Evaluating Large Language Models as Decision Support Agents in Pan-Cancer Tumor Boards: A Randomized Controlled Trial
Multidisciplinary teams (MDTs) represent the gold standard for personalized tumor treatment, but they are limited by medical resources and accessibility Limitation. Although large language models (LLMs) have shown promise in medical reasoning, their multidisciplinary practicality in pan-cancer MDTs has not been fully explored. In the early stage of this project, LLMs with high clinical application efficacy were identified through benchmark tests, and an open-label randomized controlled study (RCT) was conducted based on these LLMs. The research aims to explore whether AI-assisted assistance can enhance the accuracy and writing efficiency of MDT diagnosis and treatment reports. This study intends to prospectively collect the diagnosis and treatment information of 20 patients and MDT diagnosis and treatment information. It is planned to recruit 40 junior doctors. Doctors in the intervention group will use LLM to assist in the writing of MDT reports, while doctors in the control group will use traditional information retrieval methods for the writing of MDT reports. Three clinical experts ultimately used a standardized Likert scale to conduct comprehensive and multidisciplinary scoring of the MDT reports of the intervention group and the control group. This study quantitatively compared the diagnosis and treatment quality and efficiency of the MDT AI-assisted model and the traditional model to verify the application potential of large language models in assisting tumor diagnosis and treatment.
• A junior doctor with a practicing physician qualification certificate.
• Oncologists, surgeons, radiation oncologists, radiologists and pathologists with 3 to 5 years of clinical experience.
• Age: 25 to 33 years old, gender not limited.
• During the research period, one can participate for no less than 10 hours.
• Agree to participate in this research and sign the informed consent form.