Prediction or Prevention? Nurse Interactions with an Electronic Early Warning System for Fall Risk

Area of Science:

• Nursing informatics
• Clinical decision support systems
• Machine learning in healthcare

Background:

• Machine learning (ML) models are increasingly integrated into electronic health records for risk assessment and adverse outcome prediction.
• Limited research exists on how ML technology impacts nursing workflows, nurse behaviors, and patient outcomes.

Purpose of the Study:

• To explore nurse interactions with ML-generated fall alerts.
• To examine the relationship between alerts and patient falls within the nursing workflow.

Main Methods:

• Retrospective analysis of data from four medical/surgical units.
• Examination of nurse interactions with interruptive alerts from a predictive fall model.
• Chronological correlation of alerts with actual patient falls.

Main Results:

• 87.0% of admissions generated at least one fall alert; 1.5% resulted in a fall.
• Most alerts (57.3%) were snoozed; 22.0% were seen by non-primary nurses.
• 89.3% of falls were preceded by any staff seeing an alert, but only 38.7% by the primary nurse.

Conclusions:

• Primary nurses were often not exposed to alerts preceding patient falls.
• Alert dismissal via "Snooze to Review" was common.
• Further research is required to optimize ML technology for nursing workflows and patient safety.