SBIR Phase I: Artificial Intelligence for Competency-Based Medical Training

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to train physicians efficiently, assure high-quality patient care, and provide the United States with a robustly competent physician workforce. Current assessment practices require attending physicians and surgeons to review tens-to-hundreds of data points, removing them from clinical activities.

Integrating a machine learning model in an existing resident assessment system to predict performance can address trainees’ learning needs and identify excelling, competent, and struggling residents months earlier. This is vital to patient care: earlier identification of trainee performance can benefit patient care faster than the current human-based, semiannual process.

Improved tracking and documentation of competence may be of interest to multiple stakeholders including patients, hospitals, third-party payers such as insurance companies or the Centers for Medicare and Medicaid Services, and the residency accreditation entity, the Accreditation Council for Graduate Medical Education. Improved, automated assessment models using existing trainee data help training programs facing increasing documentation burden, as well as hospitals and third-party payers interested in reducing adverse health events for the patients they serve.

This Small Business Innovation Research (SBIR) Phase I project will integrate an artificial intelligence model to support resident physician training programs in customizing training based on individual learners’ needs. Starting with plastic and reconstructive surgery and one of the four training programs in the United States engaged in time-variable training is an efficient way to create, test, and assess the model’s efficacy.

The created machine learning model will be assessed for its predictive ability at different points during resident physicians’ training and compared with attending physicians’ assessments of trainees’ skills. Such models make time-variable training feasible enabling adaptive, needs-based scheduling of various educational rotations. This has the added advantages of keeping residents fully engaged in their training and returning faculty physicians to clinical care faster, improving job satisfaction and reducing risk of burnout.

Ultimately, time-variable training and use of their associated machine learning models will reduce the direct and indirect costs of graduate medical education; accelerate the entry of new, fully competent physicians into the workforce; and retain valuable physician educators in the workforce.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.