摘要
【目的】探讨影响广州市中老年人群缺血性脑卒中发病的交互因素,综合应用分类树法和logistic回归分析构建脑卒中风险预测模型。【方法】采用前瞻性队列研究设计,分析2013年广州社区1130例中老年体检人群的队列资料;随访1年,主结局事件为发生缺血性脑卒中;应用分类树模型构建交互项,最终和筛选主效应一起纳入多因素非条件logistic回归进行模型拟合。【结果】分类树模型筛选得出8个主效应和变量之间的10个交互项,多因素logistic回归得出影响脑卒中发病的主因素,其OR值分别为：主效应有高血压4.003（95%CI：1.948～8.223）,重体力活（h/week）3.660（95%CI：2.203～6.079）;交互项有吸烟和饮酒5.622（95%CI：2.316～13.646）,吸烟和坐位时间（h/d）4.442（95%CI：2.720～7.253）;ROC曲线下面积为0.892（95%CI：0.842～0.943）,模型拟合比较稳定。仅按非条件logistic回归分析的方法,主危险因素和修正回归系数相近,ROC曲线下面积为0.753（95%CI：0.676～0.830）,与构建的交互项模型相比AUC偏小。其中Z=3.867,P〈0.001,差异有统计学意义。【结论】引入交互项后所建立的模型效果更好。广州中老年人群吸烟和饮酒、重体力活强度过高、高血压患者、坐位时间过长缺乏运动锻炼者会加大脑卒中发生概率,吸烟、饮酒和坐位时间（h/d）之间存在的交互作用会增加发病概率;通过模型可初步预测发病概率,进而指导采取针对性干预措施。
[Objective] To analyze the main risk interaction factors of ischemic stroke among elderly residents in Guangzhou by integrating logistic regression model and classification tree model. [ Methods ] Study on 1130 follow-up middle age people by using prospective cohort study. Follow-up time in this study was one-year, and main outcome event was occur ischemic stroke. Classification tree model was used to select interaction factors, and finally with the significant main variables to build logistic regression model. [ Results] The classification tree model selected eight main risk factors and ten interaction factors, and then chose these factors to build logistic regression. Main risk factors selected from logistic regression model were hypertension and heavy work （hours/week）, the ORs were 4.003 （95%CI： 1.948-8.223） and 3.660 （95%CI： 2.203-6.079）, respectively. Interaction risk factors showed that ORs of smoking and drinking was 5.622 （95%CI： 2.316-13.646）, the other one smoking and sitting time （h/d） was 4.442 （95%CI： 2.720-7.253）. The prediction model area under the ROC curve was 0.892 （95%CI： 0.842-0.943）, which meant the model fitting were stable and well. Compared to the analysis just according to the non-conditional logistic regression, the main risk factors and correction coefficient of regression are similar. The AUC of ROC was 0.753 （95% CI： 0.676-0.830）, which was smaller than the interaction model. The Z = 3.867, P 〈 0.001, which meant the difference of the two models have statistically significant. [ Conclusion ] The harmful behaviors among community elderly people in Guangzhou were smoking, drinking, high intensity working, and sitting for too long, especially those who were high blood pressure patients would have great risk. The interaction effects among smoking, drinking and sitting time （h/d） will increase the probability of ischemic stroke. Through the model can predict the stroke risk probability and take intervention measures about it.
出处
《中山大学学报（医学科学版）》
CAS
CSCD
北大核心
2016年第4期614-620,共7页
Journal of Sun Yat-Sen University:Medical Sciences
基金
广东省科技计划项目(2013B021800035)
关键词
缺血性脑卒中
危险因素
分类树
交互项构建
多因素logistic回归
ischemic stroke
risk factors
classification tree
construction of the interaction factors
multivariate logistic regression
