以机械功为导向预测机械通气患者撤机失败风险列线图模型的建立和验证——基于MIMIC-Ⅳ数据库的研究

Development and validation of a mechanical power-oriented nomogram model for predicting the risk of weaning failure in mechanically ventilated patients:an analysis using the data from MIMIC-Ⅳ

目的建立和验证以机械功(MP)为导向的机械通气患者撤机失败风险列线图预测模型。方法收集美国重症监护医学信息数据库-Ⅳ v1.0(MIMIC-Ⅳ v1.0)中有创机械通气(IMV)超过24 h且用T管通气策略进行撤机患者的临床资料,包括人口统计学信息和合并症、首次自主呼吸试验(SBT)前4 h的呼吸力学参数、SBT前的实验室指标、SBT期间的生命体征和血气分析、重症监护病房(ICU)住院时间及IMV时间,并将其纳入模型建立组。采用Lasso方法筛选影响撤机结局的风险因素,并纳入多因素Logistic回归分析,用R软件构建列线图预测模型并制作动态网页列线图。通过受试者工作特征曲线(ROC曲线)和校准曲线评估列线图的区分度和准确性,以决策曲线分析(DCA)评估列线图的临床有效性。前瞻性收集2021年11月至2022年10月在连云港市第一人民医院、连云港市第二人民医院ICU住院的机械通气患者数据,对该模型进行外部验证。结果在模型建立组中,共有3 695例机械通气患者纳入研究,其中撤机失败率为38.5%(1 421/3 695)。Lasso回归分析最终筛选出6个变量,包括呼气末正压(PEEP)、MP、动态肺顺应性(Cdyn)、吸入氧浓度(FiO2)、ICU住院时间、IMV时间,其系数分别为0.144、0.047、-0.032、0.027、0.090、0.098;Logistic回归分析显示,6个变量均为预测撤机失败风险的独立危险因素〔优势比(OR)和95%可信区间(95%CI)分别为1.155(1.111~1.200)、1.048(1.031~1.066)、0.968(0.963~0.974)、1.028(1.017~1.038)、1.095(1.076~1.113)、1.103(1.070~1.137),均P<0.01〕。以MP为导向的机械通气患者撤机失败风险列线图预测模型在模型建立组和外部验证组均表现出准确的区分度,ROC曲线下面积(AUC)和95%CI分别为0.832(0.819~0.845)、0.879(0.833~0.925),且其预测准确性显著高于MP、Cdyn、PEEP等单一指标。校准曲线表明预测与观察结局之间存在良好的一致性。DCA分析表明列线图模型具有较高的净获益值,在临床上是有益的。结论以MP为导向构建撤机失败风险的列线图预测模型,能够合理、准确地预测机械通气患者的撤机结局,有助于临床医生进行撤机决策的判断。

Objective To develop and validate a mechanical power(MP)-oriented nomogram prediction model of weaning failure in mechanically ventilated patients.Methods Patients who underwent invasive mechanical ventilation(IMV)for more than 24 hours and were weaned using a T-tube ventilation strategy were collected from the Medical Information Mart for Intensive Care-Ⅳv1.0(MIMIC-Ⅳv1.0)database.Demographic information and comorbidities,respiratory mechanics parameters 4 hours before the first spontaneous breathing trial(SBT),laboratory parameters preceding the SBT,vital signs and blood gas analysis during SBT,length of intensive care unit(ICU)stay and IMV duration were collected and all eligible patients were enrolled into the model group.Lasso method was used to screen the risk factors affecting weaning outcomes,which were included in the multivariate Logistic regression analysis.R software was used to construct the nomogram prediction model and build the dynamic web page nomogram.The discrimination and accuracy of the nomogram were assessed by receiver operator characteristic curve(ROC curve)and calibration curves,and the clinical validity was assessed by decision curve analysis(DCA).The data of patients undergoing mechanical ventilation hospitalized in ICU of the First People's Hospital of Lianyungang City and the Second People's Hospital of Lianyungang City from November 2021 to October 2022 were prospectively collected to externally validate the model.Results A total of 3695 mechanically ventilated patients were included in the model group,and the weaning failure rate was 38.5%(1421/3695).Lasso regression analysis finally screened out six variables,including positive end-expiratory pressure(PEEP),MP,dynamic lung compliance(Cdyn),inspired oxygen concentration(FiO2),length of ICU stay and IMV duration,with coefficients of 0.144,0.047,-0.032,0.027,0.090 and 0.098,respectively.Logistic regression analysis showed that the six variables were all independent risk factors for predicting weaning failure risk[odds ratio(OR)and 95%confidence interval(95%CI)were 1.155(1.111-1.200),1.048(1.031-1.066),0.968(0.963-0.974),1.028(1.017-1.038),1.095(1.076-1.113),and 1.103(1.070-1.137),all P<0.01].The MP-oriented nomogram prediction model of weaning failure in mechanically ventilated patients showed accurate discrimination both in the model group and external validation group,with area under the ROC curve(AUC)and 95%CI of 0.832(0.819-0.845)and 0.879(0.833-0.925),respectively.Furthermore,its predictive accuracy was significantly higher than that of individual indicators such as MP,Cdyn,and PEEP.Calibration curves showed good correlation between predicted and observed outcomes.DCA indicated that the nomogram model had high net benefits,and was clinically beneficial.Conclusion The MP-oriented nomogram prediction model of weaning failure accurately predicts the risk of weaning failure in mechanical ventilation patients and provides valuable information for clinicians making decisions on weaning.