You are here

Robust statistical approaches to understanding the causal effect of air pollution mixtures

Principal Investigator: 

University of Florida

This New Investigator Award study seeks to develop statistical methodology that allows for complex relationships between air pollution and health outcomes to be used to estimate causal effects of multivariate exposures. Additionally, the proposed methodology will allow for evaluation of separate subgroups in the population to identify the most vulnerable subgroups.

Funded under

Poster abstract for HEI Annual Conference 2023

Understanding the Impact of Mobility on the Analysis of Air Pollution Mixtures

Heejun Shin1, Joseph Antonelli1

1University of Florida Department of Statistics, Gainesville, FL, USA

Background. In environmental epidemiology, exposure to air pollution is typically assigned based on an individual’s home address or geographic region, despite the fact that individuals move to different regions with different exposure levels due to daily mobility. In this study, we assess the potential bias that can occur due to mobility into neighboring geographic areas, and evaluate whether this can be corrected with cell phone mobility data.

Methods. We first formalize the above problem and show that it can be treated as an issue of interference from the causal inference literature, where the exposures of some geographic regions may affect outcomes in other regions due to travel of individuals across regions. We aim to derive expressions for the bias that one would obtain if this mobility was ignored in order to better understand the magnitude and direction of this bias in practical situations. Lastly, we develop estimation approaches that incorporate aggregate level cell phone mobility data to provide improved estimates of the health effects of the air pollution mixture.

Results. Bias formulas from ignoring mobility into neighboring regions are derived for a number of relevant estimands of interest. It is shown that in many realistic settings this bias will be towards the null, though this need not be the case in general. Additionally, we are able to show that correcting for this bias by treating this problem within an interference framework has benefits over an analogous measurement error correction that also incorporates cell phone mobility data. We develop novel estimation strategies to estimating the health effects of the air pollution mixture that utilize flexible nonparametric Bayesian methodology, which allows for nonlinear and interactive effects of the different air pollutants in the mixture. Utilizing cell phone mobility data, we show that air pollution exposure from day-to-day mobility can be quite different than exposure at home geographic areas, and therefore this issue should be addressed whenever possible.

Conclusions. Health effect estimates of the air pollution mixture can be negatively impacted by mobility of individuals into regions with different exposure levels. These issues can be alleviated if data on the movement of individuals in the population is available and is incorporated when estimating the effects of air pollutants.