使用机器学习分析脓毒症合并急性呼吸窘迫综合征患者发生急性肾损伤的影响因素及构建预测模型  

Risk factors and prediction model construction of acute kidney injury in patients with sepsis complicated with acute respira⁃tory distress syndrome

作　　者：冯超男 张钰 Feng Chaonan;Zhang Yu(Department of Anesthesiology,Nanjing Gaochun People's Hospital,Nanjing 211300,China;Department of Anesthesiology,the Affiliated Hospital of Xuzhou Medical University,Xuzhou 211002,China)

机构地区：[1]南京市高淳人民医院麻醉科,南京211300 [2]徐州医科大学附属医院麻醉科,徐州221002

出　　处：《国际麻醉学与复苏杂志》2024年第12期1256-1262,共7页International Journal of Anesthesiology and Resuscitation

摘　　要：目的分析脓毒症合并急性呼吸窘迫综合征(ARDS)患者急性肾损伤(AKI)发生的危险因素并建立AKI发生的预测模型。方法从重症监护医学信息中心(MIMIC⁃Ⅲ,版本号1.4)数据库中提取脓毒症合并ARDS患者的临床数据进行训练和内部验证,回顾性收集南京市高淳人民医院脓毒症合并ARDS患者的临床数据进行外部验证。从MIMIC⁃Ⅲ数据库中提取的病例依据7∶3的比例按照随机抽样法分为建模组(760例)和内部验证组(325例),回顾性收集的病例为外部验证组(430例)。依据建模组760例患者的数据,采用最小绝对收缩和选择算子(LASSO)回归分析探究影响脓毒症合并ARDS患者发生AKI的影响因素,将LASSO回归中系数非零的变量纳入预测模型,并使用支持向量机(SVM)、自适应提升(Adaboost)、逻辑回归(LR)和极端梯度提升(Xgboost)算法构建预测模型。通过绘制校准曲线、计算受试者操作特征曲线下面积(AUC)判断预测模型性能表现,而后在外部验证组中进行预测模型的验证。结果脓毒症相关性器官功能衰竭评价(SOFA)评分[比值比(OR)1.142,95%置信区间(CI)1.071~1.218,P<0.001]、氧合指数(OR 0.993,95%CI 0.990~0.996,P<0.001)、第1个24 h内尿量(OR 0.996,95%CI 0.993~0.999,P<0.001)、呼气末正压(PEEP)(OR 1.097,95%CI 1.038~1.160,P=0.001)、高血压(OR 1.598,95%CI 1.145~2.231,P=0.006)是脓毒症合并ARDS患者发生AKI的独立危险因素。预测模型建模组和内部验证组的AUC在SVM、Adaboost、LR、XGBoost算法中分别为0.744(95%CI 0.706~0.782)、1.000(95%CI 1.000~1.000)、0.742(95%CI 0.706~0.779)、0.887(95%CI 0.863~0.912)和0.704(95%CI 0.644~0.765)、0.781(95%CI 0.728~0.835)、0.716(95%CI 0.657~0.775)、0.799(95%CI 0.746~0.853)。预测模型在外部验证组也展示出一定的预测能力,与其对应的校准曲线也表现出较好的一致性。结论脓毒症合并ARDS患者发生AKI的独立危险因素有SOFA评分、氧合指数、第1个24 h内尿量、PEEP、�Objective To analyze the risk factors of acute kidney injury(AKI)in patients with sepsis complicated by acute re⁃spiratory distress syndrome(ARDS)and to establish a predictive model for predicting the occurrence of AKI.Methods The clinical data from patients with sepsis and ARDS were extracted from the Medical Information Mart for Intensive Care(MIMIC⁃Ⅲ,Version 1.4)to be used for training and internal validation.Additionally,data were retrospectively collected from sepsis patients with ARDS at Nan⁃jing Gaochun People's Hospital for external validation.Sepsis patients were divided in a 7∶3 ratio into a modeling group(n=760)and an internal validation group(n=325).The cases retrospectively collected formed the external validation group(n=430).Using the data from the 760 patients in the modeling group,the least absolute shrinkage and selection operator(LASSO)was employed to identify the risk factors for AKI in patients with sepsis complicated by ARDS.Variables with non⁃zero coefficients from LASSO regression were in⁃cluded in the prediction model.Support vector machine(SVM),adaptive boosting(Adaboost),logistic regression(LR),and extreme gra⁃dient boosting(Xgboost)algorithms were used to construct the prediction model.The accuracy of the model was assessed using a cali⁃bration curve and by calculating the area under the receiver operating characteristic curve(AUC).The model underwent external valida⁃tion subsequently.Results Among the 1085 patients with sepsis and ARDS included in the study,199 developed AKI,resulting in an incidence rate of 18%.Independent risk factors identified for AKI included the sepsis⁃related organ failure assessment(SOFA)score[odds ratio(OR)1.142(95%confidence interval(CI)1.071,1.218),P<0.001],oxygenation index[OR 0.993(95%CI 0.990,0.996),P<0.001],first 24⁃hour urine output[OR 0.996(95%CI 0.9993,0.999),P<0.001],positive end⁃expiratory pressure(PEEP)[OR 1.097(95%CI 1.038,1.160),P=0.001],hypertension[OR 1.598(95%CI 1.145,2.231),P=0.006].The AUCs for the modeling group a

关 键 词：脓毒症 急性呼吸窘迫综合征 机器学习 急性肾损伤 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]