肺部感染合并重症脓毒症30天死亡的预测模型

A prediction model for the 30-day mortality of the critical patients with pulmonary infection and sepsis

目的探讨肺部感染合并脓毒症重症患者30天死亡的独立危险因素并建立预测模型。方法分析MIMIC-Ⅲ数据库诊断为肺部感染与脓毒症的患者,其中CareVue数据库为训练队列(n=934),Metavision数据库为外部验证队列(n=687)。建立COX比例风险回归模型,筛选独立危险因素,绘制列线图。对模型进行内部交叉验证、外部验证,运用受试者操作特征曲线(receiver operator characteristic curve,ROC曲线)、Calibration图及决策曲线分析分别检测模型的区分度、校准度及获益,并与序贯器官衰竭评分(sequential organ failure assessment,SOFA)模型进行比较。结果年龄、SOFA评分、白细胞计数≤4×10^(9)/L、中性粒细胞百分比>85%、血小板计数≤100×10^(9)/L、血小板计数>300×10^(9)/L、红细胞体积分布宽度>15%、血尿素氮、乳酸脱氢酶水平为独立危险因素。模型的ROC曲线下面积为0.747(训练队列)及0.708(外部验证队列),在区分度、校准度、获益上优于SOFA评分模型。结论本研究建立的模型可准确、有效地预测疾病死亡风险,为早期识别高危患者提供可视化评估方法。

Objective To explore independent risk factors for 30-day mortality in critical patients with pulmonary infection and sepsis,and build a prediction model.Methods Patients diagnosed with pulmonary infection and sepsis in the MIMIC-Ⅲdatabase were analyzed.The CareVue database was the training cohort(n=934),and the Metavision database was the external validation cohort(n=687).A COX proportional hazards regression model was established to screen independent risk factors and draw a nomogram.We conducted internal cross-validation and external validation of the model.Using the receiver operator characteristic(ROC)curve,Calibration chart,and decision curve analysis,we detected the discrimination,calibration,and benefit of the model respectively,comparing with the SOFA scoring model.Results Age,SOFA score,white blood cell count≤4×10^(9)/L,neutrophilic granulocyte percentage(NEU%)>85%,platelet count(PLT)≤100×10^(9)/L,PLT>300×10^(9)/L,red cell distribution width>15%,blood urea nitrogen,and lactate dehydrogenase were independent risk factors.The areas under the ROC curve of the model were 0.747(training cohort)and 0.708(external validation cohort),respectively,which was superior to the SOFA scoring model in terms of discrimination,calibration,and benefit.Conclusion The model established in this study can accurately and effectively predict the risk of the disease mortality,and provide a visual assessment method for early identification of high-risk patients.