ATA 2023 - Identifying Opportunities For Integrating AI/Computer Vision Into Clinical Workflows
I was excited to present at the American Telemedicine Association 2023 Annual Meeting. I showed a poster that came out of our partnership between LookDeep Health and Duke University Health. Our poster was entitled: “Identifying Opportunities For Integrating AI/Computer Vision Generated Clinical Observations Into Clinical Workflows And Decision-Making,” [View Poster]
This work builds on our initial research in AI computer vision in the Duke medical intensive care unit (ATS 2022). This research demonstrated several compelling data-driven readouts of patient activity and room environment, highlighting its potential utility in sleep, sedation, agitation, and delirium. The research also resulted in approaches for auto-generated video summaries of the overnight period (MLHC’22).
While promising, we want to characterize clinical workflows to better understand opportunities for the targeted use of AI/CV. To that end, we performed an ethnographic observational study of nursing shift change on the 8E Duke stepdown unit. Using a structured form with 16 items (see above), 10 shift changes were observed (five 7AM shift changes and five 7PM shift changes). The study was determined to meet the quality improvement exemption by the Institutional Review Board.
The key result is shown below. There is a wide range in how often key items are talked about. Not surprisingly, medical condition and length of stay was discussed in all observed shift changes. Patient activity/mobility was discussed 90% of the time, suggesting that AI/CV can provide quantitative data to augment discussion of this key dimension of patient care.
At the other end of the spectrum (and somewhat surprisingly), sleep/rest was discussed 50% of the time, and hospital-acquired injury (e.g. falls, pressure injury) was discussed 30% or the time. For these critical dimensions of patient care that are not discussed as often as expected, AI/CV data may serve as a quantitative prompt to reduce barriers to discussion.
As LookDeep’s partnership with Duke moves forward, we look forward to our medical stepdown unit pilot deployment and integrating our AI/CV observations into clinical care.
Duke authors: LaDonna Shore, Armando Bedoya, Deborah “Hutch” Allen
LookDeep Health authors: Michael Choma, Narinder Singh