AI has shown promising capabilities in healthcare by demonstrating its potential to stabilize patients in complex clinical scenarios. A recent study conducted by the Mack Institute, involving researchers Christian Terwiesch, Lennart Meincke, and Arnd Huchzermeier, explored the effectiveness of the multimodal large language model Gemini Pro 2.5 in a medical training simulation known as BodyInteract.
In this simulation, the AI managed a virtual patient's evolving condition, which included dynamic vital signs and delayed test outcomes. Unlike traditional models limited to isolated tasks, this AI was responsible for the entire clinical workflow, including questioning patients, ordering tests, and administering treatments under time pressure. The study assessed the AI's performance across four acute care cases, comparing it to over 14,000 simulation runs involving medical students and an experienced emergency physician.
Results indicated that the AI was able to stabilize patients and complete cases at rates similar to, and in some scenarios even better than, medical students. Notably, the AI accomplished these tasks with similar diagnostic accuracy while operating at a significantly faster pace, highlighting its potential to enhance clinical decision-making in real-time patient care.