卵巢高反应多因素预测模型的建立及对辅助生殖技术治疗结局的影响

Multivariable Prediction Model and Clinical Outcome Impact of Ovarian Hyper-Response During Controlled Ovarian Stimulation

目的:探讨卵巢高反应对辅助生殖技术(ART)临床结局的影响并建立和验证卵巢高反应多因素预测模型。方法:对1 106个体外受精/卵胞浆内单精子显微注射(IVF/ICSI)周期患者的临床资料进行回顾性分析,分为正常反应组和高反应组。行组间比较和多因素相关性分析建立预测模型。按4∶1比例随机生成建模组和验证组,用受试者工作特征曲线(ROC)的曲线下面积(AUC)和Hosmer-Lemshow检验检测模型的判别能力和准确性。结果:①正常反应组和高反应组种植率分别为28.5%、26.7%,临床妊娠率分别为45.0%、41.9%,差异均无统计学意义。高反应组获卵数、优质胚胎率更高,中重度卵巢过度刺激综合征(OHSS)发生率及取消移植率也随之增加,差异有统计学意义。②Logistic多元逐步回归分析筛选出年龄、基础窦卵泡数(AFC)、基础FSH(bFSH)与基础LH(bLH)作为卵巢高反应的预测指标。ROC的AUC及其95%CI为0.787(0.718-0.856),Hosmer-Lemshow检验:P>0.05,模型拟合良好。结论:年龄、AFC、bFSH、bLH是卵巢高反应的预测因素,成功建立多因素预测模型。高反应不影响种植率、妊娠率等妊娠结局,但OHSS发生率和取消移植率随之增加。

Objective： To investigate the influence of ovarian hyper response during controlled ovarian stimulation （COS） in assisted reproduction and to establish and verify the multivariable prediction model of hyper response. Methods： The clinical data of 1 106 IVF/ICSI cycles in Wuhan Univer- sity Reproductive Center from Jan 2010 to Dec 2011 were retrospectively analyzed. Comparison and multivariate analysis between normal-response and hyper-response group were performed. Data were randomly generated into module and validation groups with a ratio of 4 ： 1. Area under ROC curve and Hosmer-Lemshow test were applied to validate the discriminatory power and ac- curacy of the prediction model. Results： ① Implantation rate of normal-response and hyper-re- sponse groups was 28.5% and 26.7% respectively, and clinical pregnancy rate was 45.0 % and 41.9% respectively, without statistical difference. Number of oocytes retrieved and high-quality embryo rate were significantly higher in hyper-response group, accompanied with higher trans- ferred cancellation and moderate to severe OHSS incidence rate. ②Multivariable stepwise logisticregression showed ages, basal antral follicles, basal FSH, and basal LH levels were predictive factors of ovarian hyper-response. Area under ROC curve and its 95%CI were 0. 787 （0. 718- 0. 856）. Hosmer-Lemshow test showed P〉0.05, which meant a good fit of the prediction mod- el. Conclusion. Ages, AFC, basal FSH, and basal LH were the predictive factors of high ovarian response. A multivariable prediction model had been successfully constructed. High response during COS did not affect treatment outcomes, along with an increase of OHSS and transferred cancellation rate.