摘要
目的构建2型糖尿病(type 2 diabetes mellitus,T2DM)5年发病风险(个体化)预测模型。方法选择1997-2006年首次参加中国台湾美兆自动化健检机构(简称美兆健检)的35～74岁人群24 220人,将基线未患T2DM者23 180人分为建模队列(用于建立5年发病预测模型)和验证队列(用于评估模型外部效应)。采用Logis-tic回归构建预测模型,以受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under roc curve,AUC)评价拟合优度并将人群的预测风险概率进行风险等级划分。结果去除基线患者后T2DM 5年发病率为3.25%(753/23 180)。纳入5年发病风险预测模型的变量包括年龄、有无T2DM家族史、身高、腰围、空腹血糖、甘油三酯、高密度脂蛋白胆固醇7个指标。建模队列建立的预测模型的ROC曲线下面积(AUC)约为0.873(95%CI:0.851～0.893),验证队列外部效度验证结果:桃园、台中、高雄3个体检中心的AUC依次为0.823、0.850和0.819。将建模队列划分为4个风险等级后,显示中危(占13.4%)和高危(占4.3%)的个体5年内发生T2DM的危险分别比一般人群高5.9倍和29.8倍。结论利用中国台湾美兆健检纵向数据资料建立的T2DM 5年个体风险预测模型效应与信度均较高,且简单实用,对于今后T2DM风险个体评价及群体监测均有较大的应用价值。
Objective This study aimed to provide the epidemiological model for evaluating the risk of developing T2DM within 5 years in Taiwan population aged 35-74 years. Methods After excluding 1 040 individuals who were T2DM at baseline, a cohort of about 23 180 subjects aged 35-74 years, who were nonT2DM in initial year during 1997-2006, were selected to derive a training cohort（ for risk functions） and a testing cohort （for evaluation）. We produced risk functions u- sing Logistic regression models. We evaluated model predictability by the area under the receiver-operating characteristic （ROC） curve （AUC） and testified its diagnostic property on the testing sample and then established rules to characterize 4 different degrees of risk based on cut points of these probabilities after transforming into normal distribution by log-transfor-mation. Results After excluding 1 040 T2DM individuals at baseline, we ascertained 753 （3.25% , 753/23 180） cases of incident T2DM. Final multivariable Logistic regression model included seven risk factors： age, history of diabetes, height, waist circumference, fasting plasma glucose, high density lipoprotein and triglyeeride. The area under the ROC curve （AUC） was 0. 873 （95% CI： 0. 851-0. 893 ）predicting T2DM within 5 years, and also had adequate performance in three testing samples （AUC range： 0. 819-0. 850）. We found after labeling four risk degrees, there were 13.4% and 4.3% of total subjects were for mediate and high risk population separately, and a higher 5.9 and 29.8 times when compa- ring with common risk population. Conclusions The predictability and reliability of our T2DM risk score model, derived based on Taiwan MJ Longitudinal health-checkup-based Population Database, were enough better and satisfied, and were simple and practicable. We think the model can help both individual and community surveillance for the development trend of T2DM operatively.
出处
《中华疾病控制杂志》
CAS
北大核心
2013年第5期369-373,共5页
Chinese Journal of Disease Control & Prevention
