheres. Correlations among pollutant concentrations over time and geographic space, possible interactions among pollutants, and differences in the availability of data for individual pollutants pose substantial statistical challenges. Furthermore, these efforts may be complicated by confounding or effect-modifying covariates (such as indicators of socioeconomic status) that are often correlated with health outcomes and pollution exposures.

In this session, presenters will describe their ongoing work in developing statistical methods for multiple-pollutant data analysis. Their approaches either directly incorporate health information in estimating the magnitude of health risks associated with exposures, or have been developed for use in epidemiologic studies of varying spatial and temporal design.

Session Chair: Kathryn Adams, Health Effects Institute

Source-specific exposure assessment by using Bayesian spatial multivariate receptor models for spatially correlated multi-pollutant data
Eun Sug Park, Texas A&M University

Examining spatially varying effects of multiple-pollutant air pollution exposures on birth outcomes using conditional autoregressive (CAR) models
John Molitor, Oregon State University

Dimension reduction for spatially misaligned multi-pollutant exposure data: Predictive sparse principal component analysis and predictive k-means clustering
Adam Szpiro, University of Washington

Robust distributed lag models for analyzing health effects due to multiple pollutants: a shrinkage approach
Bhramar Mukherjee, University of Michigan