The Establishment Of The Risks Assessment Model For Fetus In Intrahepatic Cholestasis Of Pregnancy

Background:Intrahepatic cholestasis of pregnancy (ICP),a pregnancy-specific disorder during the late second or third trimester, is characterized by pruritus, elevated serum bile acids and abnormal liver tests.It poses little risk on the mother but significantly increases risks to the fetuses of adverse perinatal outcome, including spontaneous preterm labor, fetal distress, meconium-staining of amniotic fluid, stillbirth. The underlying mechanisms associated with poor fetal outcome are largely unknown and the effective preventions are scarce,so it threatens fetuses seriously and makes obstetricians feel very confused.Although a series of laboratorial and clinical researches had done by domestic and overseas scholars are used to estimate the risks of fetus in uterus, the results are still not deterministic. Many factors that may be associated with fetal preterm birth, distress, stillbirth, all of the single factor can’t consistently identify these complications occurred.As yet,there has been no effective method for evaluating intrauterine environment of ICP fetus in the clinic.If quantize the parameters of risks assessment and handle them effectively,that will be helpful to formulate the treatment of ICP and choose the best time to termination of pregnancy.In recent years, there has been prevalence of using artificial neural network to deal with multifactorial,intricate and interconnected data for analysis, reasoning, recognition, and prediction in the medical science.Moreover, it has achieved certain results. Therefore, we firstly introduce the artificial neural network to the prediction system of fetus risks in intrahepatic cholestasis of pregnancy.Objective:To generate and evaluate artificial neural network (ANN) models from clinical and laboratorial criteria to predict the risk of adverse pregnancy and fetal complications in ICP.Methods:The study was used to retrospectively investigate the date of203ICP women delivered in women’s hospital school of medcine zhejiang university. Collect and screen criteria associated with the fetal risks. Generate two ANN models based on the combined parameters and biochemical parameters separately.135cases are divided as a training set and the other68cases as a test set randomly according to the proportion of meconium-staining of amniotic fluid cases. Calculate the predicting accuracy and observe the sensitivity and specificity of the two models.In addition, receiver operating characteristic (ROC) curve and the area under curve(AUC) analysis are performed.At last, analyze the effect weight of every input parameters.Results:1. The sensitivity, specificity and accuracy of ANN based on the combined parameters are80%,62.2%, and66.2%. The sensitivity, specificity and accuracy of ANN based on the biochemical parameters are73.6%,61.5%, and64.7%. The sensitivity, specificity and accuracy of the former are slightly higher than that of the latter.2. The area under curve (AUC) of all sets, train sets and test sets by ANN based on the combined parameters are0.7991,0.8417,0.7100. The area under curve (AUC) of all sets, train sets and test sets by ANN based on the biochemical parameters are0.7714,0.8110,0.7036.The prediction power of two models are moderate.3.The effect weight of pregnancy complications, CG, onset ages, itching ages, DBIL, S/D, delivery ages, ALT, age, delivery mode, fetus number, TBA, NST, amniotic fluid volume abnormality, AST, TBIL, uterine contraction is15.73%,14.63%,11.35%,10.56%,8.44%,7.24%,5.67%,4.91%,4%,3.84%,3.78%,3.64%,2.91%,2.27%,0.74%,0.14%,0.14%.Conclusion:1.An optimally trained and deployed ANN could be used to assess risks for fetus in intrahepatic cholestasis of pregnancy comprehensively and objectively.However,some relevant parameters should be further improved.2.The sensitivity, specificity and accuracy of the ANN based on the combined parameters are slightly higher than that based on the biochemical prameters.3.In the combined parameters,whose effect weight are more than10%are pregnancy complications, CG, onset ages and itching ages.