冠心病合并慢性肾脏病患者院内死亡风险预测模型的建立与验证

Establishment and validation of an in-hospital mortality risk prediction model for coronary heart disease and chronic kidney disease

目的 探讨冠心病(coronary artery disease, CAD)合并慢性肾脏病(chronic kidney disease, CKD)患者院内死亡的危险因素,建立死亡风险预测模型并验证其预测效果。方法 回顾性分析2008-2019年重症监护医学信息数据库-Ⅳ(medical information mart for intensive care-Ⅳ,MIMIC-Ⅳ)诊断为CAD合并CKD患者数据。将1 609例患者数据按照7∶3比例分为构建模型的建模组(1 126例)和内部验证组(483例),以院内死亡为结局事件,采用单因素及多因素Logistic回归分析筛选出变量,绘制预测院内死亡风险列线图,并通过受试者工作特征曲线下面积(area under curve, AUC)、敏感度、特异度、Hosmer-Lemeshow检验以及校正曲线对模型进行检验,对模型进行内部验证;收集2023年1-10月重庆市人民医院CAD合并CKD患者287例作为外部验证组,对模型进行外部验证。结果 多因素Logistic回归结果表明,年龄(OR=1.031,95%CI:1.015~1.049)、抗血小板药物(OR=0.520,95%CI:0.342~0.792)、他汀类药物(OR=0.312,95%CI:0.212~0.460)、白细胞计数(OR=1.035,95%CI:1.016~1.057)、主要不良心血管事件(major adverse cardiovascular events, MACE)(OR=2.417,95%CI:1.330~4.643)、eGFR(OR=0.986,95%CI:0.978~0.995)、磷酸盐(OR=1.226,95%CI:1.104~1.362)、碳酸氢盐(OR=0.938,95%CI:0.900~0.977)、氯化物(OR=0.969,95%CI:0.942~0.997)、PCI或CABG(OR=0.362,95%CI:0.174~0.685)均与CAD合并CKD患者发生院内死亡事件相关。由此绘制列线图,建模组AUC为0.800(95%CI:0.768~0.832),敏感度为0.693,特异度为0.760;内部验证组预测模型的AUC为0.724(95%CI:0.663~0.785),敏感度为0.689,特异度为0.682;外部验证组AUC为0.858(95%CI:0.809~0.907)、敏感度为0.800,特异度为0.787,表明该模型具有较好预测能力。建模组与外部验证组Hosmer-Lemeshow检验校准曲线显示模型校准能力良好(χ^(2)=5.975,P=0.650;χ^(2)=7.891,P=0.444)。结论 CAD合并CKD患者院内死亡与年龄、抗血小板药物、他汀类药物、白细胞计数、MACE、eGFR、磷酸盐、碳酸氢盐、PCI或CABG及氯化物有关,构建的列线图对临床CAD合并CKD患者的死亡风险具有良好的预测价值。

Objective To explore the risk factors of in-hospital death in patients with coronary artery disease(CAD)complicated with chronic kidney disease(CKD),and establish a mortality risk prediction model and verify its predictive efficacy.Methods The data of CAD patients with CKD diagnosed by Medical Information Mart for Intensive CareⅣ(MIMIC-Ⅳ)from 2008 to 2019 were collected and retrospectively analyzed.The data of 1609 patients were divided into the modeling group(1126 cases)and the internal validation group(483 cases)in a ratio of 7∶3.The in-hospital death was used as the outcome event.The variables were screened by univariate and multivariate logistic regression analyses,and a nomogram for predicting the risk of in-hospital death was drawn.The model was then assessed with receiver operating characteristic(ROC)curve analysis for area under curve(AUC),sensitivity and specificity,Hosmer-Lemeshow test,and calibration curve analysis,and the model was internally validated.A total of 287 patients with CAD and CKD admitted in Chongqing General Hospital from January to October 2023 were recruited as the external validation group,and the model was externally validated.Results Multivariate logistic regression analysis showed that age(OR=1.031,95%CI:1.015~1.049),antiplatelet drugs(OR=0.520,95%CI:0.342~0.792),statins(OR=0.312,95%CI:0.212~0.460),WBC count(OR=1.035,95%CI:1.016~1.057),major adverse cardiovascular events(MACE)(OR=2.417,95%CI:1.330~4.643),eGFR(OR=0.986,95%CI:0.978~0.995),phosphate(OR=1.226,95%CI:1.104~1.362),bicarbonate(OR=0.938,95%CI:900~0.977),chloride(OR=0.969,95%CI:0.942~0.997),and PCI or CABG(OR=0.362,95%CI:0.174~0.685)were significantly associated with in-hospital mortality in patients with CAD and CKD.The established nomogram showed an AUC value of 0.800(95%CI:0.768~0.832),a sensitivity of 0.693,and a specificity of 0.760 in prediction of in-hospital death.In the internal validation,the AUC value was 0.724(95%CI:0.663~0.785),the sensitivity was 0.689,and the specificity was 0.682;and the above values were 0.858(95%CI:0.809~0.907),0.800,and 0.787,respectively in the external validation,indicating that the model had good predictive ability.Hosmer-Lemeshow test of the modeling group and the external validation group showed that the model had good calibration ability(Chi-square=5.975,P=0.650;Chi-square=7.891,P=0.444).Conclusion The in-hospital mortality of CAD patients with CKD is related to age,antiplatelet drugs,statins,WBC count,MACE incidence,eGFR,phosphate,bicarbonate,PCI or CABG,and chloride.Our established nomogram has a good predictive value for the risk of death in CAD patients with CKD.