内科

目的比较列线图、决策树和随机森林法构建的风险预测模型对急性脑梗死（ACI）患者住院期间并发肺部感染的预测效果。方法回顾性分析2021年12月至2022年10月在广西医科大学第一附属医院治疗的380例ACI患者的临床资料，根据住院期间是否并发肺部感染，将患者分为感染组（n=97）和未感染组（n=283）。采用单因素分析和多因素logistic回归模型筛选ACI患者住院期间并发肺部感染的影响因素，分别应用列线图、决策树、随机森林法构建风险预测模型，采用准确度、灵敏度、特异度、召回率、精确率、受试者操作特征（ROC）曲线下面积（AUC）评价三种模型的预测效果。结果感染组与未感染组的吸烟史、吞咽障碍、格拉斯哥昏迷评分（GCS）、入院时白细胞计数差异均有统计学意义（均P<0.05）。多因素logistic回归分析结果显示，存在吞咽障碍、有吸烟史，GCS评分低、入院时白细胞计数高均是ACI患者住院期间并发肺部感染的独立危险因素（均P<0.05）。对于ACI患者住院期间并发肺部感染风险的预测，列线图模型和随机森林模型的ROC AUC差异无统计学意义［0.898（0.819～0.977）比0.908（0.841～0.974）］（P>0.05），但两者的ROC AUC均大于决策树模型［AUC=0.797（0.693～0.901）］（均P<0.05）；列线图模型的准确度、灵敏度、特异度、召回率、精确率均高于/等于其他两个模型。结论基于吞咽障碍、吸烟史、GCS评分与入院时白细胞计数这四个常见临床指标构建的列线图模型和随机森林模型，在预测ACI患者住院期间并发肺部感染风险中具有较好的效果，两者的预测效能均优于决策树模型。

【Abstract】 Objective To compare the prediction effects of risk prediction models, constructed by nomograms, decision trees, and random forests, for complicated pulmonary infection during hospitalization in patients with acute cerebral infarction (ACI). Methods Clinical data of 380 ACI patients treated in the First Affiliated Hospital of Guangxi Medical University from December 2021 to October 2022 were retrospectively analyzed. According to whether they had complicated pulmonary infection during hospitalization, the patients were divided into an infected group (n=97) or an uninfected group (n=283). Univariate analysis and multivariate logistic regression model were used to screen the influencing factors for complicated pulmonary infection during hospitalization in ACI patients. Risk prediction models were constructed by three statistical methods, nomograms, decision trees, and random forests, respectively, and their prediction effects were evaluated by accuracy, sensitivity, specificity, recall, precision, and area under the receiver operating characteristic (ROC) curve (AUC). Results There were statistically significant differences in the smoking history, dysphagia, Glasgow coma scale (GCS) score, and white blood cell count at admission between the infected and uninfected groups (all P<0.05). Results of multivariate logistic regression analysis showed that dysphagia, smoking history, a low GCS score, and a high white blood cell count at admission were independent risk factors for complicated pulmonary infection during hospitalization in ACI patients (all P<0.05). For predicting the risk of complicated pulmonary infection during hospitalization in ACI patients, there was no statistically significant difference in the ROC AUC between the nomogram model and the random forest model [0.898(0.819-0.977) vs 0.908(0.841-0.974)] (P>0.05), but the ROC AUCs of the two abovementioned models were greater than that of the decision tree model [AUC=0.797(0.693-0.901)] (all P<0.05); the accuracy, sensitivity, specificity, recall, and precision of the nomogram model were higher than or equal to those of the others. Conclusion Nomogram and random forest models constructed on four common clinical indexes, including dysphagia, smoking history, GCS score, and white blood cell count at admission, have superior efficacy in predicting the risk of complicated pulmonary infection during hospitalization in ACI patients, and their prediction effects are better than the decision tree model's.