A longitudinal machine-learning approach to predicting nursing home closures in the U.S

Area of Science:

• Healthcare Management
• Gerontology
• Data Science in Healthcare

Background:

• Nursing home closures disrupt access to essential skilled nursing and long-term care services.
• Previous research predominantly used descriptive analyses, lacking predictive capabilities for identifying facilities at risk of closure.

Purpose of the Study:

• To evaluate two distinct modeling paradigms for predicting nursing home closures in the United States.
• To identify influential factors contributing to nursing home closures using explainable AI methods.

Main Methods:

• Utilized longitudinal data spanning 2011-2021 for U.S. nursing homes.
• Implemented a feature-aggregated paradigm summarizing facility characteristics over time.
• Developed and compared a longitudinal paradigm incorporating time-varying features, alongside SHapley Additive exPlanations (SHAP) for feature interpretation.

Main Results:

• The longitudinal modeling approach outperformed the feature-aggregated paradigm.
• The best-performing model achieved an Area Under the Precision-Recall Curve (AUPRC) of 0.50, a Recall of 0.77, and an F1-score of 0.56.
• SHAP analysis identified key features and their impact on closure prediction, offering insights into facility vulnerabilities.

Conclusions:

• The longitudinal approach provides a robust framework for predicting nursing home closures.
• Findings support the development of proactive strategies to support at-risk facilities and ensure continuity of care.
• Targeted interventions can help maintain access to care, particularly in underserved regions.