You are here:
Evaluation of the Health Impacts of the 1990 Clean Air Act Amendments Using Causal Inference and Machine Learning
Citation:
Nethery, R., F. Mealli, J. Sacks, AND F. Dominici. Evaluation of the Health Impacts of the 1990 Clean Air Act Amendments Using Causal Inference and Machine Learning. JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION. Taylor & Francis Group, London, Uk, , 1-12, (2020). https://doi.org/10.1080/01621459.2020.1803883
Impact/Purpose:
Causal inference methodology to examine the potential impact of large-scale air pollution regulations.
Description:
We develop a causal inference framework to estimate the number of adverse health events prevented by a large-scale air quality regulation via changes in exposure to multiple pollutants. This approach is designed to analyze a regulation that brought about impacts on pollution in all areas within its purview, and it requires the following data at high spatial resolution: (1) estimates of pollution exposures in the scenario of regulation implementation (i.e., factual pollution exposure estimates), (2) estimates of pollution exposures in the scenario of no regulation (i.e., counterfactual pollution exposure estimates), and (3) observed population-level health outcome data for the areas and years under study. We introduce a causal estimand called the Total Events Avoided (TEA) by the regulation, defined as the difference in the expected number of health events under the counterfactual pollution exposures and the observed number of health events under the factual pollution exposures. We propose a matching method and a Bayesian machine learning method for estimation of the TEA and associated uncertainties. These methods use confounder-adjusted relationships between factual pollution exposures and health outcomes to inform estimation of the expected health outcomes under the counterfactual pollution exposures. We also discuss the asymptotic properties of our matching estimator. Our approach improves upon existing methods for health impact analyses of large air quality regulations by clarifying the statistical quantity being estimated and the causal identifying assumptions, leveraging population-level health data, minimizing parametric assumptions, and investigating the impacts of changes in all pollutants simultaneously rather than separately. However, in order to minimize parametric assumptions and extrapolation, our methods exclude some areas from the analysis, meaning that our estimates should provide a conservative but data-driven picture of the health impacts of regulations. This is an important complement to existing parametric approaches to regulation health impact analyses. In simulations, we ?nd that both the matching and machine learning methods perform favorably in comparison to standard parametric approaches. We apply these methods to investigate the health impacts of the 1990 Clean Air Act Amendments (CAAA) in the US Medicare population in the year 2000. We ?nd evidence that large numbers of cardiovascular and dementia hospitalizations were avoided thanks to the CAAA; however, we find limited evidence of reductions in mortality with this analysis.
URLs/Downloads:
DOI: Evaluation of the Health Impacts of the 1990 Clean Air Act Amendments Using Causal Inference and Machine Learning