北京市孕妇妊娠糖尿病发病风险模型的构建与评估

Previous studies have shown that many environmental and genetic factors are associated with gestational diabetes mellitus, so a suitable risk assessment model based on genetic and environmental risk factors will be useful for early screening of High-risk people, and then health intervention can be adopted to reduce the incidence of gestational diabetes mellitus. Accordingly, a hospital based multicenter study will be carried out in pregnant women of Beijing, a total of 800 patients with gestational diabetes mellitus and 1600 controls will be chosen into the study. In the first stage of this study, the physical activity, diet, cognition and other information will be investigated by questionnaire, and the environmental risk factors will be screened by Logistic regression. The SNPs reported in GWAS will be screened by False Positive Report Probability method, followed by genotyping in laboratory, finally the genetic scoring method will be used to assess the genetic risk, and the generalized multifactor reduction method will be used to analyze the interaction of gene-environment. In the second stage of this study, the support vector machine method will be used to construct the risk model of gestational diabetes mellitus by using selected genetic and environmental risk factors, and the area under the ROC curve will be used to evaluate the prediction effect of the risk model. The results of this study can provide theoretical support for the systematic explanation of the etiology and pathogenesis of gestational diabetes mellitus, and the effect of gene-environment interaction on gestational diabetes mellitus will be clearly explained, which will beneficial for the development of individualized prevention and health management in the future.

妊娠糖尿病病因包括遗传和环境两方面，采用合适的风险评估工具，构建综合考虑遗传和环境危险因素的风险预测模型，能够早期筛查出高危人群，通过健康干预来减少妊娠糖尿病的发生。因此，本研究拟采用以医院为基础的多中心研究，在北京市收集800例妊娠糖尿病患者和1600例对照，第一阶段，通过问卷调查研究对象体力活动、饮食、认知等信息，并采用Logistic回归对环境危险因素进行筛选；通过假阳性报告概率法对GWAS报告的SNP位点进行筛选，开展实验室检测，并对SNP位点进行遗传风险评分；采用广义多因子降维法探讨基因-环境交互作用。第二阶段，利用筛选出的环境及遗传危险因素，采用支持向量机方法，构建妊娠糖尿病发病风险模型，并采用ROC曲线下面积对模型预测效果进行评估。本研究结果能够为系统阐明妊娠糖尿病的病因和发病机制提供理论支持，明确基因-环境交互作用对妊娠糖尿病的影响，为下一步进行个性化预防和健康管理奠定基础。

既往研究表明，GDM病因包括遗传和环境两个方面。构建综合考虑遗传和环境危险因素的风险评估模型，有利于早期筛查出高危人群，通过健康干预来减少妊娠糖尿病的发生。本研究采用3种不同的建模方法，分别构建包含环境危险因素、遗传风险评分、环境危险因素和遗传风险评分的GDM风险评估模型，并采用受试者工作特征曲线（ROC）的曲线下面积（AUC）指标评价模型构建的效果。结果显示，（1）Logistic回归模型：仅纳入环境危险因素时模型的ROC曲线下面积为0.719（95%CI：0.694 ~ 0.745），仅纳入遗传风险评分时模型的ROC曲线下面积为0.612（95%CI：0.588 ~ 0.635），同时纳入环境危险因素和遗传风险评分时模型的ROC曲线下面积为0.750（95%CI：0.725 ~ 0.774）。（2）神经网络模型：仅纳入环境危险因素时模型的ROC曲线下面积为0.811（95%CI：0.788 ~ 0.833）；仅纳入遗传风险评分时模型的ROC曲线下面积为0.612（95%CI：0.584~0.639）；同时纳入环境危险因素和遗传风险评分时模型的ROC曲线下面积为0.882（95%CI：0.866~0.898）。（3）支持向量机模型：仅纳入环境危险因素时模型的ROC曲线下面积为0.791（95%CI：0.711 ~ 0.870）；仅纳入遗传风险评分时模型的ROC曲线下面积为0.616（95%CI：0.535 ~ 0.697）；同时纳入环境危险因素和遗传风险评分时模型的ROC曲线下面积为0.804（0.735-0.874）。综上，本研究发现：（1）多个环境危险因素和单核苷酸多态性位点与GDM存在关联；（2）与只纳入环境危险因素或遗传风险评分相比，同时纳入环境危险因素和遗传风险评分时GDM发病风险评估模型的效果更好。（3）与Logistic回归模型相比，采用神经网络模型和支持向量机模型构建的GDM发病风险评估模型的效果更好。