Evaluating Alternative Exposure Metrics Used for Multipollutant Air Quality and Human Health Studies

Epidemiologic studies of air pollution have traditionally relied upon surrogates of personal pollutant exposures, such as ambient concentration measurements from fixed-site pollutant monitors. This study evaluates the performance of alternative measured and modeled exposure metrics for multiple particulate and gaseous pollutants, in the context of different epidemiologic studies performed by EPA, Rutgers/Rochester/LBNL and Emory/Georgia Tech researchers. Alternative exposure estimation approaches used, included: central site or interpolated monitoring data, regional pollution levels based on measurements or models (CMAQ) and local scale (AERMOD) air quality models, hybrid models, statistically blended modeling and measurement data, concentrations adjusted by home infiltration rates based on LBNL algorithms, and population human exposure (SHEDS and APEX) model predictions. The Emory/Georgia Tech team examined the acute morbidity effects of ambient traffic-related pollutants (CO, NOx, PM2.5 and PM2.5 EC) and ozone using time series analyses of emergency department (ED) visits and case-crossover analysis of implantable cardioverter defibrillator (ICD) detected ventricular arrhythmias in Atlanta, GA. The Rutgers/Rochester/LBL team examined the associations between PM2.5 mass and its species with myocardial infarction (case-crossover study) and adverse birth outcomes (cohort study) in New Jersey. Initially, the various exposure indicators/metrics were compared in terms of their ability to characterize the spatial and temporal variations of multiple ambient air pollutants across the different study areas. These metrics were then used to examine associations between ambient air pollution and adverse health effects. Next, pollutant-specific relative risks (RRs) obtained from epidemiologic analyses of the alternative exposure metrics were evaluated against those obtained from using a conventional approach (i.e., central site data alone). Pollutant and metric dependent exposure prediction differences were found in some cases, indicating a non-uniform exposure prediction error structure across pollutants. Results suggest the need for additional refinements to methods used to estimate exposures in support of different types of air pollution epidemiologic studies.