The application of large language models (LLMs) in medicine has received increasing attention, showing significant potential in teaching, research, and clinical practice, especially in knowledge extraction, management, and understanding. However, the use of LLMs in Traditional Chinese Medicine (TCM) has not been thoroughly studied. This study aims to provide a comprehensive overview of the status and challenges of LLM applications in TCM. A systematic search of five electronic databases and Google Scholar was conducted between November 2022 and April 2024, using the Arksey and O'Malley five-stage framework to identify relevant studies. Data from eligible studies were comprehensively extracted and organized to describe LLM applications in TCM and assess their performance accuracy. A total of 29 studies were identified: 24 peer-reviewed articles, 1 review, and 4 preprints. Two core application areas were found: the extraction, management, and understanding of TCM knowledge, and assisted diagnosis and treatment. LLMs developed specifically for TCM achieved 70% accuracy in the TCM Practitioner Exam, while general-purpose Chinese LLMs achieved 60% accuracy. Common international LLMs did not pass the exam. Models like EpidemicCHAT and MedChatZH, trained on customized TCM corpora, outperformed general LLMs in TCM consultation. Despite their potential, LLMs in TCM face challenges such as data quality and security issues, the specificity and complexity of TCM data, and the nonquantitative nature of TCM diagnosis and treatment. Future efforts should focus on interdisciplinary talent cultivation, enhanced data standardization and protection, and exploring LLM potential in multimodal interaction and intelligent diagnosis and treatment.
Read full abstract