机器学习预测脓毒症患者血流动力学不稳定  

To predict hemodynamic instability with machine learning in ICU patients with sepsis

作　　者：金旭婷 李佳媚[1] 李若寒 高雅[1] 张静静[1] 任佳佳 张小玲[1] 王小闯[1] 王岗[1] Jin Xuting;Li Jiamei;Li Ruohan;Gao Ya;Zhang Jingjing;Ren Jiajia;Zhang Xiaoling;Wang Xiaochuang;Wang Gang(Department of Critical Care Medicine,the Second Affiliated Hospital of Xi′an Jiaotong University,Xi′an 710004,China)

机构地区：[1]西安交通大学第二附属医院重症医学科,陕西西安710004

出　　处：《中国急救医学》2023年第11期892-897,共6页Chinese Journal of Critical Care Medicine

基　　金：陕西省“高层次人才特殊支持计划”;西安交通大学医学“基础-临床”融合创新项目(YXJLRH2022060)。

摘　　要：目的 在脓毒症患者中建立基于生命体征监测数据的血流动力学不稳定事件预测模型。方法 从重症监护病房合作研究数据库(eICU-CRD)的脓毒症患者中识别在重症监护病房(ICU)住院期间发生的血流动力学不稳定事件。提取事件发生前6 h的连续生命体征监测数据(包括心率、呼吸和血氧饱和度)作为阳性样本,在未发生血流动力学不稳定事件的脓毒症患者中随机抽取6 h生命体征监测数据为阴性对照样本。建立并训练极致梯度提升(XGBoost)、轻量的梯度提升机(LightGBM)以及深度神经网络(DNN)模型进行建模及训练。利用受试者工作特征曲线下面积(ROC-AUC)对模型效能进行评估,使用最优的模型在脓毒症血流动力学不稳定事件发生前1 h和前2 h对事件的发生进行预测。结果 本研究共提取阳性样本2 569例,阴性对照样本7 048例。XGBoost、LightGBM以及DNN模型预测脓毒症血流动力学不稳定事件的ROC-AUC值分别为0.78、0.77和0.61。XGBoost模型在脓毒症血流动力学不稳定事件发生前1 h、前2 h进行预测的ROC-AUC值分别为0.76和0.75。结论 在ICU的脓毒症患者中,基于连续生命体征监测数据的机器学习模型可用于血流动力学不稳定事件的预测。Objective To establish machine learning models based on the records of vital signs to predict the occurrence of hemodynamic instability in ICU patients with sepsis.Methods Hemodynamic instable events in adult patients with sepsis were recognized from the eICU collaborative research database(eICU-CRD).Records of vital signs including heart rate,respiratory rate,and peripheral oxygen saturation during the 6 hours before hemodynamic instability were extracted as positive samples,and the records from the patients without hemodynamic instability were extracted as negative samples.Extreme gradient boosting(XGBoost),light gradient boosting machine(LightGBM),and deep neural network(DNN)models were built and trained to predict hemodynamic instability.Area under the receiver operating curve(ROC-AUC)values were calculated to assess the performance of models.The optimal model was employed to predict hemodynamic instability at 1 h and 2 h before the event happened.Results A total of 2569 positive samples and 7048 negative samples were included.The ROC-AUCs for hemodynamic instability prediction of the XGBoost,LightGBM and DNN models were 0.78,0.77 and 0.61,respectively.The ROC-AUCs were 0.76 and 0.75 for XGBoost in predicting hemodynamic instability at 1 h and 2 h in advance,respectively.Conclusions In critically ill patients with sepsis,machine learning model based on continuous records of vital signs could be applied in the prediction of hemodynamic instability.

关 键 词：机器学习 脓毒症 血流动力学不稳定 重症监护病房(ICU) 极致梯度提升 轻量的梯度提升机 深度神经网络 

分 类 号：R459.7[医药卫生—急诊医学]