The Potential of AI in Nursing Care: Multicenter Evaluation in Fall Risk Assessment.

Area of Science:

• Geriatric Medicine
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Falls are a leading cause of injury-related death in older adults, with limited staff hindering timely prevention.
• Artificial intelligence (AI) offers potential to enhance fall risk assessment and resource allocation in nursing care.
• Existing AI studies often use limited, single-institution data, impacting generalizability and fairness assessments.

Purpose of the Study:

• To empirically evaluate AI's potential in nursing for fall risk prediction using large, heterogeneous datasets.
• To analyze AI model performance and safety across different hospital settings (university and geriatric).
• To assess the fairness of AI models across demographic groups, specifically sex and age.

Main Methods:

• Utilized two large datasets from a university hospital (931,726 participants) and a geriatric hospital (12,773 participants).
• Trained state-of-the-art AI models using separate training, retraining, and federated learning (FL) approaches.
• Compared AI model performance against existing rule-based clinical systems and conducted fairness analyses.

Main Results:

• AI models consistently outperformed rule-based systems in fall risk prediction across both datasets.
• Federated learning (FL) did not enhance prediction performance in this specific study context.
• Fairness analysis revealed equitable performance across sex groups, but significant disparities were found across age groups.

Conclusions:

• AI models demonstrate superior performance over traditional methods for fall risk prediction in diverse clinical settings.
• Challenges in generalizing AI models due to demographic shifts and data imbalances were identified.
• Future development must address data imbalances and ensure broader demographic representation for fair and generalizable AI tools.