Preventing inpatient falls with injuries using integrative machine learning prediction: a cohort study

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Patient Safety Research

Background:

• Hospitalized patient falls are a significant safety concern, leading to adverse outcomes and increased costs.
• Current fall risk assessment methods lack the ability to predict injury severity.
• There is a need for advanced predictive models to identify patients at risk of severe fall-related injuries.

Purpose of the Study:

• To develop and validate a machine learning classifier to predict the severity of injuries resulting from inpatient falls.
• To integrate multi-source patient data for enhanced predictive accuracy.
• To provide a tool for early identification of patients at high risk for severe fall injuries.

Main Methods:

• A retrospective cohort study utilizing data from over two thousand inpatient falls.
• Development of a machine learning classifier using multi-view ensemble learning with missing data imputation (MELMV).
• Integration of demographic characteristics, diagnoses, procedural data, and bone density measurements.

Main Results:

• The MELMV classifier demonstrated superior performance compared to three baseline models.
• Achieved a cross-validated AUC of 0.713 and an AUC of 0.808 on the separate testing set.
• The model effectively predicts the severity of inpatient falls, providing a severe fall index.

Conclusions:

• Integrative machine learning models can effectively handle multi-source patient data for robust fall severity prediction.
• The MELMV classifier offers a promising approach to identify high-risk patients for severe fall injuries.
• This predictive capability can guide targeted interventions to prevent severe harm from falls, enhancing standard care.