医疗卫生装备2026，Vol.47Issue(5)：1-10,10.DOI:10.19745/j.1003-8868.2026068
基于临床大语言模型的文本与结构化数据融合型急诊分诊预测模型研究
Clinical-large-language-based emergency triage prediction model incorporating text and structured data
摘要
Abstract
Objective To develop an emergency triage prediction model based on a clinical large language model by inte-grating unstructured clinical text with structured triage information,so as to facilitate the early prediction of hospitalization and critical outcomes in emergency department(ED)patients.Methods The diagnostic text information and data on age,gender and six vital signs were extracted based on the adult emergency department visit records from the MIMIC-Ⅳ-ED database,including body temperature,heart rate,respiratory rate,oxygen saturation,systolic blood pressure and diastolic blood pressure.The semantic features from the diagnostic texts were encoded with the ClinicalBERT large language model,and then integrated with the structured triage information to build a ClinFusion predictive model.The ClinFusion predictive model had its performance verified by the comparison with various traditional triage systems and machine learning models within a unified evaluation framework.Results In the hospitalization prediction task,the ClinFusion predictive model achieved an area under the receiver operating characteristic curve(AUROC)of 0.896 and an area under the precision-recall curve(AUPRC)of 0.885,outperforming the best-performing machine learning model of multilayer perceptron(MLP)with an AUROC of 0.820 and an AUPRC of 0.791 and the best-performing traditional triage system of emergency severity index(ESI)with an AUROC of 0.708 and an AUPRC of 0.628.In the critical outcome prediction task,the model achieved an AUROC of 0.915 and an AUPRC of 0.541,again outperforming the best-performing machine learning model of gradient boosting(GB)with an AUROC of 0.881 and an AUPRC of 0.398,and the best-performing traditional triage system of ESI with an AUROC of 0.806 and an AUPRC of 0.199.Conclusion The ClinFusion model integrating clinical text with structured triage data significantly improves the ability to predict emergency patients'need for hospitalization and critical outcomes at an early stage,offering an efficient and feasible technical solution for intelligent emergency triage and clinical decision support.[Chinese Medical Equipment Journal,2026,47(5):1-10]关键词
急诊分诊/大语言模型/ClinicalBERT/临床文本/结构化数据/自然语言处理Key words
emergency triage/large language model/ClinicalBERT/clinical text/structured data/natural language processing分类
医药卫生引用本文复制引用
徐小云,王楠,韩宝石,余明,赵艳梅,冯健,胡嘉参,刘斌,张广..基于临床大语言模型的文本与结构化数据融合型急诊分诊预测模型研究[J].医疗卫生装备,2026,47(5):1-10,10.基金项目
天津市医学重点学科建设资助项目(TJYXZDXK-3-001D) （TJYXZDXK-3-001D）
