基于决策树法构建子宫肌瘤合并高血压患者术后卵巢功能早衰的风险预测模型  

Analysis on constructing a risk prediction model for premature ovarian function failure in patients with uterine fibroids complicated with hypertension after surgery based on decision tree method

作　　者：王丹[1] 张紫娟[1] 杨慧春 梁美丽 郑灵芝[1] Wang Dan;Zhang Zijuan;Yang Huichun;Liang Meili;Zheng Lingzhi(Department of Obstetrics and Gynecology,Taizhou Hospital of Zhejiang Province,Taizhou 318050,China)

机构地区：[1]浙江省台州医院妇产科,台州317000

出　　处：《中华内分泌外科杂志》2023年第3期323-326,共4页Chinese Journal of Endocrine Surgery

基　　金：浙江省基础公益研究计划项目(LGF19H040008)。

摘　　要：目的应用决策树卡方自动交互检测(chi-squared automatic interaction detector,CHAID)算法和二分类Logistic回归分析法分别构建子宫肌瘤合并高血压患者术后卵巢功能早衰(premature ovarian failure,POF)的风险预测模型,并对风险预测模型结果进行对比分析。方法回顾性分析2019年1月至2022年9月浙江省台州医院收治的子宫肌瘤合并高血压患者为研究对象,应用CHAID算法和Logistic回归分析法分别建立风险预测模型,通过受试者工作特征曲线(receiver operating characteristic,ROC)的曲线下面积(area under curve,AUC)对两种模型的预测效果进行对比评价。结果共收集患者860人,其中术后卵巢功能早衰56人,卵巢功能早衰发生率6.51%;CHAID法和Logistic回归分析法均显示子宫肌瘤手术、高血压、吸烟或被动吸烟、POF家族史、睡眠状态、体育锻炼和人流刮宫史是POF的重要影响因素;决策树模型风险预测的正确率为88.2%,模型拟合效果较好,Logistic回归模型Hosmer-Leme-show拟合优度检验显示模型拟合较好;Logistic回归模型AUC为0.893(95%CI:0.862~0.899),决策树模型AUC为0.882(95%CI:0.856~0.899),两模型预测价值均为中等,差异无统计学意义(Z=0.254,P>0.05)。结论将决策树和Logistic回归模型相结合可从不同层面发现子宫肌瘤合并高血压患者术后POF的影响因素,能更充分地了解各因素间的相互关系。子宫肌瘤合并高血压患者术后POF风险模型的建立可为子宫肌瘤合并高血压患者术后干预提供参考依据,更有效地帮助患者积极预防和减缓POF的发生发展。Objective The decision tree Chi-square automatic interactive detection(CHAID)algorithm and binary Logistic regression analysis were used to construct the risk prediction model of premature ovarian failure(POF)in patients with uterine fibroids complicated with hypertension after surgery,and the results of the risk prediction model were compared and analyzed.Methods Patients with uterine fibroids complicated with hypertension admitted to Taizhou Hospital of Zhejiang Province from Jan.2019 to Sep.2022 were retrospectively analyzed as the research objects.CHAID algorithm and Logistic regression analysis were used to establish risk prediction models,respectively.The area under the curve(AUC)of receiver operating characteristic curve(ROC)was used to compare and evaluate the prediction effects of the two models.Results A total of 860 patients were collected,including 56 patients with premature ovarian function failure after operation,and the incidence of premature ovarian function failure was 6.51%.CHAID method and Logistic regression analysis showed that uterine myoma surgery,hypertension,smoking or passive smoking,family history of premature ovarian failure,sleep status,physical exercise and history of induced curettage were important influencing factors of premature ovarian failure.The accuracy of risk prediction of decision tree model was 88.2%,and the fitting effect of the model was good.The Logistic regression model Hosmer-Leme-show goodness of fit test showed that the model fit was good.The AUC of Logistic regression model was 0.893(95%CI:0.862-0.899),and the AUC of decision tree model was 0.882(95%CI:0.856-0.899).The predictive value of the two models was moderate,and there was no significant difference between them(Z=0.254,P>0.05).Conclusions The combination of decision tree and Logistic regression model can find the influencing factors of premature ovarian function failure in patients with uterine fibroids complicated with hypertension after operation from different levels,and the relationship between the fact

关 键 词：子宫肌瘤 高血压 卵巢功能早衰 决策树预测模型 

分 类 号：R737.33[医药卫生—肿瘤] R544.1[医药卫生—临床医学] R711.75