TraumaTIQ: Online Decision Support for Trauma Management

• Authors: Abigail S. Gertner;Bonnie L. Webber
• Venue: IEEE Intelligent Systems
• Year: 1998

Imagine a patient in shock and losing blood rapidly with a gunshot wound in the abdomen arriving at a hospital. A relatively inexperienced resident surgeon decides to do a CT (computerized tomography) scan of the abdomen to find the source of the bleeding and then take the patient to the operating room. The attending physician intervenes and recommends an abdominal X ray-an effective and faster procedure. She also suggests that the resident x-ray the chest to verify that the bullet did not travel upward and cause chest-cavity injuries. What enables an advisor to provide this kind of assistance? To evaluate and address misconceptions in a physicians plan, an advisor musthave a model of the physicians beliefs and goals, understand the problem and decide on the best solution, be able to explain her reasoning, and communicate in a way that will influence the physicians future actions.This article describes an approach for providing online decision support in complex, task-oriented situations. As the above example suggests, we have applied this approach to the management of multiple trauma-a task that typically involves reasoning about multiple goals, integrating diagnosis and treatment into a single management plan, efficiently allocating resources, and acting under time pressure. While this task can benefit from intelligent decision support, we also note that the tasks heavy cognitive demands and user reluctance to accept advice from computer systems inhibit such support. To minimize these obstacles, our interface provides concise, relevant, and user-focused critiques. This interface, TraumaTIQ, comprises a plan recognizer, a plan evaluator, and a language generator. (We present TraumaTIQs plan recognition and evaluation modules in detail elsewhere.1 ) In this article, we show how the three modules work together to produce the final critique.