ICU住院患者重度急性呼吸窘迫综合征早期预测模型的构建  

Establishing an early prediction model for severe acute respiratory distress syndrome

作　　者：武俊伟 刘超[3] 王雪 杜甲珺[2] 李姣[1] 谢菲[5] WU Junwei;LIU Chao;WANG Xue;DU Jiajun;LI Jiao;XIE Fei(Institute of Medical Information,Chinese Academy of Medical Sciences&Peking Union Medical College,Beijing 100020,China;Library of Graduate School,Chinese PLA General Hospital,Beijing 100853,China;Yidu Cloud(Beijing)Technology Ltd,Beijing 100083,China;Medical Library,Xuanwu Hospital,Capital Medical University,Beijing 100053,China;College of Pulmonary and Critical Care Medicine,Chinese PLA General Hospital,Beijing 100853,China)

机构地区：[1]中国医学科学院/北京协和医学院医学信息研究所,北京100020 [2]解放军总医院研究生院图书馆,北京100853 [3]医渡云(北京)技术有限公司,北京100083 [4]首都医科大学宣武医院图书馆,北京100053 [5]解放军总医院第八医学中心呼吸与危重症医学部,北京100853

出　　处：《解放军医学院学报》2022年第11期1146-1150,共5页Academic Journal of Chinese PLA Medical School

基　　金：国家自然科学基金项目(61976223);解放军总医院医学创新研究项目(2019YFC0121703);解放军总医院转化医学重点项目(2018TM-03);解放军总医院临床扶持基金(2018FC-WJFWZX-2-04)。

摘　　要：背景急性呼吸窘迫综合征(acute respiratory distress syndrome,ARDS)发病率高,10%的ICU住院是由ARDS所致,临床特征通常在诱发事件后6~72 h出现,并迅速加重。其死亡率亦相当高且会随病情严重程度而增加。目的构建一种便捷、无创的危重症ARDS早期预测模型。方法采用麻省理工学院与飞利浦创建的eICU协作研究数据集,从中检索诊断为ARDS患者的呼吸频率、体温、心跳三种生命体征数据以及氧合指数(PaO_(2)/FiO_(2))。PaO_(2)/FiO_(2)≤100 mmHg为重度ARDS。以每个氧合指数观测点(ARDS诊断时间点)为原点,96 h为一个时间窗,应用逻辑回归、随机森林及LightGBM构建预测模型,分析诊断前6~96 h、6~48 h以及6~24 h的生命体征数据预测是否会发生重度ARDS。通过oob评分、交叉验证以及校准曲线评估模型性能,并选取2014年解放军总医院医院呼吸重症监护室的ARDS病例对模型进行外部验证。结果从eICU数据库检索纳入232例ARDS患者共3140次氧合指数测量记录,其中PaO_(2)/FiO_(2)≤100 mmHg(1 mmHg=0.133 kPa)共1042次。以6~96 h、6~48 h以及6~24 h的生命体征数据各自采用逻辑回归、随机森林及LightGBM建立了9个预测模型。不同时间窗比较,6~96 h的预测准确率及AUC最高;不同模型间比较随机森林模型的诊断性能最优;6~96 h随机森林模型准确率为0.833,AUC为0.885;6~48 h、6~24 h时间窗口的AUC分别为0.815、0.806;LightGBM、逻辑回归模型的6~96 h时间窗口AUC分为0.868、0.634。各模型在解放军总医院ARDS患者中进行验证,依然是6~96 h时间窗的随机森林模型预测性能最佳,准确率为0.834,AUC为0.843。结论基于随机森林构建的ARDS预测模型具有良好的预测能力,通过无创且易获取的心率、体温、呼吸频率三种体征指标,利用提前6~96 h时间窗数据对重度ARDS的发生进行预测,可帮助医护人员更早地进行干预和治疗。Background Acute respiratory distress syndrome(ARDS)is a disease with high morbidity and accounts for 10%of ICU admissions,with clinical features usually presenting within 6-72 hours of the pathogenesis and rapidly worsening.The mortality rate is also high and increases with the severity of the disease.Objective To establish a convenient,noninvasive early prediction model for severe ARDS.Methods The eICU Collaborative Research Database created by MIT and Philips was used to retrieve data on three vital signs(respiratory rate,temperature,and heart rate)and oxygenation index(PaO_(2)/FiO_(2))of patients diagnosed with ARDS,and PaO_(2)/FiO_(2)≤100 mmHg was considered as severe ARDS.96 h was used as a time window,and logistic regression,random forest and LightGBM were applied to establish a prediction model to analyze vital sign data from 6-96 h,6-48 h and 6-24 h before diagnosis to predict whether severe ARDS would occur.Model performance was evaluated by oob score,cross-validation and calibration curve,and also ARDS patients from Respiratory Intensive Care Unit of Chinese PLA General Hospital were selected to validate the models independently.Results A total of 232 patients were retrieved from the eICU database with 3140 oxygenation index measurements during hospitalization,including 1042 with PaO_(2)/FiO_(2)≤100 mmHg.The 6-96 h,6-48 h,and 6-24 h vital sign data were respectively used to build 9 prediction models by using logistic regression,random forest,and LightGBM.Comparing different time windows,the highest prediction accuracy and AUC were obtained for 6-96 h;the best diagnostic performance was obtained for the random forest model compared among different models;the accuracy of the random forest model for 6-96 h was 0.833 and the AUC was 0.885;the AUCs for the 6-48 h and 6-24 h time windows were 0.815 and 0.806,respectively;the AUC of LightGBM,and logistic regression models of 6-96 h time window was 0.868 and 0.634,respectively.Each model was validated in ARDS patients in Chinese PLA General Hospital,and th

关 键 词：eICU数据集 ICU 机器学习 随机森林 急性呼吸窘迫综合征 

分 类 号：R563.8[医药卫生—呼吸系统]