大语言模型在空勤人员高尿酸饮食管理中的应用研究  

Research on the application of large language model in the diet management for aircrew with hyperuricemia

作　　者：钱棪梅 徐宏伟 唐艳超 余运贤[1] QIAN Yanmei;XU Hongwei;TANG Yanchao;YU Yunxian(Department of Epidemiology and Health Statistics,School of Public Health,Zhejiang University School of Medicine,Hangzhou 310058,Zhejiang Province,China;Health Management Section of the third District of Recuperation,Special Service Recuperation Center of PLA Air Force)

机构地区：[1]浙江大学医学院公共卫生学院流行病与卫生统计学系,杭州310058 [2]空军杭州特勤疗养中心疗养三区健康管理科

出　　处：《中国数字医学》2025年第4期78-83,共6页China Digital Medicine

摘　　要：目的:探讨大语言模型在空勤人员疗养期间高尿酸饮食管理中的应用效果。方法:以通义千问2.5为测试模型并邀请1名注册营养师,以《成人高尿酸血症与痛风食养指南(2024年版)》为依据,分别针对样本食谱进行量化评分,对评分结果进行相关性、差异性和一致性分析;限定食材品类、限制摄入总热量等要求,通过大语言模型制定空勤人员高尿酸饮食一日食谱;采用NASA-TLX自我评定量表,对营养师采用大语言模型辅助完成食谱制定任务的认知负荷程度评价。结果:Spearman秩相关检验显示两者评分相关系数为0.716,存在显著相关性,P<0.001。Wilcoxon符号秩检验显示两者评分没有显著差异性,P=0.739。Bland-Altman分析两者评分差值的95%一致性界限为-2.699~2.975,t=0.638,P=0.526,具有较好的一致性;大语言模型可根据任务特点精准生成符合需求的一日食谱;营养师通过大语言模型辅助完成食谱制定任务的认知负荷明显降低。结论:大语言模型在空勤人员高尿酸饮食管理中展现出良好的实际效果和应用潜力,为个性化饮食方案地快速制定提供了有力支持。Objective To explore the application effect of large language model in the diet management of for aircrew with hyperuricemia during convalescence.Methods Using Tongyi Qianwen 2.5 as the testing model,and a registered dietitian was invited to quantitatively score the sample diet according to the"Dietary Guidelines for Adults with Hyperuricemia and Gout(2024 Edition)",and the correlation,difference and consistency of the two scores were analyzed.Requirements such as limiting the types of ingredients and restricting the total calorie intake were set,and a large language model was used to develop a daily diet plan for aircrew with hyperuricemia.The NASA-TLX selfassessment scale was used to evaluate the cognitive load level of dietitians using the LLM to assist them in formulating recipes.Results Spearman rank correlation test showed that the correlation coefficient between the two scores was 0.716(P<0.001).Wilcoxon signed rank test showed no significant difference between the two scores(P=0.739).Bland-Altman analyzed that the 95%consistency limit of the difference between the two scores was-2.699~2.975,t=0.638,P=0.526,indicating a good consistency.LLM can accurately generate daily recipes according to task characteristics.The cognitive load of dietitians completing recipe-making tasks was significantly reduced with the assistance of LLM.Conclusion LLM shows great practical effect and application potential in the diet management for aircrews with hyperuricemia,providing strong support for the rapid development of personalized diet planning.

关 键 词：人工智能 大语言模型 高尿酸血症 营养食谱 认知负荷 

分 类 号：R197.3[医药卫生—卫生事业管理] R319[医药卫生—公共卫生与预防医学]