A large language model-powered reflective AI agent for evidence-based nursing education: Design and evaluation

To develop and evaluate the Evidence-Based Nursing Expert (EBN-Expert), a domain-specific intelligent agent designed to support reflective learning in evidence-based nursing education.

Evidence-based practice (EBP) is a core competency in nursing education yet remains challenging for students to master due to its emphasis on critical thinking and decision-making. Traditional teaching methods often fall short in fostering these higher-order skills. While large language models (LLMs) are increasingly used in education, general-purpose systems offer limited support for the specific demands of EBP.

A comparative evaluation study using standardized exam questions aligned with a widely used evidence-based nursing curriculum.

The EBN-Expert was constructed with content from the textbook Evidence-Based Nursing and designed around a closed-loop workflow incorporating Retrieval, Reflection and Decision-making modules. The system was evaluated using 124 standardized test items and compared with three leading general-purpose LLMs: ChatGPT-o1, DeepSeek-R1 and Kimi.

The EBN-Expert significantly outperformed three general-purpose models (P < 0.001), achieving the highest overall score (1035.00 ± 17.32). It led in most of the 12 textbook chapters, especially in those requiring methodological or interpretive reasoning. It maintained perfect accuracy in true/false questions and excelled in multiple-choice formats. A mixed effects model further confirmed its superior accuracy and response consistency, especially at intermediate difficulty levels (Level 2 and Level 4).

The EBN-Expert illustrates the promise of domain-specific generative AI tools in nursing education. Its alignment with curriculum structure, high accuracy and reflective reasoning capabilities position it as a scalable and trustworthy approach for advancing EBP training.