Evaluating and Improving Syndrome Differentiation Thinking Ability in Large Language Models: Method Development Study

Abstract

Background: A large language model (LLM) provides new opportunities to advance the intelligent development of traditional Chinese medicine (TCM). Syndrome differentiation thinking is an essential part of TCM and equipping LLMs with this capability represents a crucial step toward more effective clinical applications of TCM. However, given the complexity of TCM syndrome differentiation thinking, acquiring this ability is a considerable challenge for the model.

Objective: This study aims to evaluate the ability of LLMs for syndrome differentiation thinking and design a method to effectively enhance their performance in this area.

Methods: We decomposed the process of syndrome differentiation thinking in TCM into three core tasks: pathogenesis inference, syndrome inference, and diagnostic suggestion. To evaluate the performance of LLMs in these tasks, we constructed a high-quality evaluation dataset, forming a reliable foundation for quantitative assessment of their capabilities. Furthermore, we developed a methodology for generating instruction data based on the idea of an "open-book exam," customized three data templates, and dynamically retrieved task-relevant professional knowledge that was inserted into predefined positions within the templates. This approach effectively generates high-quality instruction data that aligns with the unique characteristics of TCM syndrome differentiation thinking. Leveraging this instruction data, we fine-tuned the base model, enhancing the syndrome differentiation thinking ability of the LLMs.

Results: We collected 200 medical cases for the evaluation dataset and standardized them into three types of task questions. We tested general and TCM-specific LLMs, comparing their performance with our proposed solution. The findings demonstrated that our method significantly enhanced LLMs' syndrome differentiation thinking. Our model achieved 85.7% in Task 1 and 81.2% accuracy in Task 2, surpassing the best-performing TCM and general LLMs by 26.3% and 15.8%, respectively. In Task 3, our model achieved a similarity score of 84.3, indicating that the model was remarkably similar to advice given by experts.

Conclusions: Existing general LLMs and TCM-specific LLMs continue to have significant limitations in the core task of syndrome differentiation thinking. Our research shows that fine-tuning LLMs by designing professional instruction templates and generating high-quality instruction data can significantly improve their performance on core tasks. The optimized LLMs show a high degree of similarity in reasoning results, consistent with the opinions of domain experts, indicating that they can simulate syndrome differentiation thinking to a certain extent. These findings have important theoretical and practical significance for in-depth interpretation of the complexity of the clinical diagnosis and treatment process of TCM.

Keywords: RAG; TCM LLMs; instruction tuning; large language model; syndrome differentiation thinking; traditional Chinese medicine.

Figure 1.. A framework for generating instruction data is formed by integrating the local knowledge base into the Syndrome Differentiation Thinking Template.

Figure 2.. Performance comparison of RAG, CoT, and Base models across Two Tasks; Task 1 pathogenesis inference (left), Task 2 symptom inference (right). CoT: Chain-of-Thought; RAG: Retrieval Augmented Generation.

Figure 3.. Comparison of results on Task 3: CoT, RAG and our method. CoT: Chain-of-Thought; RAG: Retrieval Augmented Generation.