Towards Bridging Generalists to Subspecialists With Large Language Models

What this Trial Is About

Researchers are evaluating whether large language models (LLMs) can improve clinical decision-making among general cardiologists managing patients with cardiovascular genetic cardiomyopathies. The study compares the use of LLMs against traditional decision support tools to see if AI assistance enhances clinical reasoning and management quality. This research addresses the challenge of limited subspecialist availability in cardiology, aiming to improve triage and referral accuracy from generalists to subspecialists. General cardiologists will be randomly assigned to two groups: one with access to a large language model and one without. Both groups will manage real patient cases from a subspecialty cardiovascular genetic cardiomyopathy clinic. Each case will be reviewed by two cardiologists, one using the LLM and one relying on usual resources such as UpToDate or Dynamed. Their responses to management questions will be evaluated using a standardized rubric developed by expert subspecialists. Participants will complete their case assessments, which will then be graded by blinded subspecialty cardiologists within one month. The study will also gather participant perspectives on using the large language model shortly after assessment. This process allows researchers to measure the impact of AI assistance on clinical decision quality and gather feedback on the tool's usability. The study begins in January 2025 and is sponsored by Stanford University.

Large Language Models To Improve the Quality of Care of Cardiology Patients