Stanford AI helps clinicians follow ICU patients' activity across a whole hospital stay
How a patient moves and behaves can say a lot about how they are doing, but that picture is scattered across shifts and records. On the Stanford AI Lab blog, a team including Fei-Fei Li describes TRACER, a training-free system that lets language and vision language models reason over bedside video from Stanford Hospital ICU sensors together with electronic health records. Nurses and clinicians helped choose 15 basic actions for it to track, and it offers a visual timeline plus a chat interface for questions about a stay. Early feedback from clinicians was encouraging. Across 280 patients, the team also noticed fewer purposeful actions among those with delirium, a hint at future digital health markers.