决策树和Logistic回归模型对体外受精-胚胎移植患者妊娠结局的预测价值比较

Comparison of the predictive value of decision tree and Logistic regression models for pregnancy outcome in patients undergoing in vitro fertilisation-embryo transfer

目的:比较决策树和Logistic回归模型对体外受精-胚胎移植(in vitro fertilization and embryo transfer,IVF-ET)患者妊娠结局的预测价值。方法:纳入2021年1月至2022年10月在长治医学院附属和平医院接受IVF-ET的患者350例为研究对象,根据妊娠结局分为妊娠成功组(215例)和妊娠失败组(135例)。收集患者临床资料,建立IVF-ET患者妊娠结局Logistic回归和决策树预测模型,并在是否基于Logistic回归结果条件下建立决策树分析模型(决策树1和决策树2),采用受试者工作特征(receiver operating characteristic,ROC)曲线对模型预测效果进行评价。结果:350例患者中,妊娠成功患者占61.43%,妊娠失败者占38.57%。妊娠失败组年龄≥35岁、不孕年限≥5年、周期次数≥1次、有心理精神障碍的患者比例及HCG日血清孕酮水平均高于妊娠成功组,获卵数≥10枚、受精率≥75%的患者比例及HCG日子宫内膜厚度、优质胚胎数小于妊娠成功组(P<0.05)。多因素Logistic回归分析结果显示,年龄、HCG日血清孕酮水平、优质胚胎数及心理精神障碍均是IVF-ET患者妊娠结局的影响因素(P<0.05)。决策树模型显示,年龄、HCG日血清孕酮水平、优质胚胎数为IVF-ET患者妊娠结局的影响因素。Logistic回归模型曲线下面积(area under curve,AUC)为0.832,预测敏感度、特异度和准确度分别为87.3%、71.4%、83.5%;决策树1的AUC为0.859,预测敏感度、特异度和准确度分别为85.1%、76.8%、85.6%;决策树2的AUC为0.820,预测敏感度、特异度和准确度分别为83.7%、73.2%、82.4%。决策树1的AUC大于决策树2(P<0.05),但与Logistic回归模型的AUC比较差异无统计学意义(P>0.05)。结论:Logistic回归模型和决策树模型对于IVF-ET患者妊娠结局均有一定的预测价值。

Objective:To compare the predictive value of decision tree and Logistic regression model for pregnancy outcomes in in vitro fertilization-embryo transfer(IVF-ET)patients.Methods:A total of 350 patients undergoing IVF-ET in Heping Hospital of Changzhi Medical College from January 2021 to October 2022 were included in the study,and were divided into pregnancy success group(215 patients)and pregnancy failure group(135 patients)according to pregnancy outcome.Clinical data of patients were collected,Logistic regression and decision tree models were established to predict pregnancy outcomes of IVF-ET patients,and decision tree analysis models(decision tree 1 and decision tree 2)were established under the condition of whether the results of Logistic regression were based,and receiver operating characteristic(ROC)curve was used to evaluate the prediction effect of the model.Results:Of the 350 patients,61.43%had successful pregnancy,and the 38.57%had failed pregnancy.In the pregnancy failure group,the age≥35 years,infertility years≥5 years,cycle times≥1,the proportion of patients with mental disorders and serum progesterone level on HCG day were higher than those in the pregnancy success group,and the proportion of patients with egg number≥10,fertilization rate≥75%,endometrial thickness on HCG day and high-quality embryo number were lower than those in the pregnancy success group(P<0.05).Multivariate Logistic regression analysis showed that age,serum progesterone level on HCG day,number of high-quality embryos and mental disorders were all influencing factors of pregnancy outcome in IVF-ET patients(P<0.05).Decision tree model showed that age,serum progesterone level on HCG day and number of high-quality embryos were influencing factors of pregnancy outcome in IVF-ET patients.The AUC of Logistic regression model was 0.832,and the prediction sensitivity,specificity and accuracy were 87.3%,71.4%and 83.5%,respectively.The AUC of decision tree 1 was 0.859,and the prediction sensitivity,specificity and accuracy were 85.1%,76.8%and 85.6%,respectively.The AUC of decision tree 2 was 0.820,and the prediction sensitivity,specificity and accuracy were 83.7%,73.2%and 82.4%,respectively.The AUC of decision tree 1 was greater than that of decision tree 2(P<0.05),but there was no statistical significance compared with that of Logistic regression model(P>0.05).Conclusion:Both Logistic model and decision tree model have certain predictive value for pregnancy outcome of IVF-ET patients.