非急性期颈内动脉闭塞血管内再通治疗后成功再通的决策树预测模型

A decision tree model to predict successful endovascular recanalization of non-acute internal carotid artery occlusion

目的探讨非急性期有症状颈内动脉闭塞(symptomatic internal carotid artery occlusion,SICAO)血管内再通治疗后成功再通的预测因素,利用分类和回归树(classification and regression tree,CART)算法建立决策树模型并评价模型预测效能。方法回顾性纳入在中国8家综合卒中中心接受血管内再通治疗的非急性期SICAO患者,随机分配至训练集和验证集。在训练集通过最小绝对收缩与选择算子(least absolute shrinkage and selection operator,LASSO)算法筛选重要变量,基于CART算法构建决策树预测模型。在验证集中使用受试者工作特征(receiver operating characteristic,ROC)曲线、Hosmer-Lemeshow拟合优度检验以及混淆矩阵进行模型评价。结果最终纳入非急性期SICAO患者511例,按7∶3比例随机划分为训练集(357例)和验证集(154例),血管内再通治疗后成功再通率分别为58.8%和58.4%,差异无统计学意义(χ2=0.007,P=0.936)。采用LASSO回归筛选出的6个系数不为零变量构建CART决策树模型,最终决策树纳入5个变量,共5层,包含9条分类规则。闭塞节段数较少、近端残腔为锥形、ASITN/SIR侧支分级1~2级、缺血性事件为缺血性卒中以及最近事件至血管内再通治疗时间为1~30 d是成功再通的预测指标。ROC分析显示,决策树模型训练集曲线下面积为0.810(95%置信区间0.764~0.857),模型预测成功再通的最佳截断值为0.71;验证集曲线下面积为0.763(95%置信区间0.687~0.839),准确度为70.1%,精密度为81.4%,敏感性为63.3%,特异性79.7%。两组中Hosmer-Lemeshow检验均P>0.05。结论基于缺血性事件类型、最近事件至血管内再通治疗时间、近端残腔形态、闭塞节段数和ASITN/SIR侧支分级构建的决策树模型能有效预测非急性期SICAO血管内再通治疗后成功再通。

Objective To investigate predictive factors for successful endovascular recanalization in patients with non-acute symptomatic internal carotid artery occlusion(SICAO),to develop a decision tree model using the Classification and Regression Tree(CART)algorithm,and to evaluate the predictive performance of the model.Methods Patients with non-acute SICAO received endovascular therapy at 8 comprehensive stroke centers in China were included retrospectively.They were randomly assigned to a training set and a validation set.In the training set,the least absolute shrinkage and selection operator(LASSO)algorithm was used to screen important variables,and a decision tree prediction model was constructed based on CART algorithm.The model was evaluated using the receiver operating characteristic(ROC)curve,Hosmer-Lemeshow goodness-of-fit test and confusion matrix in the validation set.Results A total of 511 patients with non-acute SICAO were included.They were randomly divided into a training set(n=357)and a validation set(n=154)in a 7:3 ratio.The successful recanalization rates after endovascular therapy were 58.8%and 58.4%,respectively.There was no statistically significant difference(χ2=0.007,P=0.936).A CART decision tree model consisting of 5 variables,5 layers and 9 classification rules was constructed using the six non-zero-coefficient variables selected by LASSO regression.The predictive factors for successful recanalization included fewer occluded segments,proximal tapered stump,ASITN/SIR collateral grading of 1-2,ischemic stroke,and a recent event to endovascular therapy time of 1-30 d.ROC analysis showed that the area under curve of the decision tree model in the training set was 0.810(95%confidence interval 0.764-0.857),and the optimal cut-off value for predicting successful recanalization was 0.71.The area under curve in the validation set was 0.763(95%confidence interval 0.687-0.839).The accuracy was 70.1%,precision was 81.4%,sensitivity was 63.3%,and specificity was 79.7%.The Hosmer-Lemeshow test in both groups showed P>0.05.Conclusion Based on the type of ischemic event,the time from the latest event to endovascular therapy,proximal stump morphology,the number of occluded segments,and the ASITN/SIR collateral grading constructed the decision tree model can effectively predict successful recanalization after non-acute SICAO endovascular therapy.