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

Area of Science:

• Medical Informatics
• Gerontology
• Public Health

Background:

• Falls are a significant risk for hospitalized patients, leading to increased morbidity and healthcare costs.
• Traditional fall risk assessment tools, like the Morse Fall Scale (MFS), may have limitations in accurately predicting falls in specific populations.
• Developing robust, data-driven fall prediction models is crucial for enhancing patient safety in healthcare settings.

Purpose of the Study:

• To develop and evaluate machine learning (ML)-based models for predicting fall risk in military hospital inpatients.
• To compare the predictive performance of various ML models against the established Morse Fall Scale (MFS).
• To identify key factors contributing to fall risk in this specific patient cohort.

Main Methods:

• A case-control study utilizing electronic health records from 11 South Korean military hospitals (January 2020 - December 2023).
• Development and evaluation of ML models including logistic regression, decision trees, random forest (RF), and gradient boosting using 10-fold cross-validation.
• Comparison of ML model performance against the Morse Fall Scale (MFS) using metrics like accuracy, recall, F1-score, and AUC.

Main Results:

• The Random Forest (RF) model demonstrated superior predictive performance with an accuracy of 0.76, recall of 0.77, F1-score of 0.76, and AUC of 0.83.
• The Morse Fall Scale (MFS) showed poor predictive capability with an AUC of 0.02.
• Key predictors identified include admission method (on foot), medical department (orthopedics), MFS score, activity status, and body mass index.

Conclusions:

• Machine learning models, particularly Random Forest, offer significantly higher accuracy in predicting falls among military inpatients compared to the Morse Fall Scale.
• The study highlights the limitations of the MFS in this population and the potential of ML for improving fall prevention strategies.
• Future research should focus on integrating real-time patient data and clinical decision support systems for enhanced fall risk management.