BP人工神经网络应用于探讨2型糖尿病/糖耐量低减的发病危险因素

The Application Study of BPANN on Etiology of DM/IGT

目的探讨 BP人工神经网络 (BPANN)在 2型糖尿病 (2型 DM) /糖耐量低减 (IGT)的发病危险因素研究中的应用特点。方法采用 BPANN和多因素非条件 L ogistic回归方法对某矿区 2型 DM现况调查资料及在某综合性医院收集的病例 -对照资料进行分析比较。结果针对 2型 DM和 IGT,BPANN网络结构分别为 32→ 12→ 1,2 0→ 6→ 1。比较网络输出的平均影响值 (MIV)与 L ogistic回归分析结果 ,发现两者均选出母亲 DM史、腰臀比 (WHR)、同胞 DM史、离退休者、高梁等为影响 2型 DM发病的主要因子 ;标准体重百分数 (SWP)、子女 DM史、职业体力活动、年龄和脉率是影响IGT发病的主要因子。结论 BPANN同样能够用于 2型 DM/ IGT的发病危险因素研究 ;BPANN分析不要求资料的分布类型以及变量的独立性 ,能较好地处理数据协变量间的共线性问题 ,在病因研究中具有 L

Objective To explore the applied characteristics of BPANN on the study of risk factors for type 2 DM/IGT.Method BPANN was applied to analyze the data from the prevalence survey in a mining district and from the case control study in a general hospital,and compared with the results from Logistic regression analysis.Results The structures of BPANN were 32→12→1 to type2 DM,and 20→6→1 to IGT.With both the BPANN and Logistic regression analyses,DM history of mother,WHR,the retiree,DM history of brethren,and durra were selected as the leading risk factors for DM;and SWP,DM history of children,occupational physical activity,age and pulse for IGT.Conclusions BPANN also can be utilized in studying the risk factors for type 2 DM/IGT.BPANN needn't variables to meet normal distribution and be independence,and can deal with the collinearity between covariables preferably.It has the unique advantage over the Logistic regression analysis.