A Spatiotemporal Causal Machine Learning framework for quantifying the Impact of PM2.5 on Asthma at the Community level

This repository contains the cleaned project files needed to run Run_Main_Raw.py, along with the associated public input datasets used by the workflow. Paper Description A Spatiotemporal Causal Machine Learning Framework for Quantifying the Impact of PM2.5 on Asthma at the Community Level Jing Li¹, Xuantong Wang², Hong Wei Chu³ Department of Geography and the Environment, University of Denver, Denver, Colorado, USA; [e-mail] Department of Geosciences, Texas Tech University, Lubbock, Texas, USA; [e-mail] National Jewish Health, Denver, Colorado, USA; [e-mail] Corresponding author: Jing Li ([e-mail]) Key Points Developed a spatiotemporal cross-graph CEVAE model to estimate counterfactual asthma outcomes under reduced PM2.5. Identified heterogeneous causal effects of PM2.5 on asthma that extend beyond traditional regression-based analyses. Generated causal maps revealing where PM2.5 reductions yield the greatest health gains, helping guide equitable air-quality policy. Project Description This project develops a spatiotemporal causal machine learning framework to quantify the impact of fine particulate matter (PM2.5) on asthma outcomes at the county level. Using monthly hospitalization and emergency department data from Colorado (2011–2019), the model integrates spatial and temporal dynamics to capture how pollution exposure varies across regions and over time. Unlike traditional statistical approaches that focus on correlation, this framework applies causal inference to estimate counterfactual scenarios, such as how asthma hospitalizations would change under reduced PM2.5 levels. Results reveal significant spatial heterogeneity, with the greatest health benefits occurring in communities facing overlapping environmental and social stressors. By identifying where pollution reduction would have the largest impact, this work provides a scalable, data-driven tool to inform targeted and equitable air quality and public health interventions. Data Sources All datasets used in this study are publicly available from public data repositories. Asthma hospitalization and emergency department data (2004–2023): Colorado Department of Public Health and Environment (CDPHE), Colorado Health Information Dataset (COHID). Source: https://cdphe.colorado.gov/cohid Air quality (PM2.5, NOx, O3, SO2): U.S. Environmental Protection Agency (EPA) Air Quality System (AQS). Source: https://www.epa.gov/aqs Meteorology: PRISM Climate Group monthly climate datasets used for county-level weather covariates. Source: https://prism.oregonstate.edu Sociodemographic indicators: CDPHE Public Health Data Portal and county-level American Community Survey derivatives. Sources: https://data-cdphe.opendata.arcgis.com and https://data-cdphe.opendata.arcgis.com/datasets/CDPHE::cdphe-organized-american-community-survey-estimates-2017-2021-by-county/about Environmental justice indicators: Colorado EnviroScreen v2 county-level screening indicators. Source: https://cdphe.colorado.gov/enviroscreen Repository Contents Included code: Code/Run_Main_Raw.py Code/DataLoader_Combine.py Code/Model_STCrossGraph_Causal.py Code/Analysis_Causal_Validation.py Included data: Only the data files referenced by the shared code Asthma, air quality, weather, demographic, environmental justice, policy, boundary, and imputation inputs used by the pipeline