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

Nursing home closures affect access to skilled nursing and long-term care. Prior research on nursing home closures focused largely on cross-sectional or descriptive analyses, few studies have leveraged predictive or forecasting approaches to identify facilities at risk of closure. In this study, we evaluate two paradigms for predicting nursing home closures in the U.S. Leveraging longitudinal data on nursing homes from 2011 to 2021, we first implement a feature-aggregated paradigm that summarizes facility characteristics over time. Second, we implement a longitudinal paradigm incorporating time-varying features. The longitudinal approach produced the best performing model, with an AUPRC of 0.50, Recall of 0.77, and an F1-score of 0.56. SHapley Additive exPlanations (SHAP) were used to identify influential features and their directional associations. The findings in this paper offer a promising path toward proactive approaches that could be targeted at facilities at risk of closure to ensure care access, particularly in underserved areas.