血管性认知障碍影响因素的决策树模型研究

发布时间：2018-06-06 03:36

本文选题：血管性认知障碍 + 决策树　；参考：《青岛大学》2017年硕士论文

【摘要】：目的收集脑血管病患者社会人口学、生活模式和临床疾病因素,分析导致血管性认知障碍(VCI),非痴呆型血管性认知障碍(VCIND),血管性痴呆(Va D)的影响因素,建立VCI、VCIND、Va D的影响因素模型。方法依据脑血管病诊断标准,选取2014年10月至2016年10月于本院老年医学科和神经内科住院治疗的脑血管病患者505例,进行社会人口学、生活模式和临床疾病因素问卷调查和数据采集,随机数字表法将患者分为训练集组(421例)与测试集组(84例),依据VCI等诊断标准将训练集患者分为非VCI组(225例)和VCI组(196例),VCI组又分为Va D组(98例)和VCIND组(98例),采用决策树方法和Logistic回归分析脑血管病患者发生VCI、VCIND、Va D的影响因素,建立发生VCI、VCIND、Va D的影响因素决策树模型和Logistic回归模型。结果1.通过训练集建立VCI决策树模型,其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%,模型稳定,拟合较好。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、体育锻炼是发生VCI的10个分类节点变量影响因素,根节点为饮酒。Logistic回归分析显示,饮酒、受教育程度、体育锻炼、糖尿病4个因素为脑血管病患者发生VCI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%。决策树模型的受试者工作特征曲线(AUC)为0.737(95%CI 0.688~0.786),Logistic回归模型的AUC为0.664(95%CI 0.612~0.717)。2.通过训练集建立VCIND决策树模型,其交叉验证模型识别准确度为81.73%,对测试集的预测准确度为58.49%。受教育程度、高血脂、业余爱好、糖尿病、睡眠、饮食、体育锻炼发生VCIND的7个分类点变量影响因素,根节点是受教育程度。Logistic回归分析显示,饮酒、受教育程度、体育锻炼、糖尿病4个因素为脑血管病患者发生VCIND的影响因素,其预测准确度为72.14%,用测试集数据检验模型,对测试集的预测准确度为54.72%。VCIND决策树模型的AUC为0.716(95%CI 0.648~0.784),Logistic回归模型的AUC为0.596(95%CI 0.525~0.666)。3.通过训练集建立的VaD模型,交叉验证模型识别准确度为71.43%,对测试集的预测准确度为69.42%,模型稳定,拟合较好。饮茶、业余爱好、睡眠、年龄、糖尿病、饮酒等影响Va D发生分类点变量影响因素,其中根节点为饮茶。Logistic回归分析结果显示,业余爱好、饮茶、睡眠3个变量是发生Va D的影响因素,该模型预测准确度为64.80%,对测试集的预测准确度为53.92%。决策树模型的AUC为0.714(95%CI 0.641~0.788),Logistic回归模型的AUC为0.648(95%CI 0.571~0.725)。结论1.在对不同程度的血管性认知障碍发生预测方面,决策树模型优于Logistic回归模型。2.过量饮酒、糖尿病、高血压、高脂饮食、失眠因素是脑血管病发生VCI的危险因素;有业余爱好、高教育水平、参加体育锻炼、饮茶是其保护因素。3.高血脂、糖尿病、失眠、高脂饮食是VCIND发生的危险因素;有业余爱好、高教育水平、参加体育锻炼是其的保护因素。4.失眠、高龄、有糖尿病、饮酒是影响脑血管病患者发生Va D的危险因素;饮茶、有业余爱好是其保护性因素。
[Abstract]:Objective to collect social demography, life pattern and clinical disease factors in patients with cerebrovascular disease, analyze the factors that lead to vascular cognitive impairment (VCI), non dementia vascular cognitive impairment (VCIND) and vascular dementia (Va D), and establish the influence factor model of VCI, VCIND, Va D. Methods according to the diagnostic criteria of cerebrovascular disease, select from October 2014 to 20 In October, 505 cases of cerebrovascular disease were hospitalized in the Department of geriatrics and neurology of our hospital. Social demography, life pattern and clinical disease factors were investigated and data collected. The patients were divided into training set (421 cases) and test set (84 cases) by random digital table, and the patients were divided into non - training sets according to VCI and other diagnostic criteria. In group VCI (225 cases) and group VCI (196 cases), group VCI was divided into Va D group (98 cases) and VCIND group (98 cases). The decision tree method and Logistic regression analysis were used to analyze the influencing factors of VCI, VCIND, Va D in patients with cerebrovascular disease, and the decision tree model and regression model were established. Results 1. set up the decision tree by training set. The accuracy of the cross validation model was 73.63%, the accuracy of the test set was 73.81%, the model was stable, and the fitting was good. Drinking, hobbies, drinking tea, education, hypertension, sleep, age, diet, diabetes, physical exercise were the 10 factors influencing the occurrence of VCI, and the root node was.Logistic regression. The analysis showed that drinking, education, physical exercise, and diabetes were 4 factors affecting the incidence of VCI in patients with cerebrovascular disease. The predictive accuracy of the model was 66.98%, and the predictive accuracy of the test set was 0.737 (95%CI 0.688~0.786) for the 53.57%. decision tree model (95%CI 0.688~0.786), and the AUC of the Logistic regression model was 0.66. 4 (95%CI 0.612~0.717).2. established the VCIND decision tree model through the training set. The accuracy of the cross validation model was 81.73%. The prediction accuracy of the test set was 58.49%. education, hyperlipidemia, hobbies, diabetes, sleep, diet, and physical exercise of the 7 classification point variables of VCIND. The root node was educated. .Logistic regression analysis showed that drinking, education, physical exercise, and diabetes were 4 factors affecting VCIND in patients with cerebrovascular disease, with a predictive accuracy of 72.14%, a test set data test model, and a AUC of 0.716 (95%CI 0.648~0.784) and Logistic regression for the prediction accuracy of the test set of the 54.72%.VCIND decision tree model (95%CI 0.648~0.784). The model's AUC is the VaD model of 0.596 (95%CI 0.525~0.666).3. through the training set. The accuracy of cross validation model is 71.43%, the accuracy of the test set is 69.42%, the model is stable, and the fitting is good. Tea drinking, hobbies, sleep, age, diabetes, drinking alcohol and so on influence the factors of the Va D classification point variables, among which the root node is the root node. The results of.Logistic regression analysis showed that 3 variables of hobby, drinking tea and sleep were the factors affecting Va D. The prediction accuracy of the model was 64.80%, the prediction accuracy of the test set was 0.714 (95%CI 0.641~0.788) for the 53.92%. decision tree model (95%CI 0.641~0.788), and AUC (95%CI 0.571~0.725) of the Logistic regression model was 0.648 (95%CI 0.571~0.725). The conclusion was 1. in the conclusion. The decision tree model is better than the Logistic regression model.2. excessive drinking, diabetes, hypertension, high fat diet, and insomnia factors are the risk factors for the occurrence of VCI in cerebrovascular disease, and there are hobbies, high education, participation in physical training, and tea drinking is the protective factor of.3. hyperlipidemia, diabetes, and loss. Sleep, high fat diet is a risk factor for the occurrence of VCIND; hobbies, high education, and physical exercise are the protective factors of.4. insomnia, age, diabetes, and drinking are the risk factors for Va D in patients with cerebrovascular disease; tea drinking and hobbies are protective factors.
【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R749.1

【参考文献】