基于Logistic回归模型及卡方自动交互探测决策树模型预测胎儿染色体异常的研究  

Prediction of fetal chromosomal abnormalities based on Logistic regression model and chi-squared automatic interaction detector decision tree model

作　　者：阳蓉 罗孝勇[2] 李陶 YANG Rong;LUO Xiaoyong;LI Tao(Department of Ultrasound,Daping Hospital,Army Military Medical University,Chongqing 400042,China)

机构地区：[1]陆军军医大学大坪医院超声科,重庆市400042 [2]四川省遂宁市中心医院超声科

出　　处：《临床超声医学杂志》2022年第12期895-899,共5页Journal of Clinical Ultrasound in Medicine

基　　金：重庆市技术创新与应用示范社会民生类一般项目(cstc2018jscx-msybX0018);重庆市影像医学与核医学临床医学研究中心(CSTC2015YFPT-gcjsyjzx0175)。

摘　　要：目的 应用Logistic回归模型及卡方自动交互探测(CHAID)决策树模型分析胎儿染色体异常的影响因素,比较两种模型对其的预测价值。方法 回顾性分析10种常见超声软指标阳性并获得羊水穿刺结果的642例单胎孕妇资料,以胎儿染色体结果为因变量,超声软指标为自变量,建立Logistic回归模型和CHAID决策树模型,筛选胎儿染色体异常的影响因素;绘制受试者工作特征(ROC)曲线分析两种模型对胎儿染色体异常的预测价值。结果 单因素Logistic回归模型显示,NT增厚、鼻骨缺失、侧脑室增宽均为胎儿染色体异常的影响因素(OR=4.942、2.558、2.463,均P<0.05);多因素Logistic回归筛选NT增厚(OR=7.511,P<0.05)、鼻骨缺失(OR=4.819,P<0.05)、侧脑室增宽(OR=4.789,P<0.05)用于回归模型的拟合,获得回归方程:Y=-2.888+2.016×NT增厚+1.572×鼻骨缺失+1.566×侧脑室增宽。CHAID决策树模型显示,NT增厚、鼻骨缺失均为胎儿染色体异常的影响因素。ROC曲线分析显示,Logistic回归模型预测胎儿染色体异常的曲线下面积大于CHAID决策树模型(0.712 vs. 0.675),差异有统计学意义(Z=2.267,P<0.05)。结论 Logistic回归模型和CHAID决策树模型均可预测胎儿染色体异常,Logistic回归模型的预测价值优于CHAID决策树模型。Objective To explore the predictive value of fetal chromosomal abnormalities by Logistic regression model and chi-squared automatic interaction detector(CHAID)decision tree model,and to analyze the influencing factors of fetal chromosomal abnormalities.Methods The data of 642 singleton pregnant women with 10 common positive ultrasound soft markers and amniocentesis results were retrospectively analyzed,fetal chromosomal abnormalities as the dependent variables and ultrasound soft marker as independent variables to establish Logistic regression model and CHAID decision tree model,the influence factors affecting fetal chromosomal abnormalities were screened. Receiver operating characteristic(ROC)curve was drawn to analyze the predictive value of the two models for fetal chromosomal abnormalities.Results The univariate Logistic regression model showed that the nuchal transluency(NT)thickening,nasal bone loss,and lateral ventricle widening were influence factors for fetal chromosomal abnormalities(OR=4.942,2.558,2.463,all P<0.05).Multivariate Logistic regression analysis were used to screen NT thickening(OR=7.511,P<0.05),nasal bone loss(OR=4.819,P<0.05)and lateral ventricle widening(OR=4.789,P<0.05)for fitting the regression model. The regression equation was obtained:Y=-2.888+2.016×NT thickening+1.572×nasal bone loss+1.566×lateral ventricle widening.The CHAID decision tree model showed that NT thickening and nasal bone loss were the influencing factors for fetal chromosomal abnormalities.ROC curve analysis showed that the area of the Logistic regression model was higher than that of the CHAID decision tree model(0.712 vs. 0.675),the difference was statistically significant(Z=2.267,P<0.05).Conclusion Logistic regression model and CHAID decision tree model have certain predictive value for fetal chromosomal abnormalities,and the Logistic regression model is better than the CHAID decision tree model.

关 键 词：LOGISTIC回归模型 卡方自动交互探测决策树模型 染色体异常 胎儿 超声软指标 

分 类 号：R445.1[医药卫生—影像医学与核医学] R714.5[医药卫生—诊断学]