Initial Experience With AI-based Video Monitoring for Delirium-Associated Factors and Room Environment

Abstract: Introduction


Delirium is highly prevalent and a major focus of consultation-liaison psychiatry. The symptoms of delirium characteristically fluctuate, yet patient assessments are intermittent. There are limited ways to continuously monitor the room environment, despite its role in the development and treatment of delirium [1]. Here, we present our initial experience with an AI-based video method for continuous monitoring of patient activity and room environment for patients at high-risk for developing delirium. We discuss the findings in the context of a patient who was thought to exhibit evidence of long-term cognitive decline but through AI video monitoring, instead exhibited symptoms often associated with delirium.


The presented work here comes from an ongoing study at OHSU for patients admitted after a trauma. Patients between 65 and 90 were eligible with exclusion criteria of an existing diagnosis of delirium, major neurocognitive disorder, or major affective/psychotic disorders. Once enrolled, a custom video cart with color and night vision cameras was deployed to the patient’s room. Video data were continuously streamed to an encrypted hard drive for up to 7 days. AI analysis and image blurring was then performed on the video, quantifying patient and bedside activity, and number of people in the room. Data streams were used to generate visual and graphical overnight reports.


Mini-Case Report


An 80-year-old female was admitted after a motor vehicle collision with rib, spinal, and hip fractures. No head trauma was noted. She remained on the trauma service for non-operative management of these fractures. Recording started on day 2 of her admission. AI-assisted video actigraphic analysis showed significant nighttime patient activity with rest-wake inversion, demonstrated by video summaries generated of the nighttime period as well as by motion “heat maps” that visually display patient activity over a given time interval. AI analysis also revealed significant bedside activity through the daytime and nighttime on most days, consistent with frequent staff visits seen on video summaries. Information was presented via a visual overnight report, with summary images, heatmaps, and video of the busiest 5-minutes in each hour, enabling rapid review of the overnight period.


Conclusion and Next Steps


Initial experiences with AI video monitoring of patient activity and room environment highlighted its potential in the care of patients at-risk for delirium. AI-generated insight suggested manifestations of acute delirium that may have been misinterpreted as worsening overall cognition. Continuous monitoring may help shift from a dichotomous paradigm of delirious/not delirious to a paradigm that recognizes that it exists on a continuum.



References
1. Kang J, Lee M, Ko H, et al. Effect of nonpharmacological interventions for the prevention of delirium in the intensive care unit: A systematic review and meta-analysis. J Crit Care. 2018;48:372-384. doi:10.1016/j.jcrc.2018.09.032