心力衰竭重症患者院内死亡率的预测模型:基于MIMIC-Ⅲ数据库的回顾性研究  

Nomogram for predicting hospital mortality in critically ill patients with heart failure:a retrospective study based on MIMIC-Ⅲdatabase

作　　者：王羽 刘志玄 邱洪斌[1] 张艺潆[1] WANG Yu;LIU Zhixuan;QIU Hongbin;ZHANG Yiying(School of Public Health,Jiamusi University,Jiamusi,Liaoning 154007,China)

机构地区：[1]佳木斯大学公共卫生学院,黑龙江省佳木斯市154007

出　　处：《中国动脉硬化杂志》2023年第3期245-252,共8页Chinese Journal of Arteriosclerosis

基　　金：黑龙江省自然科学基金重点项目(ZD2022H006);黑龙江省博士后科学基金(LBH-Q21047);痛风病因学与功能食品研究创新团队、黑龙江省北药与功能食品优势特色学科建设项目(HLJTSXK-2022-03);佳木斯大学科技创新团队(cxtd202101);佳木斯大学国家基金培育项目(JMSUGPZR2022-022);佳木斯大学校级重点课题(JMSUSKZX-ZD002)。

摘　　要：[目的]基于重症监护医学数据库(MIMIC-Ⅲ)分析心力衰竭重症患者院内死亡率的预后因素并构建预测模型。[方法]从MIMIC-Ⅲ数据库中提取心力衰竭患者的相关数据。随机将研究对象(n=8604)按7∶3分为训练组(n=6022)和验证组(n=2582),结局为院内死亡率。对训练组进行LASSO-Logistic回归分析,确定心力衰竭患者院内死亡率的预后因素,并据此构建列线图模型。受试者工作特征(ROC)曲线评估列线图模型的区分度,校准曲线评估列线图模型的校准能力,决策曲线分析(DCA)和临床影响曲线(CIC)评估列线图模型的临床疗效。[结果]LASSO-Logistic分析表明,红细胞分布宽度(RDW)、呼吸频率、血氧饱和度、急性生理评分Ⅲ(APSⅢ)评分和简化急性生理评分Ⅱ(SAPSⅡ)是心力衰竭重症患者院内死亡率的独立预测因素。在训练组和验证组中,ROC曲线下面积(AUC)分别为0.775(95%CI:0.757~0.792)和0.767(95%CI:0.742~0.793),校准曲线与对角线均高度重合,平均绝对误差为0.009和0.016,表明预测模型具有较好的区分度和校准度。同时,DCA和CIC曲线显示,预测模型在大部分的阈值概率范围内提供了显著的净收益。[结论]列线图模型能简单而准确地预测心力衰竭重症患者院内死亡率。Aim Construct a nomogram for predicting hospital mortality in critically ill patients with heart failure(HF)from Medical Information Mart for Intensive CareⅢ(MIMIC-Ⅲ)database.Methods Data were extracted involving critically ill patients with HF from MIMIC-Ⅲdatabase.All eligible patients(n=8604)were randomly classified into the training set(n=6022)and validation set(n=2582)with a ratio of 7∶3,and the outcome was hospital mortality.LASSO-Logistic analysis in the training set was used to determine the prognostic factors and the nomogram for predicting hospital mortality was thereby constructed.Receiver operating characteristic(ROC)curve,calibration curve,decision analysis curve(DCA)and clinical impact curve(CIC)were generated to assess the discrimination,calibration and clinical utility of the nomogram,respectively.Results LASSO-Logistic analysis showed that red blood cell distribution width(RDW),respiratory rate,oxygen saturation,acute physiology scoreⅢ(APSⅢ)and simplified acute physiology scoreⅡ(SAPSⅡ)were independent predictors.In both training and validation set,the area under the curve(AUC)was 0.775(95%CI:0.757~0.792)and 0.767(95%CI:0.742~0.793),respectively.Calibration curve was highly consistent with the diagonal line,and the mean absolute error was 0.009 and 0.016.AUC and calibration curve showed great performance of the predictive model in discrimination and calibration.Meanwhile,DCA and CIC revealed that the predictive model provided significant net benefits for most threshold probabilities.Conclusion Nomogram is a simple and accurate tool for predicting hospital mortality in critically ill patients with HF.

关 键 词：重症监护医学数据库 列线图 最小绝对收缩选择算法 决策曲线分析 临床影响曲线 

分 类 号：R5[医药卫生—内科学]