Establishing a Classification System for High Fall-Risk Among Inpatients Using Support Vector Machines.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Patient Safety

Background:

• Hospital falls are a significant patient safety concern.
• Existing fall risk assessment methods can be burdensome for healthcare staff.
• Predictive modeling offers a potential solution for proactive fall prevention.

Purpose of the Study:

• To develop and validate a predictive model for daily inpatient falls.
• To utilize routinely collected electronic health record (EHR) data for fall risk assessment.
• To create a fall risk model that minimizes additional data collection burden.

Main Methods:

• A support vector machine (SVM) model was constructed.
• A dataset of approximately 1.2 million patient-days was created using fall reports and nursing care needs data.
• A multistep grid search optimized model parameters, with sensitivity and specificity used for evaluation.

Main Results:

• The optimized SVM model achieved a sensitivity of 64.9% and a specificity of 69.6%.
• The model effectively predicted fall risk based on the previous day's patient status.
• The approach leverages existing EHR data, avoiding new data collection.

Conclusions:

• A data-driven, machine learning model can effectively predict daily inpatient fall risk.
• Utilizing EHR data streamlines fall risk assessment, reducing the burden on nursing staff.
• This predictive model supports proactive interventions to enhance patient safety and reduce falls.