异体骨髓移植后早发性卵巢功能不全患者首疗程激素补充治疗月经预测模型的建立OA北大核心CSTPCD
Establishment of a Prediction Model for Menstruation after the First Course of Hormone Replacement Therapy in Premature Ovarian Insufficiency Patients af-ter Allogeneic Hematopoietic Stem Cell Transplantation
目的:研究建立异体骨髓移植(allo-HSCT)后早发性卵巢功能不全(POI)患者首疗程激素补充治疗(HRT)的月经预测模型,为制定激素补充治疗方案提供一定参考价值.方法:选择2017 年1 月至2022 年10 月在苏州大学附属第一医院就诊的allo-HSCT后POI患者154 例进行回顾性分析,根据首疗程HRT后月经来潮情况分为理想月经组(116 例)和非理想月经组(38 例).单因素分析比较两组一般特征和临床资料差异后,选择纳入的预测因子.将纳入人群随机拆分为训练集和验证集后,利用随机森林算法构建训练集的月经预测模型,并通过验证集验证模型的预测效率.最后将模型制作成用户交互界面并部署至服务器共享.结果:单因素分析结果示,两组患者就诊年龄、体质量指数(BMI)、孕次、产次、血液病诊断、移植年龄、供体性别、卵泡刺激素(FSH)、黄体生成素(LH)、腰椎骨密度(BMD)、HRT方案差异均有统计学意义(P<0.05).依据平均准确度下降程度选择纳入模型的预测因子为就诊年龄、移植年龄、BMI、FSH、HRT方案、产次、孕次.初步构建随机森林模型后,优化模型参数,决策树数量(ntree)=500,特征数(mtry)=6,以80%和20%划分训练集和验证集,使模型拟合度高的同时误差率稳定,采用十倍交叉验证降低过度拟合.最终构建的月经预测模型曲线下面积为 0.768,灵敏度为 0.695,特异度为 0.735.结论:本研究成功建立了allo-HSCT后POI患者首疗程HRT的月经预测模型,该模型假阳性率较低,提示当模型预测结果为非理想月经时,可考虑调整拟定的HRT方案,以促进早期月经来潮.
Objective:To establish a menstrual prediction model after the first course of hormone replacement therapy(HRT)in premature ovarian insufficiency(POI)patients after allogeneic hematopoietic stem cell transplan-tation(allo-HSCT),and to provide certain reference value for formulating HRT plans.Methods:The retrospective analysis recruited 154 POI patients after allo-HSCT in the First Affiliated Hospital of Soochow University from Jan-uary 2017 to October 2022.They were divided into ideal menstruation group(n=116)and unideal menstruation group(n=38)according to menstruation after the first course of HRT.Basic characteristics and clinical data were compared in single-factor analysis to select predictive factors.Patients were randomly divided into training set and test set.The menstrual prediction model was developed based on random forest algorithm on the training set and the prediction efficiency was verified by the test set.Finally,we made a user interaction interface and deployed to the server for sharing.Results:The single-factor analysis suggested statistic difference of age of visit,body mass index(BMI),gravidity,parity,hematologic diseases,transplantation age,donor gender,follicle-stimulating hormone(FSH),Luteinizing Hormone(LH),lumbar bone mineral density(BMD)and HRT plan(P<0.05).According to mean decrease accuracy,the predictive factors included visit age,transplantation age,BMI,FSH,HRT plans,gravidity and parity.After the initial establishment of the random forest model,we improved it by adjusting ntree to 500,mtry to 6 and training/test set division to 80%/20% .We also used tenfold cross validation to reduce over-fitting.The area under curve(AUC)of the final constructed menstrual prediction model was 0.768,a sensitiv-ity of 0.695 and a specificity of 0.735.Conclusions:This study successfully established a menstrual prediction model for amenorrhea patients after allo-HSCT when finished the first course of HRT.The false positive rate was low,suggesting that if the prediction result of the model is non-ideal menstruation,we may consider adjusting HRT plans to promote menstruation in time.
张宁;刘卫泽宇;张婧婧;李晓宇;孙芳璨;陈慧赟;马骁;韩冰
苏州大学附属第一医院:妇产科,江苏 苏州 215000||苏州大学附属第二医院输血科,江苏 苏州 215000苏州大学附属第一医院:妇产科,江苏 苏州 215000苏州大学附属第一医院:血液科,江苏 苏州 215000
临床医学
异体骨髓移植早发性卵巢功能不全激素补充治疗随机森林预测模型
Allogeneic hematopoietic stem cell transplantationPremature ovarian insufficiencyHormone re-placement therapyRandom forestsPrediction model
《实用妇产科杂志》 2024 (007)
577-581 / 5
评论