In previous work, we have developed, deployed and evaluated a novel intelligent medical training system for pathologic diagnosis and reporting. SlideTutor is an individualized, adaptive, simulation environment that provides explanations and assistance specific to each student's needs. Our results show that the system produces dramatic improvements in diagnostic and reporting accuracy. On average, students achieve a 400% gain in performance, and retain these skills over time. A unique aspect of our system is that it dynamically models student skills, knowledge and misconceptions, so that it can adapt its interventions. In the process, the SlideTutor system captures an enormous amount of information about the intermediate steps in the physician reasoning. We have developed an elaborate research infrastructure that allows us to make use of this abundant information as research data. With this novel research infrastructure, we are poised to make far more general statements about how to create adaptive and individualized systems for training physicians. We now propose to leverage the SlideTutor infrastructure to focus on three foundational areas where there is nearly no previous research in medical domains: metacognition, performance prediction, and learning behaviors. Work in these areas has the potential to deeply impact the fields of patient safety, medical simulation and competency- based assessment, in addition to guiding the development of future medical training systems.
