急诊重症监护室患者早期死亡动态列线图预测模型的建立与验证

Establishment and validation of a dynamic nomogram model predicting early mortality risk among patients in emergency intensive care unit

目的探讨急诊重症监护室患者入院3 d死亡的独立危险因素,以此建立动态网页列线图预测模型并进行验证。方法回顾性收集2018年1月至2020年12月南京医科大学附属无锡第二医院急诊重症监护病房(EICU)收治的634例患者为研究对象,用其临床资料建立预测模型和内部验证。收集2021年1月至12月该院EICU收治的189例患者,用其临床数据作为验证队列。采用单因素和多因素logistic回归分析确定EICU病人早期死亡的危险因素,并构建列线图模型。受试者工作特征曲线(ROC)、C指数及校准曲线评估列线图模型的预测效能,决策曲线分析评估模型的临床获益。在验证队列中对模型进行外部验证。结果全组634例EICU患者中,61例(9.62%)入院3 d内死亡。多因素logistic回归分析显示,危重症营养风险(NUTRIC)评分(OR:1.490,95%CI:0.436~2.116,P<0.001)、国家早期预警评分(NEWS)(OR:1.304,95%CI:1.113~1.528,P=0.001)、合并急性呼吸窘迫综合征(ARDS)(OR:2.183,95%CI:1.220~3.905,P=0.009)和使用机械通气(OR:2.320,95%CI:1.249~4.308,P=0.008)是EICU病人早期死亡的独立危险因素。据此建立静态及网页版动态列线图预测模型(网址:https://wxeysunyutinggrxjbk.shinyapps.io/DynNomEICUapp/),C指数和曲线下面积(area under the curve,AUC)分别为0.832和0.895,校准曲线显示预测概率和实际概率之间具有较好的一致性。模型的外部验证也显示出良好的区分度(C指数=0.862,AUC=0.892)和一致性。决策曲线分析表明该预测模型的临床获益性和应用价值。结论基于4个独立危险因素构建的动态列线图模型能够便捷、可靠的预测EICU患者早期死亡风险,有助于精准预警患者的病情严重程度及预后,优化临床干预决策,提高救治成功率。

Objective To investigate the independent risk factors for 3-day mortality among patients in emergency intensive care unit(EICU),establish and verify a dynamic web-based nomogram model.Methods The clinical data of 634 patients hospitalized in the EICU of Nanjing Medical University Affiliated Wuxi Second Hospital from January 2018 to December 2020 were retrospectively collected as the main cohort for modeling and internal validation.The information 189 patients admitted to the EICU of this hospital from January to December 2021 were enrolled as the validation cohort.Potential risk factors for early death in EICU patients were identified by univariate and multivariate logistic regression analysis,and a nomogram model was constructed.The receiver operator characteristic curve(ROC),C index,and calibration curve were conducted to evaluate the prediction efficiency of the nomogram model,and the decision curve analysis was operated to assess the clinical benefits of the model.External validation of the predictive model is performed in the validation cohort.Results Among 634 patients,61(9.62%)died within 3 days after admission.Multivariate logistic regression analysis revealed that NUTRIC score(OR:1.490,95%CI:0.436~2.116,P<0.001),NEWS score(OR:1.304,95%CI:1.113~1.528,P=0.001),acute respiratory distress syndrome(ARDS)(OR:2.183,95%CI:1.220~3.905,P=0.009),and mechanical ventilation(OR:2.320,95%CI:1.249~4.308,P=0.008)were significant independent risk factors for early death in EICU patients.Based on these results,a static and dynamic online nomogram was established(https://wxeysunyutinggrxjbk.shinyapps.io/DynNomEICUapp/).The C index and area under the curve(AUC)of the nomogram were 0.832 and 0.895,respectively.The calibration curve indicated that the predicted probability was in good consistency with the actual probability.The external validation of the model also presented good discrimination(C index=0.862,AUC=0.892)and predictive consistency.Decision curve analysis demonstrated the clinical benefit and application value of the prediction model.Conclusion This easy-to-use dynamic nomogram based on four independent risk factors has the preferable predictive performance for early death in EICU patients,which can contribute to accurately warning the severity and prognosis of patients,optimize clinical intervention strategies,and eventually improve the success rate for emergency treatment.