基于学生视角的大语言模型生成护理硕士跨学科连续照护模拟案例的评价研究

An Evaluation Study on the Generation of Interdisciplinary Continuous Care Simulation Cases for Master of Nursing by LLM from a Student Perspective

摘要: 目的：本研究旨在系统比较DeepSeek R1、GEMINI和豆包三种大语言模型(LLM)生成三幕式连续照护案例的质量差异。方法：采用标准化提示词驱动三款LLM各生成5例覆盖常见临床场景的案例，由23名护理硕士研究生基于验证过的评价指标体系进行盲法评价，涵盖案例结构、内容(含跨学科协作与循证思维)及应用效果维度。结果：DeepSeek R1、GEMINI和豆包三种模型在综合总分依次为(258.43 ± 67.68)、(245.96 ± 74.01)、(245.43 ± 74.56)，无显著统计学差异(P > 0.788)，且所有二级指标也均无统计学差异(P > 0.05)；然而，跨学科协作深度仅达中等水平，在该维度的得分DeepSeek R1、GEMINI和豆包依次为(25.57 ± 7.04)、(24.87 ± 6.70)、(23.91 ± 7.67)。结论：本研究所测试的三种LLM在结构化提示下生成案例的整体质量趋于一致且基本满足教学需求，可作为高效的护理模拟案例辅助生成工具，但在跨学科互动真实性与伦理安全性上仍需人工审校。未来研究应聚焦提示词工程优化以增强情境深度，并通过真实教学场景试验验证其对学生临床推理能力的实际影响，从而构建“人机协同”的护理教育新范式。

Abstract: Objective: This study aims to systematically compare the quality differences in generating three-act continuous care case scenarios by three large language models (LLMs): DeepSeek R1, GEMINI, and DouBao. Methods: Standardized prompts were used to generate five cases each from the three LLMs, covering common clinical scenarios. Twenty-three master’s students in nursing conducted blinded evaluations using a validated assessment framework, evaluating case structure, content (including interdisciplinary collaboration and evidence-based thinking), and application effectiveness. Results: The overall composite scores of DeepSeek, GEMINI, and DouBao were 258.43 ± 67.68, 245.96 ± 74.01, and 245.43 ± 74.56, respectively, with no statistically significant differences (P > 0.788); all secondary indicators also showed no significant differences (P > 0.05). However, the depth of interdisciplinary collaboration was only moderate, with scores of 25.57 ± 7.04,24.87 ± 6.70, and 23.91 ± 7.67 for DeepSeek R1, GEMINI, and DouBao, respectively. Conclusion: Current mainstream LLMs produce cases of consistent overall quality under structured prompts, meeting basic teaching requirements and serving as efficient auxiliary tools for nursing simulation case generation. However, manual review remains necessary to ensure the authenticity of interdisciplinary interactions and ethical safety. Future research should focus on optimizing prompt engineering to enhance contextual depth and validate its practical impact on students’ clinical reasoning abilities through real-world teaching scenario experiments, thereby establishing a new paradigm of “human-machine collaboration” in nursing education.