Estimating the Causal Effect of Lowering Particulate Matter Levels Below the National Ambient Air Quality Standards on Health Outcomes

• Activity Number: 375
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Maggie Makar* and Joseph Antonelli and Qian Di and Joel Schwartz and David Cutler and Francesca Dominici
• Companies: MIT and Harvard T.H. Chan School of Public Health and Harvard T.H. Chan School of Public Health and Harvard T.H. Chan School of Public Health and Harvard and Harvard T.H. Chan School of Public Health
• Keywords: Causal inference ; statistical methods ; epidemiology ; Marginal structural models ; air pollution ; human health

Abstract

We will present methods for causal inference incorporating model uncertainty to estimate the health benefits of compliance with the NAAQS standard for PM2.5 (12 ug/m3). By linking data from Medicare claims, the Medicare Current Beneficiaries Survey (MCBS), remote sensing data, monitor level pollution data, and others, we created a retrospective study of 68,789 interviews, spanning 5138 zip codes. This study has an unprecedented number of confounders and high accuracy in the spatial resolution of PM2.5 exposure over a large time and geographical span. We assigned to every individual a binary zip code level exposure denoting levels higher and lower than 12 ug/m3. We considered the following outcomes: death, all-cause hospitalization, hospitalization for circulatory diseases, and hospitalization for respiratory diseases within a year of the patients' MCBS interview. To adjust for measured confounding we gathered information on more than 100 individual level confounders capturing smoking, body mass index, behavioral and socio-economic factors. We will develop methods for causal inference to adjust for this high dimensional set of potential confounders in a marginal structural model.