Code Blue for Rural America: Predicting Hospital Closures with Machine Learning

• In Event: The Role of Big Data and Machine Learning in Health Care: Challenges and Opportunities

Abstract

Rural hospital closures have accelerated over the past two decades in the United States, disrupting access to care, weakening local economies, and exacerbating health disparities. Despite growing policy and public concern, the factors that predict these closures remain poorly understood. This study is the first to apply machine learning techniques to predict rural hospital closures across the U.S. Using a comprehensive dataset from the RAND Hospital Data (2000–2025), the Center for Healthcare Quality and Payment Reform, and the Area Resource Files, I developed high-dimensional predictive models that incorporate hospital-level financial and operational indicators, county-level workforce and demographic characteristics, and state-level policy environments. To address class imbalance resulting from the relatively low number of closures, the analysis employed upsampling to enhance model performance. I compared four supervised learning algorithms: logistic regression, support vector machines, random forest, and extreme gradient boosting (XGBoost). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). The best-performing model—XGBoost—achieved an AUC of 0.883, followed by random forest (AUC 0.839). In addition to predictive accuracy, the study highlighted key features associated with hospital vulnerability. These findings demonstrate the value of machine learning tools in identifying at-risk rural hospitals and informing timely policy interventions to prevent further closures.

Author

• KUN HUANG, University at Albany
  Presenting Author