A Workflow Time Study to Estimate Savings from the Implementation of a Machine Learning Algorithm to Automate Fall Risk Assessment

• Activity Number: 149 - Statistical Learning for Decision Support
• Type: Contributed
• Date/Time: Monday, August 8, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Gregg M Gascon* and Erika Braun and Victoria Zigmont and Amy Tracy and Michelle Murray and Shalunda Tyler and Jason Kaufman and Stephanie Risteff and Eleanora Vassileva and Bradley Waller and J. Michael Kramer
• Companies: OhioHealth and OhioHealth and OhioHealth and OhioHealth and OhioHealth and OhioHealth and OhioHealth and OhioHealth (retired) and OhioHealth and The Ohio State University and OhioHealth
• Keywords: Machine learning; External validation; Workflow study; Time and motion; UX research; Task automation

Abstract

The Epic Inpatient Risk of Falls (EIRF) machine learning model was designed to automate inpatient fall risk assessment. The EIRF uses ordinal logistic regression to categorize patients into ranked risk categories using the levels of 38 variables. A fall risk score is produced every 4 hours which queues provider use of fall interventions based on risk levels. At a large midwestern health system, the EIRF was selected to replace a manual assessment. An external validation was performed. Two thresholds were chosen to create three risk levels. The model was put into production but not made visible to providers. To determine time saved through automation, researchers developed and implemented a workflow study to produce time motion data through external observation of a purposive sample of units and facilities. The model was turned on for a pilot study of mobility that used the automated fall assessment model to ensure that no subsequent increase in the rate of falls was observed. The fall rate on the pilot units fell, and a significant amount of time was liberated through fall assessment automation.