利用决策树算法建立严重脓毒症院内死亡的风险预测模型及其效能验证  

Establishing a risk prediction model for in-hospital mortality in severe sepsis using decision tree algorithm and its effectiveness verification

作　　者：朱震寒 姜婷婷 徐宪辉 ZHU Zhenhan;JIANG Tingting;XU Xianhui(Department of Emergency Medicine,Navy 971 Hospital of the People's Liberation Army,Qingdao 266071,China)

机构地区：[1]中国人民解放军海军第九七一医院急诊医学科,山东青岛266071

出　　处：《东南大学学报(医学版)》2025年第1期90-98,共9页Journal of Southeast University(Medical Science Edition)

摘　　要：目的:探讨严重脓毒症患者院内死亡的风险因素,并利用决策树算法建立院内死亡风险预测模型且验证模型效能,以指导临床工作。方法:回顾性选取本院243例严重脓毒症患者,根据院内死亡情况将其分为死亡组(n=98)与存活组(n=145)。比较两组临床资料,通过多因素Logistic回归分析影响患者院内死亡的相关因素。另根据2∶1比例,将243例严重脓毒症患者随机分为模型组(n=162)与验证组(n=81),基于模型组数据利用决策树算法建立严重脓毒症患者院内死亡的风险预测模型,利用验证组数据对该模型进行验证。结果:经多因素Logistic回归分析,年龄≥60岁、入院时心率>100次·min^(-1)、入院时平均动脉压<70 mmHg、入院时氧合指数<300 mmHg、急性生理学与慢性健康状况评分系统Ⅱ(APACHEⅡ)评分≥30分、序贯器官衰竭评分(SOFA)≥8分、中性粒细胞计数与淋巴细胞计数比值(NLR)>3、红细胞压积(HCT)<0.3、ρ(白蛋白)<35 g·L^(-1)、c(血肌酐)>176.8μmol·L^(-1)、活化部分凝血活酶时间(APTT)>45 s、ρ[B型脑钠肽前体(NT-proBNP)]≥2000 pg·mL^(-1)、c(血乳酸)>4.0 mmol·L^(-1)、ρ[C反应蛋白(CRP)]>50 mg·L^(-1)、ρ[降钙素原(PCT)]>2 ng·mL^(-1)均是严重脓毒症患者院内死亡的危险因素(P<0.05);基于模型组数据,建立含SOFA、APACHEⅡ评分、血乳酸、白蛋白、NT-proBNP、NLR、CRP、PCT共8个解释变量在内的严重脓毒症患者院内死亡的风险预测决策树模型,其中SOFA最为重要。基于验证组数据对风险预测决策树模型进行验证,结果显示该风险预测决策树模型预测严重脓毒症患者院内死亡的灵敏度、特异度、准确度分别为87.10%、84.00%、85.19%。结论:利用决策树算法建立的严重脓毒症院内死亡风险预测模型包含SOFA、APACHEⅡ评分、血乳酸、白蛋白、NT-proBNP、NLR、CRP、PCT共8个变量,其中SOFA是最重要影响因素,该决策树模型对严重脓毒症院内死亡�Objective:To explore the risk factors of in-hospital mortality in patients with severe sepsis,and establishing an in-hospital mortality risk prediction model using decision tree algorithm and validating its effectiveness,to guide clinical work.Methods:243 patients with severe sepsis were retrospectively selected and divided into a death group(n=98)and a survival group(n=145)based on in-hospital mortality.The clinical data between the two groups was compared,and the relevant factors affecting in-hospital mortality of patients was screened using multivariate Logistic regression analysis.In addition,243 patients with severe sepsis were randomly divided into a model group(n=162)and a validation group(n=81)in 2∶1 ratio,and a risk prediction model for in-hospital mortality in severe sepsis patients was established using decision tree algorithm based on model group data,and the model was validated using validation group data.Results:According to multiple Logistic regression analysis,age≥60 years old,heart rate>100 beats·min^(-1)at admission,mean arterial pressure<70 mmHg at admission,oxygenation index<300 mmHg at admission,acute physiology and chronic health score systemⅡ(APACHEⅡ)score≥30 points,sequential organ failure assessment(SOFA)score≥8 points,ratio of neutrophil count to lymphocyte count(NLR)>3,hematocrit(HCT)<0.3,ρ(albumin)<35 g·L^(-1),c(blood creatinine)>176.8μmol·L^(-1),activated partial thromboplastin time(APTT)>45 s,ρ[N terminal pro B type natriuretic peptide(NT-proBNP)]≥2000 pg·mL^(-1),c(blood lactate)>4.0 mmol·L^(-1),ρ[C-reactive protein(CRP)]>50 mg·L^(-1)andρ[procalcitonin(PCT)]>2 ng·mL^(-1)were all risk factors for in-hospital mortality in patients with severe sepsis(P<0.05).Based on model group data,a risk prediction decision tree model for in-hospital mortality in severe sepsis patients was established,and it included eight explanatory variables of SOFA score,APACHEⅡscore,blood lactate,albumin,NT proBNP,NLR,CRP and PCT,among which SOFA score was the most important.Decision

关 键 词：严重脓毒症 死亡 风险因素 决策树模型 

分 类 号：R631.2[医药卫生—外科学]