神经外科ICU病人呼吸机相关性肺炎风险预测模型的构建

Construction of risk prediction model of ventilator⁃associated pneumonia in neurosurgical ICU patients

目的:探讨神经外科ICU病人发生呼吸机相关性肺炎的危险因素,构建风险预测模型并进行验证。方法:选取2017年1月—2021年12月在湖州市2所三级综合性医院ICU住院的534例神经外科术后病人,分为建模组425例和验证组109例。通过单因素分析和Logistic回归分析筛选神经外科ICU病人发生VAP的危险因素,构建风险预测模型并绘制列线图。采用Bootstrap法对列线图模型进行内部验证,验证组数据对模型进行外部验证,最后通过临床决策曲线分析(DCA)评价模型的临床有效性。结果:最终纳入创伤性脑损伤、ICU住院时间、丙泊酚累积剂量、气管切开、深静脉置管、血清清蛋白值6个因素构建风险预测模型,该模型受试者工作特征(ROC)曲线下面积为0.966,灵敏度为89.9%,特异度为94.9%,最佳临界值为0.461。内部验证ROC曲线下面积为0.958,外部验证为0.934,Calibration图示预测概率与实际概率之间一致性较高,DCA决策曲线图表明模型具备较高的临床效益和实用价值。结论:构建的神经外科ICU病人呼吸机相关性肺炎风险预测模型具有良好的预测效能,可为临床预测神经外科ICU病人VAP发生风险提供参考。

Objective:To explore the risk factors of ventilator⁃associated pneumonia in neurosurgical ICU patients,and to construct and verify the risk prediction model.Methods:A total of 534 postoperative neurosurgery patients who were hospitalized in the ICU of two tertiary general hospitals in Huzhou city from January 2017 to December 2021 were selected,and divided into a modeling group of 425 cases and a validation group of 109 cases.The risk factors of VAP in neurosurgical ICU patients were screened by univariate analysis and logistic regression analysis,and risk prediction model was constructed and nomogram was drawn.The nomogram model was internally validated by Bootstrap method,and the model was validated externally by validation data.Finally,clinical decision curve analysis(DCA)was used to evaluate the clinical validity of the model.Results:Finally,6 factors including traumatic brain injury,ICU stay,cumulative dose of propofol,tracheotomy,deep vein catheterization,and serum albumin value were included to construct a risk prediction model.The area under the receiver operating characteristic(ROC)curve of the model was 0.966,the sensitivity was 89.9%,the specificity was 94.9%,and the optimal critical value was 0.461.The area under the ROC curve was 0.958 for internal validation and 0.934 for external validation.The Calibration chart showed that the predicted probability and the actual probability were highly consistent,and the DCA decision curve chart showed that the model had high clinical benefit and practical value.Conclusions:The risk prediction model of ventilator⁃associated pneumonia in neurosurgical ICU patients had good predictive performance,and which could provide reference for clinical prediction of the risk of VAP in neurosurgical ICU patients.