【作者】 李宁；

【导师】 张学红；

【作者基本信息】 兰州大学 ， 临床医学·妇产科学（专业学位）， 2024， 硕士

【摘要】 背景:妊娠期高血压疾病(hypertensive disorders of pregnancy,HDP)是一种常见的妊娠并发症。妊娠期间的发生高血压疾病,包括妊娠期高血压、子痫前期和子痫,是新生儿、胎儿和产妇死亡的主要原因,并会对后代的健康结果产生不利影响。辅助生殖技术,尤其是体外受精-胚胎移植技术(in vitro fertilization and embryo transfer,IVF-ET)为不孕患者带来了希望,在全球得到了广泛的应用。然而,与自然受孕相比,IVF-ET与HDP风险增加相关,尤其是冻融胚胎移植(frozen embryo transfer,FET)导致HDP的风险高于新鲜胚胎移植。但是接受IVF-ET的患者本身存在的生殖障碍(糖脂代谢异常、子宫内膜相关疾病、生殖衰老、排卵障碍等)是否会增加HDP的发病率以及FET中具体哪个因素可能导致HDP的风险增加还在探索中。因此,本研究针对FET人群,通过回顾性分析探索与HDP发生相关的危险因素,进一步筛选预测因子并建立FET妊娠后HDP发生的预测模型。第一部分:FET妊娠中HDP发生的危险因素研究目的:探索FET妊娠人群中与HDP发生相关的危险因素。方法:纳入2019年1月至2023年5月于兰州大学第一医院生殖医学中心行FET的周期临床资料,根据排除标准剔除不符合的周期,对所纳入患者的一般临床资料、FET相关资料和HDP发病情况进行差异性分析,随后运用单因素Logistic回归分析评估与HDP发生相关的危险因素。结果:1、本研究共纳入符合纳排标准的周期4079例,HDP病例组263例,同时期对照组3816例,6.45%(263/4079)为HDP患者。2、HDP组和对照组在内膜准备方案、AFC、AMH、BMI、PCOS/DOR/EMs患病与否、高血压家族史、剖宫产史、移植胚胎数量和孕囊数方面有统计学差异(P<0.05)。3、FET妊娠的单因素Logistic回归分析显示,无排卵内膜准备方案、AFC≤5或>40、BMI、PCOS、DOR、EMs、高血压家族史、剖宫产史、移植胚胎数量、孕囊数量与HDP发病显著相关(P<0.05)。结论:BMI增大、患有PCOS/DOR/EMs、AFC过低(0～5)或过高(>40)、高血压家族史、剖宫产史、无排卵的内膜准备方案(包括人工周期和降调节激素替代周期)、移植胚胎数量(≥2)、孕囊数量(≥2)是FET周期HDP发生的危险因素(P<0.05)。第二部分:冻胚移植后妊娠期高血压疾病发生的预测模型构建目的:构建冻融胚胎移植后妊娠人群妊娠期高血压疾病发生的孕早期预测模型。方法:筛选出单因素Logistic回归分析中P值较小(P<0.1)的因子,采用最小绝对收缩和选择算子(LASSO)回归技术对这些因子进一步筛选。按照7:3的比例将最终纳入的周期随机分配为训练集和验证集并比较两组间各因素是否具有差异。分别使用Logistic回归、随机森林(RF)、极端梯度增强树分类器(XGB)和支持向量机分类器(SVM)建立HDP发病的预测模型,将表现最好的模型纳入最终模型中,将模型可视化并评估其效能、拟合程度和临床应用范围。结果:1、基于LASSO回归lambada.1se的Log Lambda值,最终筛选出7个非零系数变量,这些变量包括内膜准备方案、BMI、PCOS、EMs、DOR、高血压家族史和孕囊数量。2、将数据集按照7∶3的比例随机,得到训练集(n=2855)和验证集(n=1224),对两组间观察因素进行差异性分析,结果表明两组间各个变量均不存在统计学差异(P>0.05)。3、采用Logistic回归、RF、XGB和SVM算法建立HDP预测模型,在训练集中模型的f1-score分别为93、97、92、91,验证集中模型的f1-score分别为90、88、89、89。最终选用Logistic回归模型作为本研究最终模型并用列线图直观展示。4、Logistic回归模型的训练集和验证集AUC分别为0.728和0.734;模型Hosmer-Lemeshow拟合优度检验,MAE=0.004,P=0.867,这说明模型没有过拟合现象;模型的DCA提示当HDP阈值概率为7%～64%时,与全部“干预”或者不“干预”相比,使用该模型具有净收益。结论:1、本研究最终纳入BMI、PCOS、EMs、DOR、高血压家族史、内膜准备方案和孕囊数量共7个预测因子建立FET妊娠后HDP发生的预测模型。这7个因子在单因素Logistic回归分析和多因素Logistic回归分析中均具有统计学意义(P<0.05),是HDP发生的独立危险因素。2、本研究中Logistic回归模型优于RF、XGB和SVM,因而作为最终预测模型,采用列线图展示较为直观简洁。3、该模型拟合较好,具有良好的区分能力,DCA显示对HDP阈值概率设置为7%～64%时,使用该模型可以获得净收益。更多还原

【Abstract】 Background: Hypertensive disorders of pregnancy(HDP)is a common pregnancy complication.Hypertensive disorders during pregnancy,including gestational hypertension,preeclampsia,and eclampsia,are the leading causes of neonatal,fetal,and maternal mortality and can adversely affect the health outcomes of future generations.Assisted reproductive technology,especially in vitro fertilization and embryo transfer(IVF-ET),has brought hope to infertile patients and has been widely used around the world.However,compared with natural conception,IVF-ET is associated with an increased risk of HDP,especially frozen embryo transfer(FET),which leads to a higher risk of HDP than fresh embryo transfer.However,whether the reproductive disorders(abnormal glucose and lipid metabolism,endometrial related diseases,reproductive aging,ovulation disorders,etc.)in patients undergoing IVF-ET will increase the incidence of HDP and which specific factors in FET may lead to the increased risk of HDP are still being explored.Therefore,this study aimed at the FET population,explored the risk factors related to the occurrence of HDP through retrospective analysis,further screened predictors and established a prediction model for the occurrence of HDP after FET pregnancy.Part Ⅰ: Risk factors of HDP in FET pregnancyObjective: To explore the risk factors associated with HDP in FET pregnant women.Methods: The clinical data of FET cycles in the Reproductive Medicine Center of the First Hospital of ** from January 2019 to May 2023 were included.According to the exclusion criteria,the cycles that did not meet the criteria were excluded.The general clinical data,FET-related data and HDP incidence of the included patients were analyzed.Then,univariate logistic regression analysis was used to evaluate the risk factors associated with HDP.Results:1.This study included 4079 cycles that met the inclusion criteria,263 in the HDP case group and 3816 in the control group at the same time,6.45 %(263/4079)were HDP patients.2.There were significant differences between the HDP group and the control group in endometrial preparation,AFC,AMH,BMI,PCOS,DOR,endometriosis,family history of hypertension,history of cesarean section,number of high-quality embryos transferred and number of gestational sacs(P<0.05).3.Univariate logistic regression analysis of FET pregnancy showed that endometrial preparation regimen,AFC,AMH,BMI,PCOS,DOR,endometriosis,family history of hypertension,history of cesarean section,number of embryos transferred,and number of gestational sacs were significantly correlated with the incidence of HDP(P<0.05).Conclusion : Increased BMI,PCOS,DOR,endometriosis,low AFC(0～5)or high AFC(>40),family history of hypertension,history of cesarean section,anovulatory endometrial preparation(including artificial cycles and down-regulation hormone replacement cycles),number of embryos transferred(≥ 2),number of gestational sacs(≥2)were risk factors for HDP in FET cycles(P<0.05).Part Ⅱ: Establishment of a predictive model for hypertensive disorder complicating pregnancy after frozen embryo transferObjective: To construct a prediction model of hypertensive disorder complicating pregnancy in early pregnancy after frozen embryo transfer.Methods: The factors with small P value(P < 0.1)in single factor Logistic regression analysis were screened out,and these factors were selected by the least absolute shrinkage and selection operator(LASSO)regression technique.According to the ratio of 7: 3,the final included cycles were randomly assigned to the training set and the test set,and the differences between the two groups were compared.Logistic regression,random forest,extreme gradient enhancement tree classifier and support vector machine classifier were used to establish the prediction model of HDP incidence.The best performing model was included in the final model,and the model was visualized and its efficacy,fitting degree and clinical application range were evaluated.Results:1.Based on the Log Lambda value of LASSO regression lambada.1 se,seven non-zero coefficient variables were finally selected,including endometrial preparation plan,BMI,PCOS,endometriosis,DOR,family history of hypertension and gestational sac number.2.The data set was randomly divided into training set(n = 2855)and validation set(n = 1224)according to the ratio of 7 : 3,and the difference analysis of the observation factors between the two groups showed that there was no statistical difference between the two groups(P>0.05).3.The HDP prediction model was established by Logistic regression,RF,XGB and SVM algorithms.The f1-scores of the model in the training set were 93,97,92 and91,respectively,and the f1-scores of the model in the test set were 90,88,89 and 89,respectively.Finally,the Logistic regression model was selected as the final model of this study and displayed intuitively with a nomogram.4.The AUC of the training set and the validation set of the Logistic regression model were 0.728 and 0.734,respectively,model Hosmer-Lemeshow goodness of fit test,MAE=0.004,P=0.867,indicating that the model has no over-fitting phenomenon,the DCA of the model suggests that when the HDP threshold probability is 7 %-64 %,the use of the model has a net benefit compared to all ’ intervention ’ or no ’ intervention ’.Conclusion:1.This study finally included 7 predictors of BMI,PCOS,endometriosis,DOR,family history of hypertension,endometrial preparation regimen and number of gestational sacs to establish a predictive model for HDP after FET pregnancy.These seven factors were statistically significant in univariate logistic regression analysis and multivariate logistic regression analysis(P<0.05),and were independent risk factors for HDP.2.In this study,the Logistic regression model is superior to RF,XGB and SVM,so as the final prediction model,the nomogram is more intuitive and concise.3.The DCA shows that when the threshold probability of HDP is set to 7 % ～64 %,the net income can be obtained by using this model.更多还原