A Spatiotemporal Causal Machine Learning Framework for Quantifying the Impact of PM_2.5 on Asthma at the Community Level

Air pollution, particularly fine particulate matter with an aerodynamic diameter of 2.5 μm or smaller (PM_2.5), is a leading environmental risk factor for respiratory diseases. While previous studies have investigated the biological mechanisms linking PM_2.5 to asthma exacerbation, these studies are unable to scale for population-level assessments. In contrast, most regional analyses rely on traditional statistical models that primarily quantify associations, limiting estimation of health changes under hypothetical pollution interventions. This study introduces a spatiotemporal causal machine learning framework to estimate PM_2.5-related changes in asthma outcomes at the county level. Using county-level monthly hospitalization and emergency department records from 2011 to 2019 in Colorado, the framework integrates spatial and temporal representation learning to model geographic interdependence and temporal dynamics in exposure-response relationships. Further, through a causal inference approach, the model estimates model-based counterfactual health outcomes under hypothetical PM_2.5 interventions, moving beyond regression-based correlation analysis. Results show that the model-based counterfactual estimates exhibit spatial and population heterogeneity. Larger estimated effects were observed in several communities experiencing compounded environmental and social stressors, while uncertainty was greater in data-sparse counties.