Machine Learning Risk Predictions at OR-ICU Handoffs

The Opportunity

Machine learning risk prediction (MLRP) is widely recognized for its potential to decrease medical errors and failure-to-rescue, which is an urgent need as medical errors contribute to nearly a third of deaths. However, provider uptake is slow, as simple scoring systems are often preferred. Scalable interventions that address this need include innovations in machine learning display and further explanation of machine learning to improve provider adoption and understanding. The goal of this project was to reduce errors during OR and ICU handoff by improving adoption and understanding of machine learning by providers.

Our Approach

Using machine learning prediction models to assess postoperative risks to patients, the team incorporated the risk information into nursing handoff reports using an iterative, user-centeric approach to design the report. Simulation encounters and think-aloud sessions were used to test prototypes, and the report design was tested to assess its effect on clinicians’ mental model of patient status.

The Solution

Statistically significant improvements were reported in time to analgesia dosing, and there were fewer information omissions, fewer technical errors, and greater information-sharing scores for the handoff interventions group.

Additional outcomes are not complete for this project, so please check back at a later date for more information.