目的 对比分析以ChatGPT-4o和DeepSeek-V3为代表的生成式人工智能(generative artificial intelligence,GenAI)在肠造口患者教育方面生成信息的质量，为护理领域健康教育的智能化改革提供参考。方法 将临床常见的34个肠造口护理问题分为3类并编写为结构化指令，依次向2种GenAI进行交互，由6位国内肠造口护理专家对模型输出结果进行盲法评价；再使用改良的确保患者信息质量(ensuring quality information for patients,EQIP)评估工具对2种GenAI生成的信息进行系统评价，分数≥19分为高分模型。结果 ChatGPT-4o和DeepSeek-V3生成肠造口护理信息的准确率分别为82.4%(28/34)、67.6%(23/34)，EQIP评分分别为22、24分。结论 2种GenAI均可作为肠造口患者初步了解医疗信息的辅助工具，涉及决策问题时仍需结合专科医护团队的经验和判断。垂直领域大语言模型的开发和应用是肠造口患者教育智能化改革的重要方向。

Objective To compare and analyze the quality of information generated by generative artificial intelligence (GenAI) represented by ChatGPT-4o and DeepSeek-V3 in patient education for enterostomy,providing reference for the intelligent reform of health education in the nursing field.Methods A total of 34 common questions of clinical enterostomy care were categorized into 3 types and formulated as structured prompts.These prompts were sequentially input into the 2 GenAI models for interaction.The outputs were then blindly evaluated by 6 domestic experts in enterostomy care.Additionally,a modified Ensuring Quality Information for Patients (EQIP) tool was used to systematically evaluate the information generated by the 2 GenAI models,with a score ≥19 indicating a high-quality model.Results The accuracy rates of information on enterostomy care generated by ChatGPT-4o and DeepSeek-V3 were 82.4% (28/34) and 67.6% (23/34),respectively.Their EQIP scores were 22 and 24 points,respectively.Conclusions Both GenAI models can serve as auxiliary tools for enterostomy patients to gain preliminary understanding of medical information.However,it is still necessary to integrate the experience and judgment of specialized healthcare teams in terms of decisions.The development and application of large language models in vertical domains represent an important direction for the intelligent reform of patient education in enterostomy care.

R473.6

西安交通大学基本科研业务团队和人才支持-学科交叉团队项目（xtr062025009）