血管性认知功能障碍高危因素的决策树模型研究  

Study on the decision tree model for risk factors of vascular cognitive impairment

作　　者：王潇[1] 郭宗君[1] 张文青[2] 吴钦娟[3] 于焕清 张凤香 肖琳 

机构地区：[1]青岛大学附属医院老年医学科,青岛266003 [2]莱州市人民医院神经内科,莱州261400 [3]莱芜市人民医院老年医学科,莱芜271100 [4]青岛市第五人民医院老年医学科,青岛266002 [5]青岛市中心医院重症医学科,青岛266042 [6]北京康复研究中心,北京100068

出　　处：《中华行为医学与脑科学杂志》2017年第6期534-538,共5页Chinese Journal of Behavioral Medicine and Brain Science

基　　金：基金项目：青岛市科技局基金项目（Kzd-03;09-1-1-33-nsh;KZJ-28;15-9-2-74,nsh）;青岛市黄岛区科技局计划项目（2014-1-73）

摘　　要：目的收集脑血管病患者人口学、生活模式和临床疾病因素，分析导致血管性认知功能障碍（VCI）的高危因素，建立VCI高危因素的决策树模型。方法选取2014年lO月至2016年10月于老年医学科和神经内科住院治疗脑血管病患者505例，进行人口学、生活模式和临床疾病因素问卷调查和数据采集，分为训练集组（421例）与测试集组（84例），其中训练集组患者分为非VCI患者组（225例）和VCI患者组（196例）。采用决策树算法分析脑血管病患者发生VCI的影响因素，并与Logistic回归分析和卡方检验比较，建立VCI高危因素的决策树模型。结果训练集组构建的决策树模型交叉验证模型识别准确度为73．63％，对测试集的预测准确度为73．81％。饮酒、业余爱好、饮茶、受教育程度、高血压、睡眠、年龄、饮食、糖尿病、体育锻炼是决策树模型的分类节点变量，饮酒作为根节点变量。卡方检验分析根据分类节点分类的患者，其VCI发病率差异有统计学意义（P〈0．05）。多因素Logistic回归分析显示，文化水平、饮酒、体育锻炼、糖尿病4个因素为VCI发生的影响因素，该模型预测准确度为66．98％，对测试集的预测准确度为53．57％。决策树模型与Logistic回归模型的ROC曲线显示，决策树模型AUC为0．737（95％CI：0．688—0．786），Logistic回归模型AUC为0．664（95％CI：0．612～0．717）。结论在对脑血管病患者发生VCI预测准确度方面，决策树模型优于Logistic回归模型。过量饮酒、糖尿病、高血压、高脂饮食、失眠是VCI发生的危险因素；业余爱好、高受教育程度、体育锻炼、饮茶是VCI的保护因素。Objective To collect the demographic,lifestyle and clinical factors of patients with ce- rebrovascular disease,and analyze the vascular cognitive impairment（YCf） factors and set up high-risk fac- tors model. Methods 505 patients with cerebrovascular disease hospitalization in department of geriatrics and neurology in hospital from October 2014 to October 2016 were enrolled.According to the questionaire sur- vey data of demographics, lifestyle and clinical factors, the patients were divided into training set （421 cases） and test set （84 cases） ,and training set were divided into the non-VCI set （225 cases） and VCI set （196 cases）. Analyzed the influence factors of VCI in patients with cerebrovascular disease by decision tree algo- rithm, and compared it with the Logistic regression analysis and chi-square and established the decision tree model for risk factors of VCI. Results According to the VCI decision tree model, cross validation model rec-ognition accuracy was 73.63%, while test set prediction accuracy was 73.81%. Alcoholism, hobbies, educa- tion level, tea drinking, diabetes, hypertension, diet, age, sleep and physical exercise were classification of node variables ,while drinking was the root. The probability of VCI had significant difference （P〈0.05） in the crowds with different risk factors. According to results of Logistic regression analysis, education level, drink- ing, exercise and diabetes were independent risk factors for VCI, while the model prediction accuracy was 66.98%, and test set prediction accuracy was 53.57%. According to the ROC curve of the decision tree model and the Logistic regression model,the decision tree model AUC was 0.737 （95%CI 0.688 to 0.786） , and the Logistic regression model AUC was 0.664 （95%CI 0.612 to 0.717）. Conclusion It is suggested that the decision tree model might be superior to logistic regression model in the prediction accuracy for VCI of pa- tients with cerebrovascular disease. The alcoholism, diabetes, high blood pressure, hig

关 键 词：血管性认知障碍 危险因素 决策树 预测模型 

分 类 号：R749.13[医药卫生—神经病学与精神病学]