You are here:
A simulation study to determine the attenuation and bias in health risk estimates due to exposure measurement error in bi-pollutant models
Citation:
Dionisio, K., L. Baxter, AND H. Chang. A simulation study to determine the attenuation and bias in health risk estimates due to exposure measurement error in bi-pollutant models. Presented at ISES 2014 Conference, Cincinnati, OH, October 12 - 16, 2014.
Impact/Purpose:
The National Exposure Research Laboratory (NERL) Human Exposure and Atmospheric Sciences Division (HEASD) conducts research in support of EPA mission to protect human health and the environment. HEASD research program supports Goal 1 (Clean Air) and Goal 4 (Healthy People) of EPA strategic plan. More specifically, our division conducts research to characterize the movement of pollutants from the source to contact with humans. Our multidisciplinary research program produces Methods, Measurements, and Models to identify relationships between and characterize processes that link source emissions, environmental concentrations, human exposures, and target-tissue dose. The impact of these tools is improved regulatory programs and policies for EPA.
Description:
To understand the combined health effects of exposure to ambient air pollutant mixtures, it is becoming more common to include multiple pollutants in epidemiologic models. However, the complex spatial and temporal pattern of ambient pollutant concentrations and related exposures introduces the potential for exposure measurement error that is dependent on the exposure metric used in the epidemiologic analysis. This exposure measurement error can lead to effect attenuation and reduced statistical power in the epidemiologic model results. Using central site measurements, air quality model estimates, and population exposure model estimates as exposure metrics for CO, NOx, PM2.5, EC, SO4, and O3, for 193 ZIP codes in the Atlanta, GA metropolitan area, we calculated empirically determined measurement error and variance/covariance parameters for multiple pollutant pairs. Results show that pollutant concentrations and their exposure errors are moderately to highly correlated (typically >0.5), especially for CO, NOx, and EC. In addition, spatial variability exists, and variance of exposure error ranges from 0.25-0.83 for these same pollutants, indicating that attenuation of model coefficients differs across type of exposure error, pollutant, and space. We use the empirically determined parameters and actual health data in a classical exposure measurement error framework accounting for both additive and multiplicative bias to conduct a simulation study to determine the degree of effect attenuation in health risk estimates. Results from the simulation study indicate significant possible attenuation of model coefficients in a bi-pollutant model.