Machine-learning-based Fall-prediction Model for Inpatients in Military Hospitals

This case-control study develops a machine-learning (ML)-based fall-risk-prediction model for military-hospital inpatients, using electronic records (January 2020 to December 2023) from 11 military hospitals in South Korea. The dataset includes 1470 fall and 3478 nonfall cases, and 25 predictive factors are selected based on data completeness and previous studies. ML models, including logistic regression, decision trees, random forest (RF), and gradient boosting, are developed and evaluated using 10-fold cross-validation. The predictive performances of these models are compared with that of the Morse fall scale (MFS). RF exhibits superior performance, with accuracy, recall, F1-score, and area under the curve (AUC) of 0.76, 0.77, 0.76, and 0.83, respectively. The MFS exhibits a poor predictive capability of 0.02. The key predictive variables include the admission method (on foot), medical department (orthopedics), MFS score, activity status, and body mass index. The findings reveal the limitations of MFS and higher accuracies of ML models in predicting falls among young military inpatients, highlighting the suitability of ML models for improving fall-prevention strategies. Future research should integrate real-time patient conditions and key fall-related factors into prediction models and embed these into clinical decision-making systems.