Machine learning models predicting inpatient falls.

Area of Science:

• Healthcare Informatics
• Clinical Prediction Models
• Patient Safety

Background:

• Inpatient falls pose a significant risk to patient safety and can lead to injuries.
• The Hester-Davis score (HD), a common tool for assessing fall risk, demonstrates limited predictive accuracy.
• There is a critical need for more precise methods to predict inpatient falls.

Purpose of the Study:

• To develop and evaluate dynamic machine learning models for predicting inpatient fall risk.
• To compare the performance of machine learning models against the traditional Hester-Davis score.
• To identify key predictors of inpatient falls using comprehensive patient data.

Main Methods:

• A retrospective analysis of 46,695 patients from 17 hospitals (January 2018 - July 2022).
• Development of four dynamic machine learning models incorporating Hester-Davis variables, socio-demographics, comorbidities, physiological measures, medications, and time-series data.
• Models were updated at 8- and 24-hour intervals.

Main Results:

• The Extreme Gradient Boosting (EGB) model achieved a superior Area Under the Curve (AUC) of 0.87 (95% CI 0.86-0.88).
• The Hester-Davis score showed significantly lower AUCs: 0.57 (95% CI 0.56-0.58) and 0.62 (95% CI 0.59-0.61).
• Key predictors identified include neurological conditions, behavioral abnormalities, oxygen saturation, heart rate, and IV furosemide use.

Conclusions:

• Machine learning models, particularly EGB, offer enhanced accuracy in predicting inpatient falls compared to the Hester-Davis score.
• Dynamic models integrating diverse patient data show promise for improving clinical fall risk assessment.
• Further prospective validation is necessary for clinical implementation of these advanced prediction tools.