Curving Emissions: Comparing Methods for Evaluating Exposure-Response Curves

• Activity Number: 337 - Environmental Epidemiology and Analysis of Large Database
• Type: Contributed
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Michael Cork* and Francesca Dominici and Daniel Mork
• Companies: Harvard University and Harvard and Harvard University
• Keywords: environmental epidemiology ; air pollution; causal inference; pm2.5

Abstract

Determining the causal relationship between a continuously varying exposure and outcome is a crucial scientific endeavor with significant policy implications. While there is broad consensus on air pollution as a hazard, the shape of the exposure response curve (ERC) remains uncertain. In this paper, we evaluate the performance of three general classes of epidemiological methods for assessing ERC via simulation. These classes consists of regression methods that use nonparametric penalized spline models, parametric functions that impose thresholds or are used for quantifying disease burden, and novel causal inference methods. For each approach, we evaluate these methods with several criteria that are influential in creating policy relevant information. Given the uncertainty and noise present in epidemiological data, simulation studies with plausible ERC relationships subjected to current epidemiological methods could elucidate potential biases and provide guidance on the appropriate conditions for application. Finally, we look to use each proposed method to estimate the effect of long term PM2.5 exposure on all-cause mortality using a Medicare enrollee cohort from 2000 to 2016.