What does this research mean for the field?

The risk of rural hospital closures in the U.S. is primarily driven by accumulated long-term financial and operational instability, policy context, and community sociodemographics rather than short-term distress, and can be accurately predicted using explainable machine learning models. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The study aims to identify risk factors contributing to rural hospital closures in the U.S. using explainable machine learning techniques.

June 1, 2026Open Access

Why are rural hospitals closing in the U.S.? Predictors identified using explainable machine learning

Key Points

The study aims to identify risk factors contributing to rural hospital closures in the U.S. using explainable machine learning techniques.
Conducted a retrospective longitudinal study of rural hospitals from 2011 to 2022 using national datasets.
Integrated various data types including financial, operational, and community characteristics.
Utilized an XGBoost classifier with stratified three-fold cross-validation for model training and explainability assessment via SHAP.
Out of 2,683 hospitals, 139 (5.2%) closed during the study period.
Model achieved an accuracy of 75%, recall of 86%, and AUC of 89%.
Key predictors of closure risk included sustained financial weakness, operational instability, and community sociodemographic factors.

Abstract

OBJECTIVE: Rural hospital closures in the U.S. reduce access to essential healthcare services and worsen health and economic outcomes in rural communities. This study uses national longitudinal data and explainable machine learning (XML) to identify and interpret risk factors of rural hospital closures. MATERIALS AND METHODS: We conducted a retrospective longitudinal study of U.S. rural hospitals using national datasets from 2011 to 2022, integrating hospital-level financial, operational, policy, population, and community data. Longitudinal feature engineering and statistical summarization were performed, class imbalance and multicollinearity were addressed using random undersampling and variance inflation factor (VIF) analysis, and an XGBoost classifier was trained using stratified three-fold cross-validation and Optuna for hyperparameter tuning. Model explainability was assessed using SHapley Additive exPlanations (SHAP). RESULTS: Among 2,683 rural hospitals, 139 (5.2%) closed during the study period. The model achieved a recall of 86%, accuracy of 75%, and an AUC of 89%. The key variables linked to closure risk are sustained financial weakness and volatility (e.g., total margin, variability in return on equity, and current ratio), operational instability (low patient volume and variability in bed-days available), policy context (e.g., years since Medicaid expansion), and community context, including sociodemographic characteristics of the service area (e.g., proportion of Black or African American residents), which may reflect underlying structural and socioeconomic conditions rather than direct predictors or causal drivers of closure. DISCUSSION: These findings indicate that the risk of closure of rural hospitals reflects accumulated long-term financial and operational instability, as well as policy and community context, rather than short-term distress, highlighting the value of longitudinal and explainable modeling approaches. CONCLUSION: Longitudinal data integrated into XML provides a practical approach for early warning systems that enable proactive monitoring and timely interventions, particularly for hospitals serving historically underserved communities. CLINICAL TRIAL NUMBER: Not applicable.

Mark Helpful

Bookmark

Relay

View Full Paper