基于贝叶斯网络学习的胎儿心脏病非遗传相关因素分析

Analysis of non-genetic related factors of fetal heart disease based on Bayesian network learning

目的本研究旨在基于贝叶斯网络的大数据分析探讨胎儿心脏病(FHD)的相关因素,并定量分析从单因素到多因素累积暴露的对FHD的相对风险比。方法连续入组从2010年6月至2018年7月于首都医科大学附属北京安贞医院母胎医学会诊中心接受胎儿超声心动图检查的孕妇(n=16086,包括孕有异常胎儿的孕妇3312例),获得26个孕妇和胎儿因素,包括年龄,合并症,药物暴露,引产史和先天性心脏病病史,近亲结婚,孕妇和配偶的不良习惯,以及是否是双胎及是否伴有心律失常,基于所有变量构建贝叶斯网络,并通过联合树推理算法,预测两组的胎儿CHD患病率,从而得到不同暴露因素组合条件下的胎儿CHD的风险比(RR)。考虑到孕周对模型预测准确度的提升作用,将孕周分组进行了敏感性分析。结果单因素分析显示,双胎妊娠、自发流产、配偶吸烟的RR分别为1.50、1.38、1.11;当多个因素结合在一起时风险逐渐升高。当是否为双胎与流产史或吸烟配偶相结合时,我们发现FHD的RR更高(RR=1.96或1.63)。以此类推,当同时存在五个因素时,包括是否为双胎、孕早期上呼吸道感染、孕妇精神压力、贫血以及自发流产史或配偶吸烟时FHD风险高于小于5个的因素组合(RR=2.62或2.28)。除上述因素外,其他因素不会继续增加FHD风险。我们进一步根据孕周分组(A组:≥16周,<28周;B组≥28周,<40周)进行敏感性分析,结果发现趋势同上。结论基于贝叶斯网络学习结果,我们发现与FHD直接相关因素包括自发流产,妊娠早期上呼吸道感染,贫血和孕妇精神压力,以及双胎和吸烟配偶。上述组合因素越多,FHD的风险就越高。这些发现提示对存在这些危险因素的孕妇加强管理及产前咨询是非常重要的。

Objective The purpose of this study is to explore the related factors of fetal heart disease(FHD) based on big data analysis of the Bayesian network and to quantitatively analyze the relative risk ratio of a single factor to multi-factor cumulative exposure to FHD. Methods Pregnant women who underwent fetal echocardiography at the maternal-fetal medical consultation center of Beijing Anzhen Hospital Affiliated to Capital Medical University from June 2010 to July 2018(n=16 086, including 3312 pregnant women with abnormal fetuses) obtained 26 maternal and fetal factors, including age, complications, drug exposure, history of induced labour and congenital heart disease, and close relative marriage, According to the bad habits of pregnant women and spouses, whether twins and arrhythmias are accompanied, a Bayesian network is constructed based on all variables, and the prevalence of fetal CHD in the two groups is predicted through the joint tree reasoning algorithm, to obtain the risk ratio(RR) of fetal CHD under different combinations of exposure factors. Considering the effect of gestational age on the model’s prediction accuracy, we grouped gestational age into groups for sensitivity analysis. Results Univariate analysis showed that the RR of twin pregnancy, spontaneous abortion and spouse smoking were 1.50, 1.38 and 1.11, respectively;When multiple factors are combined, the risk increases gradually. We found that the RR of FHD was higher when twins were combined with a history of abortion or a smoking spouse(RR=1.96 or 1.63). By analogy, when there are five factors at the same time, including twins, upper respiratory tract infection in early pregnancy,the mental stress of pregnant women, anemia, history of spontaneous abortion or spouse smoking, the risk of FHD is higher than the combination of fewer than five factors(RR=2.62 or 2.28). In addition to the above factors, other factors will not continue to increase the risk of FHD. We further grouped according to gestational weeks(group A:≥16 weeks, <28 weeks;Sensitivity analysis was performed in group B(≥28 weeks, <40 weeks). The results showed that the trend was the same as that in group B. Conclusion Based on the Bayesian network learning results, we found that the factors directly related to FHD included spontaneous abortion, upper respiratory tract infection in early pregnancy, anemia and maternal stress, as well as twins and smoking spouses. The more these combined factors, the higher the risk of FHD. These findings suggest that it is essential to strengthening pregnant women’s management and prenatal consultation with these risk factors.