广西医学

目的构建梯度提升机（GBM）模型预测呼吸重症监护病房（RICU）机械通气患者并发呼吸机相关性肺炎（VAP）的短期预后情况。方法回顾性分析RICU收治的350例机械通气并发VAP患者的临床资料，根据随访结局将患者分为死亡组（n=110）和生存组（n=240），筛选影响其预后的危险因素。按8∶2比例将患者随机分为训练集280例和验证集70例，采用R语言4.2.1软件构建GBM预测RICU机械通气并发VAP患者的短期预后模型并评价模型的预测效能。结果年龄、机械通气时间、C反应蛋白（CRP）水平、血清降钙素原水平、急性生理学与慢性健康状况评分系统Ⅱ（APACHEⅡ）评分及序贯器官衰竭评估（SOFA）评分是RICU机械通气并发VAP患者预后的影响因素（P<0.05）。基于上述影响因素构建的GBM模型在训练集中的受试者工作特征曲线下的面积（AUC）为0.926（95% CI：0.894，0.958），敏感度为85.4%、特异度为86.2%，在验证集中的AUC为0.880（95% CI：0.779，0.980），敏感度为85.7%、特异度为86.4%；校准曲线显示，GBM模型的预测概率与实际发生率基本一致；决策曲线分析显示，训练集和验证集的阈概率分别在0.10~0.98、0.10~0.80。结论基于年龄、机械通气时间、CRP水平、血清降钙素原水平、APACHEⅡ评分及SOFA评分构建的GBM模型对RICU机械通气并发VAP患者的短期预后情况有较好的预测价值。

Objective To establish the gradient boosting machine (GBM) model for predicting short⁃term prognosis of patients with mechanical ventilation and concomitant ventilator⁃associated pneumonia (VAP) in respiratory intensive care unit (RICU). Methods The clinical data of 350 patients with mechanical ventilation and concomitant VAP admitted to RICU were retrospectively analyzed, and they were divided into death group (n=110) or survival group (n=240) according to patients' follow⁃up outcome. The risk factors affecting their prognosis were screened. According to the ratio of 8 : 2, patients were randomly assigned to training set (280 cases) or validation set (70 cases). The R language 4.2.1 software was used to establish a model of GBM for predicting short⁃term prognosis in patients with mechanical ventilation and concomitant VAP in RICU, and prediction efficiency of the model was evaluated. Results Age, mechanical ventilation duration, C⁃reactive protein (CRP) level, serum procalcitonin level, Acute Physiology and Chronic Health Evaluation Ⅱ (APACHEⅡ) score, and Sequential Organ Failure Assessment (SOFA) score were the influencing factors for prognosis of patients with mechanical ventilation and concomitant VAP in RICU (P<0.05). Area under the curve (AUC) of receiver operating characteristic for the GBM model established based on the aforementioned influencing factors in the training set was 0.926 (95% CI:0.894, 0.958), the sensitivity was 85.4%, and the specificity was 86.2%; furthermore, AUC in the validation set was 0.880 (95% CI: 0.779, 0.980), the sensitivity was 85.7%, and the specificity was 86.4%. The calibration curve revealed that the predicted probability of GBM model was basically consistent with the actual incidence rate. The decision curve analysis indicated that the threshold probabilities of the training set and the validation set were 0.10-0.98 and 0.10-0.80, respectively. Conclusions The GBM model established based on age, mechanical ventilation duration, CRP level, serum procalcitonin level, APACHEⅡ score and SOFA score exerts a favorable predictive value for short⁃term prognosis of patients with mechanical ventilation and concomitant VAP in RICU.