基于XGBoost算法构建地屈孕酮治疗先兆流产效果模型的研究

Efficacy of didroxyprogesterone in the treatment of threatened abortion based on XGBoost algorithm

目的采用极端梯度上升(XGBoost)算法构建地屈孕酮治疗先兆流产的有效性及安全性预测模型。方法选取2017年1月至2018年1月乐山市人民医院诊治的2811例先兆流产患者的临床资料作为研究对象。根据其临床资料建立地屈孕酮用药数据库,依据妊娠结局分为保胎成功组、保胎失败组,以及阿氏评分>7组、阿氏评分≤7组,利用XGBoost算法构建预测模型。结果地屈孕酮保胎成功率为79.97%,新生儿阿氏评分≤7的占比为1.14%。基于XGBoost算法建立的地屈孕酮治疗先兆流产有效性的全部变量预测模型准确率为0.990,F1值为0.975,受试者工作特征(ROC)曲线下面积(AUC)为0.993;安全性的全部变量预测模型准确率为0.986,F1值为0.000,AUC为0.609。结论基于XGBoost算法的地屈孕酮治疗先兆流产有效性的最终预测模型能准确可靠地预测地屈孕酮治疗先兆流产的有效性。

Objective To establish a predictive model for the efficacy and safety of didroxyprogesterone in the treatment of threatened abortion using extreme gradient Boosting(XGBoost)algorithm.Methods The clinical data of 2,811 cases of threatened abortion in Leshan People′s Hospital from January 2017 to January 2018 were selected,according to the clinical data of threatened abortion patients,the patients were divided into successful tocolysis group and failed tocolysis group,Apgar scores>7 group and Apgar scores≤7 group according to pregnancy outcome,and then use XGBoost algorithm to build prediction model.Results The success rate of didroxyprogesterone was 79.97%,and the rate of neonatal Apgar score≤7 was 1.14%.Based on XGBoost algorithm,the accuracy of all variables prediction model for the efficacy of didroxyprogesterone in the treatment of threatened abortion was 0.990,F1 value was 0.975,AUC value was 0.993.The accuracy rate of all variables prediction model for the safety was 0.986,F1 value was 0.000,AUC value was 0.609.Conclusions The final prediction model based on XGBoost algorithm can accurately and reliably predict the effectiveness of didroxyprogesterone in the treatment of threatened abortion.