Using machine learning models to predict falls in hospitalised adults.

Area of Science:

• Healthcare Informatics
• Clinical Risk Management
• Machine Learning in Healthcare

Background:

• Effective fall prevention programs require accurate identification of high-risk patients.
• Integrating health information systems presents challenges in analyzing data for patient safety.

Purpose of the Study:

• To identify factors associated with patient falls.
• To develop high-performance prediction tools for at-risk patients in Australian acute and sub-acute care settings.

Main Methods:

• Retrospective study of over 670,000 patients in Victoria, Australia (2019-2021).
• Utilized data from multiple sources including electronic health records.
• Applied machine learning models (Random Forest, Deep Neural Network) to predict falls and identify risk factors.

Main Results:

• Both Deep Neural Network and Random Forest models demonstrated high accuracy in predicting patient falls.
• Identified 12 common risk factors across patient, staffing, and admission categories.
• Models achieved high accuracy (e.g., 0.989 for RF) and specificity (e.g., 1.000 for RF).

Conclusions:

• Machine learning effectively predicts patient falls and identifies critical risk factors.
• Findings highlight the potential of AI in enhancing patient safety and informing fall prevention interventions.
• Further validation in diverse settings is recommended for wider implementation.