摘要
目的通过基于多模态数据的机器学习算法构建发生创伤性凝血病(TIC)进展预测模型,并评估其预测效能,为早期识别TIC进展高风险患者、及时干预提供依据。方法采用便利抽样法,选取2022年1月至2024年12月入住首都医科大学附属北京积水潭医院创伤中心的734例创伤患者为研究对象。收集患者的人口统计学资料、临床检验指标、影像学数据、创伤严重程度评分(ISS)等多模态数据。运用Python 3.8软件中的scikit-learn库构建随机森林、支持向量机和神经网络3种机器学习预测模型,计算各模型的准确率、精确率、召回率、F1值及AUC。结果单因素分析显示,收缩压、舒张压、血红蛋白(Hb)、血小板计数(PLT)、凝血酶原时间(PT)、D-二聚体、ISS、是否合并颅脑损伤、是否合并胸部损伤以及是否合并腹部损伤均与发生TIC进展相关(均P<0.05)。多因素logistic回归分析结果显示,舒张压、Hb、PLT、PT、D-二聚体以及合并胸部损伤均是预测发生TIC进展的独立危险因素(均P<0.05)。随机森林、支持向量机与神经网络预测模型的AUC分别为0.812、0.856与0.893。结论构建的神经网络预测模型在预测发生TIC进展方面效能最优,能够有效预测创伤患者发生TIC进展的风险,有助于临床医务人员早期识别TIC高危患者,为及时干预提供决策支持。
Objective To construct a predictive model for the progression of traumatic coagulopathy(TIC)using machine learning algorithms based on multimodal data,and to evaluate its predictive efficacy,so as to provide evidence for early identification of patients at high risk of TIC progression and timely intervention.Methods A convenience sampling method was adopted to select 734 trauma patients admitted to the Trauma Center of Beijing Jishuitan Hospital,Capital Medical University from January 2022 to December 2024 as the research subjects.Multimodal data including demographic information,clinical laboratory indicators,imaging data,and Injury Severity Score(ISS)of the patients were collected.Three machine learning predictive models(random forest,support vector machine,and neural network)were constructed using the scikit-learn library in Python 3.8 software,and the accuracy,precision,recall,F1 score,and AUC of each model were calculated.Results Univariate analysis showed that systolic blood pressure,diastolic blood pressure,hemoglobin(Hb),platelet count(PLT),prothrombin time(PT),D-dimer,ISS,complicated craniocerebral injury,complicated thoracic injury,and complicated abdominal injury were associated with the progression of TIC(all P<0.05).Multivariate logistic regression analysis indicated that diastolic blood pressure,Hb,PLT,PT,D-dimer,and complicated thoracic injury were independent risk factors for predicting TIC progression(all P<0.05).The AUC values of random forest,support vector machine,and neural network models were 0.812,0.856,and 0.893,respectively.Conclusion The constructed neural network model has the optimal efficacy in predicting TIC progression,which can effectively predict the risk of TIC progression in trauma patients.It helps clinical medical staff to early identify high-risk patients with TIC and provides decision support for timely intervention.
作者
刘颖
吴俊
LIU Ying;WU Jun(Fourth Clinical School,Peking University,Beijing 100035,China;不详)
出处
《浙江医学》
2026年第5期471-475,共5页
Zhejiang Medical Journal
基金
北京市自然科学基金-大兴创新联合基金项目(L246006)
北京市卫生系统高层次公共卫生技术人才建设项目(02-18)。
关键词
创伤性凝血病
多模态数据
机器学习
预测价值
Traumatic coagulopathy
Multimodal data
Machine learning
Predicted value
