基于三种统计学方法构建的超重及肥胖人群高血压发病预测模型的分析比较  

Screening risk factors and interaction analysis of hypertension in overweight and obesity population based on three statistical models

作　　者：李禄伟 黄倩 施佳成 沈艳明 刘晓玲[1] 王彩梅[2] 于萍[1] 吴岚[3] 覃洋 江仁美[1] 于健[1] LI Lu-wei;HUANG Qian;SHI Jia-cheng;SHEN Yan-ming;LIU Xiao-ling;WANG Cai-mei;YU Ping;WU Lan;QIN Yang;JIANG Ren-mei;YU Jian(Department of Endocrinology,The Affiliated Hospital of Guilin Medical College,Guilin,Guangxi 541001,China;不详)

机构地区：[1]桂林医学院附属医院内分泌科,广西桂林541001 [2]桂林医学院附属医院检验科,广西桂林541001 [3]桂林医学院附属医院神经内科,广西桂林541001

出　　处：《现代预防医学》2021年第11期2061-2066,共6页Modern Preventive Medicine

基　　金：广西医疗卫生适宜技术开发与推广应用项目(S2019062);桂林市科学研究与技术开发计划项目(20190218-5-1)。

摘　　要：目的利用CRT分类树、logistic回归、BP神经网络构建超重及肥胖人群高血压发病预测模型。方法抽取出3150名超重及肥胖人群(体质指数≥24kg/m^(2))。分别应用CRT分类树、logistic回归、BP神经网络构建超重及肥胖人群高血压发病预测模型,筛选出高危因素,采用受试者工作特征曲线(ROC)对三种统计学方法构建的预测模型进行特异性、敏感性及准确性评估。结果三种方法构建的预测模型筛选出的高危因素包括非酒精性脂肪性肝病(NAFLD)、空腹血糖(FPG)、年龄、甘油三酯(TG)、尿酸(UA)、低密度脂蛋白胆固醇(LDL-c)。CRT分类树模型、logistic回归模型、BP神经网络模型ROC曲线下面积(AUC)值分别为0.721、0.734、0.733,敏感性分别为61.63%、76.59%、82.85%,特异性分别为77.58%、60.44%、52.00%,Youden指数分别为39.20%、37.02%、34.85%。结论本研究筛选的危险因素包括NAFLD、FPG、年龄、TG、UA、LDL-c,基于危险因素应用三种统计学方法构建的预测模型具有中等预测价值,对超重及肥胖人群高血压发病具有较好的预测能力。Objective To predict hypertension in the overweight and the obese population was constructed by using a CRT classification tree,Logistic regression,and BP neural network.Methods 3150 overweight and obese people(body mass index≥24 kg/m^(2))were selected.The prediction models of hypertension in overweight and obese populations were constructed by using CRT classification tree,Logistic regression,and BP neural network respectively.The high-risk factors were selected.The prediction models constructed by three statistical methods were evaluated by ROC.Results The high-risk factors selected by the three methods were NAFLD,FPG,Age,TG,UA,LDL-c.The AUC values of the ROC curve of CRT classification tree model,Logistic regression model and BP neural network model were 0.721,0.734 and 0.733,respectively,with sensitivity of 61.63%,76.59%,82.85%,specificity of 77.58%,60.44%,52.00%,under index 39.20%,37.02%and 34.85%,respectively.Conclusion The risk factors selected in this study include NAFLD,FPG,Age,TG,UA,LDL-c.The prediction model based on the three statistical methods has a medium predictive value and has a good predictive ability for hypertension in overweight and obese people.

关 键 词：超重及肥胖 高血压 预测模型 CRT分类树 LOGISTIC回归 BP神经网络 ROC曲线 

分 类 号：R544.1[医药卫生—心血管疾病] R589.2[医药卫生—内科学]