摘要
目的基于人工神经网络(ANN)算法构建2型糖尿病发病风险预测模型,为预测一般人群的糖尿病发病风险提供依据。方法对象来自2011-2012年北京市房山区心脑血管慢性病队列调查,共纳入3 153名调查对象。基线调查和2014年的随访调查均由经过统一培训的调查员对调查对象进行问卷调查、体格检查和实验室检测。数据集按照7∶3的比例随机分为训练集和测试集。利用ANN预测一般人群2型糖尿病3年发病风险,模型纳入20项环境因素(包括研究对象的一般信息、既往病史和体检指标)以及28个单核苷酸多态性(SNP)位点信息,包含1个输入层、1个输出层、2个密集连接(dense)隐藏层,3个丢弃(dropout)层。结果模型在训练集中准确率为0.936,敏感度为0.783,特异度为0.892;在测试集中准确率为0.940,敏感度为0.721,特异度为0.882。特征重要度排名前5位的输入变量为空腹血糖(FPG)和口服葡萄糖耐量试验(OGTT)的检测结果。结论 ANN可以准确预测糖尿病发病风险,为识别糖尿病高危人群,给予早期干预提供依据。
Objective To construct the predicting model for type 2 diabetes mellitus(T2 DM) risk on the basis of artificial neural network(ANN) approach,and to provide the basis for predicting T2 DM risk in general population. Methods The subjects(3 153 cases) were from the cohort survey of chronic cardiovascular and cerebrovascular diseases in Fangshan district from 2011 to 2012.The basic investigation and following up investigation(2014) were performed by trained investigators with questionnaire,physical examinations and laboratory tests. The data sets were randomly divided into training and test sets. The T2 DM risk for 3 years in general population was predicted by ANN,the model included 20 environmental factors(the general information,disease history and physical examination indexes) and 28 single nucleotide polymorphism(SNP) site information(a input layer,an output layer,2 dense hidden layers and 3 dropout hidden layers). Results Concentration accuracy,sensitivity and specificity of model in the training set were 0.936,0.783 and 0.892,respectively and those of model in the test set were 0.940, 0.721 and 0.882,respectively.The input variables of top 5 feature importance were the results of fasting plasma glucose and oral glucose tolerance test.Conclusion ANN can accurately predict the risk of T2 DM,and provide the basis for recognizing high risk population or early intervention.
作者
车前子
郑启文
陈思
马雨佳
周泽宸
武轶群
吴涛
胡永华
陈大方
CHE Qian-zi;ZHENG Qi-wen;CHEN Si;MA Yu-jia;ZHOU Ze-chen;WU Yi-qun;WU Tao;HU Yong-hua;CHEN Da-fang(Department of Epidemiology and Biostatistics,School of Public Health,Peking University,Beijing 100191,China)
出处
《中国慢性病预防与控制》
CAS
CSCD
北大核心
2020年第4期274-279,共6页
Chinese Journal of Prevention and Control of Chronic Diseases
基金
北京市自然基金(7182085)
国家自然基金(81172768,81872692)。
关键词
人工神经网络
糖尿病
2型
风险预测
Artificial neural network
Diabetes mellitus,type 2
Risk prediction
