急性缺血性脑卒中患者卒中相关感染风险预测模型构建与验证

Construction and validation of risk prediction model for stroke-related infection in patients with acute ischemic stroke

摘要:

目的 探讨急性缺血性脑卒中患者发生卒中相关感染的危险因素，并构建决策树预测模型。 方法 回顾性选取2020年6月—2021年6月某院神经内科病房收治的急性缺血性脑卒中患者为研究对象。将其以一定比例分配为训练组与验证组。通过Lasso回归筛选预测因子，基于CHAID算法构建急性缺血性脑卒中患者卒中相关感染的决策树模型。内部验证采用随机拆分验证法，使用受试者工作特征(ROC)曲线下面积(AUC)对模型效果进行评价。 结果 共收治693例AIS患者，训练组484例，验证组209例。训练组卒中相关感染发病率为17.8%(86例)，验证组卒中相关感染发病率为20.1%(42例)。年龄、空腹血糖、糖尿病史、甘油三酯、吸烟、合并呼吸系统疾病、合并心血管系统疾病、意识障碍、住院时长是急性缺血性脑卒中患者发生卒中相关感染的危险因素。将以上因素纳入并构建决策树模型，决策树模型包含3层，共7个节点。合并呼吸系统疾病、糖尿病史、吸烟是发生卒中相关感染的预测指标。验证组决策树模型ROC的AUC为0.980，灵敏度为97.0%，特异度为97.6%，Youden指数为0.946，Kappa值为0.914。 结论 本研究构建的模型可以较好的预测急性缺血性脑卒中患者发生卒中相关感染的风险，可作为临床护理人员对患者进行风险预测的评估工具。

Abstract:

Objective To explore the risk factors for the occurrence of stroke-related infection in patients with acute ischemic stroke (AIS), and construct a decision tree prediction model. Methods AIS patients admitted to the department of neurology of a hospital from June 2020 to June 2021 were retrospectively selected as the research objects. They were divided into training group and validation group in a certain proportion. The predictors were screened by Lasso regression, and a decision tree model for stroke-related infection in AIS patients was constructed based on the CHAID algorithm. Random split validation method was adopted for internal validation, and the area under the receiver operating characteristic (ROC) curve (AUC) was used to evaluate the effect of the model. Results A total of 693 AIS patients were treated, 484 in training group and 209 in validation group. Incidence of stroke-related infection in training group and validation group were 17.8% (n=86) and 20.1% (n=42) respectively. Age, fasting blood glucose, history of diabetes, triglycerides, smoking, complicated respiratory diseases, complicated cardiovascular diseases, disturbance of consciousness, and long length of hospitalization were risk factors for stroke-related infection in AIS patients. The above factors were included and a decision tree model was constructed. The decision tree model contained 3 layers and a total of 7 nodes. Complicated respiratory disease, history of diabetes, and smoking were predictors of stroke-related infection. The AUC of ROC of validation group decision tree model was 0.980, the sensitivity and specificity were 97.0% and 97.6% respectively, Youden index was 0.946, Kappa value was 0.914. Conclusion The model constructed in this study can better predict the risk of stroke-related infection in AIS patients, and can be used as an evaluation tool for clinical nurses to predict the risk of patients.