摘要目的:构建机械通气（mechanical ventilation, MV）患者呼吸机相关性肺炎（ventilator-associated pneumonia, VAP）风险预测列线图模型，并验证模型预测性能的稳定性。方法:根据研究对象入院顺序，回顾性选取2019年1月至2022年12月入住宁夏医科大学总医院重症医学科的MV患者为研究对象。根据是否发生VAP将MV患者分为非VAP组与VAP组，收集两组一般资料、疾病、用药、病情及操作指标等临床资料作为模型候选预测因子进行组间比较。使用多因素Logistic逐步向前回归分析筛选最终进入模型的预测因子，并构建列线图模型。通过受试者工作特征曲线下面积（area under the curve, AUC）评价模型区分度，计算模型在预测临界值下的诊断性试验结果，Hosmer-Lemeshow检验评价模型拟合度，采用Bootstrap重抽样1 000次进行内部验证，通过校准曲线和决策分析曲线分别评价模型校准度和临床适用度。结果:本研究共纳入1 250例MV患者，其中非VAP组1 102例，VAP组148例，VAP发病率为11.8%。检出多重耐药菌、慢性肾脏疾病、颅脑损伤、氧合指数、气管插管场所、再次气管插管、使用纤维支气管镜、使用抗菌药物情况和MV时间为发生VAP的模型预测因子。列线图模型AUC为0.917（95% CI: 0.895~0.939），最大约登指数0.697对应预测截断值为0.096，模型准确度为0.836，敏感度为0.865，特异度为0.832，阳性预测值及阴性预测值分别为0.409、0.979。Hosmer-Lemeshow检验提示模型拟合程度良好（ P=0.938）。内部验证结果显示，校准曲线的模型预测风险与实际风险基本一致，决策分析曲线的决策阈值概率为2%~90%。 结论:本研究构建的列线图模型简洁便利且具有较为稳定的预测性能，可开展外部验证评价模型可外推性，为临床个体化预测MV患者VAP发生风险提供依据。

abstractsObjective:To develop a nomogram model for predicting the risk of ventilator-associated pneumonia (VAP) in patients with mechanical ventilation (MV) and to validate the stability of the prediction performance of the model.Methods:The patients with MV admitted to the Department of Critical Care Medicine of General Hospital of Ningxia Medical University from January 2019 to December 2022 were retrospectively selected according to the order of admission. The patients with MV were divided into the non-VAP group and the VAP group according to whether VAP occurred. The clinical data of the two groups, including general information, disease, medication, condition, and operation-related indicators were collected as candidate predictors of the model for comparison. Multivariate logistic stepwise forward regression analysis was used to screen the predictors that finally entered the model, and a nomogram model was constructed. The model discrimination was evaluated by the area under the receiver operating characteristic curve (AUC), the diagnostic test results of the model at the predicted threshold were calculated, the Hosmer-Lemeshow test was used to evaluate the model fit, and the Bootstrap resampling was used 1 000 times for internal validation, and model calibration and clinical applicability were evaluated by calibration curve and decision analysis curve, respectively.Results:A total of 1 250 patients with MV were included, including 1 102 patients in the non-VAP group and 148 patients in the VAP group, and the prevalence of VAP was 11.8%. The detection of multidrug-resistant organisms, chronic kidney disease, brain injury, oxygenation index, the place of tracheal intubation, reintubation, use of bronchoscopy, use of antibiotics, and MV duration were model predictors of VAP. The AUC of the nomogram model was 0.917 (95% CI: 0.895-0.939), the maximum Youden index of 0.697 corresponded to a prediction threshold of 0.096. The model accuracy, sensitivity and specificity were 0.836, 0.865, and 0.832, respectively. The positive predictive value and the negative predictive value were 0.409 and 0.979, respectively. The Hosmer- Lemeshow test indicated that the model fit well ( P=0.938). The results of the internal validation of the model showed that the predicted risk of the calibration curve was generally consistent with the actual risk, and the decision threshold probability of the decision analysis curve ranged from 2% to 90%. Conclusions:The nomogram model developed in this study is simple, convenient and has relatively stable prediction performance, which can be externally validated to evaluate the extrapolation of the model, and provide a basis for individualized clinical prediction of the risk of VAP in patients with MV.