Abstract:

Objective To develop an intelligent question-answering（Q&A） agent for evidence-based nursing and systematically compare the performance of knowledge base and multi-model collaborative-driven agent with that of general large language models（LLMs）. Methods An intelligent Q&A system was developed using the low-code Coze platform. A targeted knowledge base was constructed by integrating content from the Evidence-Based Nursing textbook. Retrieval-Augmented Generation（RAG） technology was employed for knowledge retrieval. A collaborative workflow was designed involving two base models，one expert model，and one analytical model. Final outputs were generated using decision rules based on “hard voting” and “expert model priority”. Evaluation was conducted using a standard in-class test from the Fudan University Evidence-Based Nursing Center to compare performance differences among different models（the intelligent Q&A agent for evidence-based nursing，DeepSeek，Kimi，ChatGPT-4o Mini） across overall questions，questions from various chapter categories，and questions with varying difficulty levels（1-5）. Results In terms of overall answer accuracy，the intelligent Q&A agent for evidence-based nursing performed better than the three mainstream models（P<0.05）. In terms of questions across different difficulty levels，the intelligent Q&A agent for evidence-based nursing showed superior performance than the three mainstream models on level-3 questions（P<0.05）. Conclusion The constructed intelligent Q&A agent for evidence-based nursing demonstrates professionalism and reliability in evidence-based nursing. Future efforts should focus on expanding the breadth and depth of the knowledge base，building a dynamically updated structured knowledge graph，and developing a multimodal data support system to continuously make the intelligent Q&A agent for evidence-based nursing adapt to the evolution of evidence-based nursing practice.

Key words: Artificial Intelligence, Large Language Models, Agent, Evidence-based nursing, Knowledge base