IVF-ET妊娠结局预测模型的研究进展

Research progress of IVF-ET pregnancy outcome prediction model

不孕不育是指夫妻在进行无保护措施的性行为后1年内未获得妊娠,这种现象的普遍存在对患者和社会都造成了不同程度的影响。准确预测妊娠结局是生殖医学中一个非常有意义和具有挑战性的任务。本文针对辅助生殖结局预测的建模方法进行了综述和探讨分析。研究发现,建模过程包括6个步骤,即业务理解、数据理解、数据准备、建模、模型评估和模型发布。建模方法主要包括传统的统计模型和基于大数据的新技术(机器学习)两大类方法。与机器学习预测模型相比,传统预测建模方法能较好地探析各影响因素对妊娠结局的作用机制,但预测能力相对较低;机器学习方法可以大幅度提高预测能力,但不能生成透明的、可解释的规则。因此,建模方法的使用需根据临床实践目的不同进行选择。随着大数据技术的发展,预测模型的效能已得到极大的提高。与此同时,也面临着更多的挑战,包括解决大数据衍生出的问题,识别新的信息特征,平衡预测能力和结果解释性,以及用金标准研究设计验证模型等。

Infertility is defined as the inability of a couple to have a child within one year after unprotected sex.The prevalence of this phenomenon has affected both patients and society to varying degrees.Accurate prediction of pregnancy outcome is a very meaningful and challenging task in reproductive medicine.In this paper,the modeling methods of assisted reproductive outcome prediction are reviewed and analyzed.The studies found that the modeling process consists of six steps:business understanding,data understanding,data preparation,modeling,model evaluation,and model release.Modeling methods mainly include traditional statistical models and new techniques based on big data(machine learning).Compared with the machine learning predictive model,the traditional predictive modeling method can better explore the action mechanism of each influencing factor on pregnancy outcome,but its predictive ability is relatively weak.Machine learning methods can greatly improve predictive power,but cannot generate transparent and interpretable rules.Therefore,the use of modeling methods should be selected according to different clinical practice purposes.With the development of big data technology,the efficiency of prediction models has been greatly improved.At the same time,there are more challenges,including solving the problems derived from big data,identifying new information features,balancing predictive power and interpretability of results,and researching design and validating models with gold standard research and design.