AI-driven fall risk prediction in inpatients: Development, validation, and comparative evaluation

Abstract

Background & Aim: Falls among hospitalized patients pose severe consequences, necessitating accurate risk prediction. Traditional assessment tools rely on cross-sectional data and lack dynamic analysis, limiting clinical applicability. This study developed an AI-based fall risk prediction model using supervised learning techniques to enhance predictive accuracy and clinical integration.
Methods & Materials: This study was conducted at a medical center in Taiwan, excluding pediatric patients due to non-disease-related fall factors. Fall cases were obtained from hospital records, and non-fall cases were stratified based on age and gender to create a balanced 1:1 dataset.
A total of 52 predictive variables were identified and refined to 39 through expert review. The AI model was compared with MORSE, STRATIFY, and HII-FRM using supervised learning with 10-fold cross-validation. Performance was evaluated based on accuracy, sensitivity, and specificity.
Results: The results demonstrated that the AI-based model significantly outperformed traditional fall risk assessment tools in accuracy, sensitivity, and specificity. More importantly, the model’s superior predictive power allows for real-time risk assessment and seamless integration into clinical decision support systems. This integration can enable timely interventions, optimize patient safety protocols, and ultimately reduce fall-related incidents in hospitalized settings.
Conclusion: By automating risk assessment, the AI model can alleviate the workload of healthcare professionals, reducing the time required for manual evaluations and minimizing subjective biases in clinical decision-making. This not only enhances operational efficiency but also allows nursing staff to allocate more time to direct patient care. These findings underscore the transformative potential of AI-driven approaches in healthcare, improving patient safety through data-driven.