Multi-Pollutant Measurement Error in Air Pollution Epidemiology Studies Arising from Predicting Exposures with Penalized Regression Splines

• Activity Number: 656
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Silas Bergen*+ and Adam Szpiro
• Companies: University of Washington and University of Washington
• Keywords: Measurement error ; Air pollution epidemiology ; Penalized regression splines ; Spatial modeling

Abstract

Air pollution epidemiology studies are trending toward a multi-pollutant approach, as health outcomes are more realistically influenced by a mixture of pollutants rather than any single pollutant. In these studies the relevant exposures at subject locations are often unobserved and must be predicted with spatial models built using observed exposures at misaligned monitoring locations. Estimating health effects with these predicted rather than true exposures induces measurement error that can bias the estimated health effects and impact their standard errors. We characterize the measurement error when the exposure models are penalized regression splines, developing an asymptotic bias expression that can be used for bias correction. Our simulations show that, contrary to conventional wisdom that downward bias of a poorly-measured pollutant's coefficient will translate to upward bias in the coefficient of a better-measured pollutant, the estimated health effects can be simultaneously biased upward or downward. Our bias correction can be combined with a simple non-parametric bootstrap to attain correct coverage of 95% confidence intervals.