脑卒中患者静脉血栓栓塞症风险预测模型的构建及验证

Development and validation of a risk prediction model for venous thromboembolism in stroke patients

目的构建并验证脑卒中患者发生静脉血栓栓塞症(venous thromboembolism,VTE)的风险预测模型,为脑卒中患者VTE的预防控制提供科学依据。方法研究对象为河南省脑卒中队列的675例脑卒中患者,按7∶3随机划分为训练集(473例)和测试集(202例)。使用随机森林算法筛选变量、logistic回归模型分析方法构建模型,并绘制列线图。通过受试者工作特征曲线下面积(area under curve,AUC)、Hosmer-Lemeshow检验等评价模型的预测效能,使用决策曲线分析(decision curve analysis,DCA)评估模型的临床使用价值;并采用五折交叉验证对模型进行内部验证。结果最终纳入模型的预测因子为年龄、住院天数、日常生活活动能力(activity of daily living,ADL)、肌力、尿酸(uric acid,UA)、D-二聚体、纤维蛋白原(fibrinogen,Fib)和总胆固醇(total cholesterol,TC)。模型在训练集中的Hosmer-Lemeshow检验P=0.872,AUC=0.924(95%CI:0.898~0.950);测试集Hosmer-Lemeshow检验P=0.597,AUC=0.902(95%CI:0.852~0.951)。DCA曲线表明,模型在训练集和测试集中均具有较高的临床净获益。五折交叉内部验证结果显示,模型在训练集和测试集中的平均AUC分别为0.913和0.929。结论该研究构建的脑卒中患者VTE风险预测模型能有效预测VTE的发生,可为高风险患者早期识别和预防性治疗提供参考。

Objective To develop and validate the risk prediction model of venous thromboembolism(VTE)in stroke patients,so as to provide a scientific basis for the prevention and control of VTE.Methods A total of 675 stroke patients were enrolled from our stroke cohort of Henan Province.The data were randomly divided into a training(473 patients)and a testing dataset(202 patients)by a ratio of 7∶3.Then,we used a random forest algorithm for variable selection and logistic regression analysis to construct the model,and a nomogram was drawn.The prediction efficiency of the model was evaluated by the area under the receiver operating characteristic curve and Hosmer-Lemeshow test.Decision curve analysis(DCA)was used to evaluate the clinical application value of the model and the five-fold cross-validation was utilized to verify the model internally.Results The predictors that ultimately entered the prediction model were age,hospital stays,ADL,myodynamia,uric acid,D-dimer,fibrinogen,and total cholesterol.In the training dataset,the Hosmer-Lemeshow test yielded P=0.872 and the AUC was 0.924(95%CI:0.898-0.950).The testing dataset showed that the Hosmer-Lemeshow test yielded P=0.597 and the AUC was 0.902(95%CI:0.852-0.951).DCA curves indicated that the model had high clinical net benefits in both datasets.Internal verification presented that the average AUCs of the model in the training and testing datasets were 0.913 and 0.929,respectively.Conclusions The risk prediction model developed in this study can effectively predict VTE occurrence in stroke patients,offering a valuable tool for identifying high-risk individuals and implementing early preventive measures.