妊娠期肝内胆汁淤积症胎儿风险评估模型的建立与应用研究

发布时间：2018-04-23 19:25

本文选题：妊娠 + 胆汁淤积　；参考：《浙江大学》2014年硕士论文

【摘要】：背景妊娠期肝内胆汁淤积症(intrahepatic cholestasis of pregnancy, ICP)是妊娠中晚期的特发性疾病,以皮肤瘙痒,胆汁酸升高及肝功能异常为临床特点。其对孕妇的影响极小,主要危及胎儿,可引起早产、宫内窘迫、羊水粪染,围产儿死亡等。ICP导致围产儿不良结局发生的机制目前尚未阐明,同时缺乏有效的防治措施,对胎儿威胁较大,也给广大产科临床工作者带来许多困惑。国内外学者已进行了一系列从实验室到临床的研究,帮助判断胎儿在子宫内的环境,但是准确性不甚理想。ICP发病时有众多因素可能与胎儿出现早产、窘迫、死胎死产等有关,所有单一因素均无法完全解释胎儿出现这些并发症。目前临床工作中没有一套有效的评估ICP孕妇胎儿宫内环境的方法。如果把评估ICP孕妇胎儿宫内环境状况的参数量化并进行有效地数据处理,相关结果用于胎儿风险评估,将有助于合理制订ICP治疗方案、及时选择终止妊娠时机。近年来,在医学领域兴起了采用人工神经网络来处理多参数的、复杂的、相互关联的问题,从中进行分析、推理、识别和预测,并已取得了一定的成效。因此,我们首次将人工神经网络引入妊娠期肝内胆汁淤积症胎儿风险预测体系中。目的通过ICP孕妇生化指标和临床资料作为参数建立预测妊娠期肝内胆汁淤积症胎儿风险的人工神经网络模型,并探讨其预测价值。方法选取203例在浙江大学医学院附属妇产科医院住院分娩的ICP孕妇作为研究对象,收集和筛选与胎儿风险相关的指标。构建联合参数法和生化参数法两种人工神经网络模型,按羊水混浊比例随机将其中135例作为训练集,另外68例作为测试集,分别计算预测胎儿风险的准确率,观测两种模型的敏感度和特异度,并绘制ROC曲线及计算曲线下面积分析预测效能。分析各个输入参数对预测结果的影响权重。结果 1、联合参数法建立的ANN模型的预测敏感度为80%,特异度为62.2%,准确率为66.2%；生化参数法建立的ANN模型的预测敏感度为73.6%,特异度为61.5%,准确率为64.7%。联合参数法ANN模型的准确率、敏感度及特异度均较高于生化参数法ANN模型。 2、联合参数法ANN模型的所有集、训练集及测试集的ROC曲线下面积分别为0.7991,0.8417,0.7100。生化参数ANN模型的所有集、训练集及测试的ROC曲线下面积分别为0.7714,0.8110,0.7036。两种模型的预测效能为中等。 3.妊娠合并症的影响系数最高(15.73%),其次为CG(14.63%),其余为发病孕周(11.35%),皮肤瘙痒病程(10.56%),DBIL(8.44%),S/D比值(7.24%),分娩孕周(5.67%),ALT(4.91%),年龄(4%),分娩方式(3.84%),胎数(3.78%),TBA(3.64%),NST分数(2.91%),羊水量(2.27%),AST(0.74%),宫缩、TBIL(0.14%)。结论 1、应用ANN技术可以全面客观评估ICP胎儿风险,但相关参数仍需进一步改进和完善。 2、联合参数法ANN模型的准确率、敏感度及特异度优于单纯生化参数法建立的ANN模型。 3、联合参数法中参数权重在10%以上的有：妊娠合并症(15.73%)、CG(14.63%)、发病孕周(11.35%)及皮肤瘙痒病程(10.56%)。
[Abstract]:Background

ICP ( ICP ) is an idiopathic disease in the middle and late stage of pregnancy . It is characterized by skin itching , bile acid elevation and abnormal liver function . The mechanism of ICP in pregnant women is very small , which can cause premature labor , intrauterine distress , amniotic fluid econium staining , perinatal death , etc . The mechanism of ICP leading to perinatal adverse outcome has not yet been clarified yet , meanwhile , there is a lack of effective control measures .

A series of studies from laboratory to clinic have been carried out by scholars at home and abroad to help judge the fetus ' s environment in the womb , but the accuracy is not ideal . There are many factors which may be related to the premature birth , distress , stillbirth and stillbirth of the fetus . There is no effective method to evaluate the intrauterine environment of the ICP pregnant woman .

In recent years , artificial neural networks have been used in the medical field to deal with multi - parameter , complex and interrelated problems , from which analysis , reasoning , identification and prediction have been made , and some results have been achieved . Therefore , we first introduced the artificial neural network into the fetal risk prediction system of intrahepatic calculosis of pregnancy .

Objective To establish an artificial neural network model for predicting the risk of fetal stasis in pregnancy during pregnancy by using biochemical indexes and clinical data of ICP pregnant women as parameters , and to discuss its predictive value .

method

In this paper , 203 ICP pregnant women who were hospitalized in the Affiliated Hospital of Zhejiang University Medical College were selected as the research subjects , and the indexes related to the risk of the fetus were collected and screened . 135 cases were randomly divided into training sets according to the proportion of amniotic fluid turbidity , and the sensitivity and specificity of the two models were calculated respectively .

Results

1 . The sensitivity of ANN model established by joint parameter method is 80 % , the specificity is 62.2 % , the accuracy rate is 66.2 % ;
The ANN model has a sensitivity of 73.6 % , a specificity of 61.5 % and an accuracy of 64.7 % . The accuracy , sensitivity and specificity of ANN model are higher than those of ANN model .

2 . The area of the ROC curves of all sets , training sets and test sets of the ANN model of the joint parameter method is 0.7991 , 0.8417 and 0.771 respectively . The area under the ROC curve of the ANN model is 0.7714 , 0.8110 and 0.7036 , respectively . The prediction efficiency of the two models is medium .

3 . The coefficient of pregnancy complications was the highest ( 15.73 % ) , followed by CG ( 14.63 % ) , the rest being the gestational week ( 11.35 % ) , the skin itching course ( 10.56 % ) , the birth control week ( 3.84 % ) , the fetal number ( 3.78 % ) , the TBA ( 3.64 % ) , the birth mode ( 2.91 % ) , the amniotic fluid volume ( 2.27 % ) , AST ( 0.74 % ) , uterine contraction , TBIL ( 0.14 % ) . Conclusion

1 . The risk of ICP fetus can be evaluated objectively by ANN technology , but the related parameters need to be further improved and improved .

2 . The accuracy rate , sensitivity and specificity of ANN model of combined parameter method are superior to ANN model established by simple biochemical parameter method .

3 . The weight of parameters in the joint parameter method was more than 10 % : pregnancy complications ( 15.73 % ) , CG ( 14.63 % ) , gestational week ( 11.35 % ) and skin pruritus ( 10.56 % ) .

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R714.255

【参考文献】