2型糖尿病危险因素及患病风险预测模型研究

A Research on Risk Factors and Risk Prediction Models of Type 2 Diabetes Mellitus

探讨2型糖尿病发生的危险因素以及logistic回归、BP神经网络模型在其患病风险预测中的应用.作者回顾性收集糖尿病相关体检数据,分别应用logistic回归和BP神经网络建立2型糖尿病预测模型,通过受试者工作特征曲线(ROC)评价模型的预测性能.logistic回归分析结果显示,年龄、T2DM家族史、体质指数、甘油三酯、C反应蛋白是2型糖尿病发生的危险因素.研究结果表明,BP神经网络模型的预测准确率为88.6%,ROC曲线下面积为0.826(95%CI:0.816~0.835)优于logistic回归模型[准确率=81.8%、AUC(95%CI)=0.764(0.749~0.780)].因此,老年、肥胖、有糖尿病家族史、甘油三酯和C反应蛋白含量高的个体更容易患2型糖尿病;在2型糖尿病个体患病风险预测方面,BP神经网络模型要优于logistic回归模型.

This paper explores the risk factors of type 2 diabetes mellitus（ T2 DM） and the application of logistic regression and BP neural network in T2 DM risk prediction. DM-related physical examination data are collected and predictive models of type 2 diabetes mellitus are established by logistic regression and BP neural network respectively. And then predictive performance of the two models is evaluated by the receiver operating characteristic（ ROC） curve. Logistic regression analysis shows that age,T2 DM family history,body mass index（ BMI）,triglyceride and C-reactive protein are risk factors for T2 DM. Our research shows that,for BP neural network model,the accuracy and AUC（ 95% CI） are 88. 6%,0. 826（ 95% CI： 0. 816 - 0. 835） respectively,showing more accurate predictive performance than logistic regression model [accuracy = 81. 8%,AUC（ 95% CI） =0. 764（ 0. 749 - 0. 780） ]. Thus,people who have the factors including advanced age,obesity,T2 DM family history,high levels of triglyceride and C-reactive protein are prone to develop T2 DM,and the BP neural network model has a better performance than logistic regression model in predicting the risk of T2 DM.