Revolutionizing Inpatient Fall Prevention with AI Technology
When it comes to preventing patient falls in hospitals, clinical care teams often rely on traditional tools such as bed alarms, gait belts, and closer nursing station placement. However, these interventions can sometimes disrupt clinical workflows and lead to alert fatigue, making them less effective than intended.
To address these challenges and ensure that high-risk patients receive the necessary precautions while reducing clinician alarm fatigue, UCHealth in Aurora, Colorado has developed an innovative user interface. This interface leverages artificial intelligence to deliver specific fall prevention interventions based on each patient’s unique risk profile.
The tool analyzes mobility data, behavioral health indicators, and other risk factors to predict the likelihood of inpatient falls with injury. By integrating these predictions directly into the Epic electronic health records system using cloud tools, UCHealth aims to provide clinicians with clear recommendations for patient care.
Creating the Model and Setting Objectives
A multidisciplinary team at UCHealth conducted a thorough literature review to identify 12 risk domains and 92 potential variables related to fall risk. By mapping these variables to Epic data elements and analyzing data from over 181,000 inpatient admissions, the team developed a simplified logistic regression model that runs every four hours.
Instead of a simple high/low risk indicator, the model classifies patients into three tiers: Highest Risk, Elevated Risk, and Universal Risk. This classification system generates a list of patient-specific precautions to help keep individuals safe during their hospital stay.
Challenges of Integration into Clinical Workflows
Integrating the AI model into clinical workflows presented several challenges for the UCHealth team. They needed to define clear, tiered precautions for each risk level to guide staff on appropriate actions. Additionally, the team focused on meaningful display of risk, data availability in the first 12 hours, user trust and adoption, and workflow integration.
- Meaningful display of risk: Clinicians needed to understand why a patient was flagged as high risk and which precautions were recommended.
- Data availability in the first 12 hours: The model required sufficient clinical data entry, and staff had to exercise clinical judgment in the initial hours of a patient’s admission.
- User trust and adoption: Building confidence in AI recommendations among nurses and care team members was crucial for successful implementation.
- Workflow integration: The model’s outputs and recommended precautions were seamlessly integrated into existing Epic flowsheets and care plans to streamline the process for clinicians.
Insights and Lessons Learned
The session at HIMSS25 will offer valuable insights into using AI to reduce fall injury risks, including strategies for model adoption, performance measurement, addressing disparities among patient populations, and continuous improvement. Attendees can expect practical takeaways on deploying, monitoring, and refining AI tools for fall prevention, with an emphasis on organizational engagement and nurse involvement.
If you’re interested in learning more about UCHealth’s innovative approach to fall prevention using AI technology, don’t miss Drew and Cyriacks’ session, “Fall Injury Risk Model: From AI to Clinical Interventions,” scheduled for Tuesday, March 4 at 3:15 p.m. at HIMSS25.
