Predicting asthma prevalence across US communities by integrating social vulnerability and environmental risk with machine learning

Thinh H Nguyen; Saleh Alsulami; Teresia T Pham; Diane Story; Marissa Hauptman; Wanda Phipatanakul

doi:10.1016/j.jaci.2026.04.015

Predicting asthma prevalence across US communities by integrating social vulnerability and environmental risk with machine learning

J Allergy Clin Immunol. 2026 Apr 29:S0091-6749(26)00297-6. doi: 10.1016/j.jaci.2026.04.015. Online ahead of print.

Authors

Thinh H Nguyen¹, Saleh Alsulami¹, Teresia T Pham², Diane Story³, Marissa Hauptman⁴, Wanda Phipatanakul⁵

Affiliations

¹ Division of Immunology, Boston Children's Hospital, Harvard Medical School, Boston, Mass.
² Department of Pediatrics, Tufts School of Medicine, Boston, Mass; Atrius Health, Medford, Mass.
³ Pediatric Environmental Health Center, Division of General Pediatrics, Boston Children's Hospital, Boston, Mass.
⁴ Pediatric Environmental Health Center, Division of General Pediatrics, Boston Children's Hospital, Boston, Mass; Region 1 New England Pediatric Environmental Health Specialty Unit, Boston, Mass; Department of Pediatrics, Harvard Medical School, Boston, Mass.
⁵ Division of Immunology, Boston Children's Hospital, Harvard Medical School, Boston, Mass. Electronic address: wanda.phipatanakul@childrens.harvard.edu.

Abstract

Background: Asthma is one of the most common chronic conditions in the United States, disproportionately affecting racial and ethnic minority populations and low-income communities. Although both social and environmental factors are known contributors to asthma prevalence, the extent to which they interact to shape geographic disparities remains insufficiently understood.

Objective: This study aimed to explore how these factors vary across regions with differing asthma prevalence and thereby identify patterns that may inform targeted public health responses.

Methods: We used the US Centers for Disease Control and Prevention's 2022 Environmental Justice Index, which includes standardized social and environmental measures at the census tract level. K-means clustering identified geographic clusters with varying asthma prevalence based on asthma rates and related health outcomes. A random forest model was trained to predict asthma rates by using social and environmental variables, and feature importance was analyzed.

Results: We identified 9 clusters, with 3 showing the highest asthma prevalence. High-prevalence areas had greater social vulnerability, including higher poverty rate, minority population rate, limited English proficiency, unemployment, and housing cost burden. They also experienced greater environmental exposures, such as elevated level of fine particulate matter with an aerodynamic diameter of 2.5 μm or less, ozone, diesel pollution, more land near hazardous waste sites, more chemical facilities, and older housing. The random forest model predicted asthma rates with strong predictive performance (R² = 0.81). Socioeconomic variables consistently ranked as stronger predictors than environmental exposures.

Conclusions: Geographic areas with high asthma prevalence are shaped by overlapping social and environmental vulnerabilities. Socioeconomic disparities emerged as particularly influential. These findings support integrated, data-driven public health strategies to address asthma inequities.

Keywords: Asthma prevalence; air pollution; census tract; environmental exposures; environmental justice; health disparities; machine learning; social determinants of health.

Abstract

Grants and funding