基于随机森林算法构建模型预测小儿肠套叠空气灌肠复位失败的应用价值  

Application Value of Building a Model Based on Random Forest Algorithm to Predict the Failure of Air Enema Reduction of Intussusception in Children

作　　者：赵莲芳 何豪杰 郑宗杰 Zhao Lianfang;He Haojie;Zheng Zongjie(Qingliu County General Hospital,Qingliu Fujian 365300,China)

机构地区：[1]清流县总医院,福建清流365300

出　　处：《医疗装备》2023年第14期6-10,共5页Medical Equipment

摘　　要：目的 基于随机森林算法构建模型分析影响小儿肠套叠空气灌肠复位失败的因素。方法 选取2021年9月至2022年8月于医院小儿外科就诊的86例小儿肠套叠患儿作为研究对象,其中空气灌肠复位成功58例,空气灌肠复位失败28例。分析上述患儿临床资料及超声影像特征,应用Lasso回归筛选预测空气灌肠失败的独立预测变量,基于上述独立预测变量使用随机森林算法构建模型,并进行内部验证,运用评分分析独立预测变量与空气灌肠复位失败的相关性。结果 发病时间、是否血便、套鞘厚径为小儿肠套叠空气灌肠复位失败的独立预测变量。基于上述3个变量构建模型及ROC曲线,AUC为0.980,灵敏度为100.0%,特异度为90.7%,根据约登指数确定截断值为0.629。内部验证显示该模型具有较好的预测能力,评分显示影响小儿肠套叠空气灌肠复位失败的危险因素依次为套鞘厚径、是否血便、发病时间。结论 基于临床资料与超声影像特征构建随机森林算法模型在预测小儿肠套叠空气灌肠复位失败中具有一定的临床应用价值,且显示影响小儿肠套叠空气灌肠复位失败的危险因素依次为套鞘厚径、是否血便、发病时间。Objective The model was built based on random forest algorithm to analyze the factors affecting the failure of air enema reduction of intussusception in children.Methods A total of 86 children with intussusception who saw a doctor in pediatric surgery in the hospital from September 2021 to August 2022 was selected as the study objects.Among them,58 cases received successful air enema reduction and the other 28 cases' reduction were failed.The clinical data and ultrasonic image characteristics of the above patients were analyzed,Lasso regression was applied to screen independent variables of predicting air enema failure,a model was built using random forest algorithm based on the above independent predictors,and internal validation was conducted,and scores were used to analyze the correlation between independent predictors and failed air enema reduction.Results The onset time,blood stool or not and sheath thickness were independent predictors of the failure of air enema in children with intussusception.The model and ROC curve were constructed based on the above three variables.AUC was 0.980,sensitivity was 100.0%,specificity was 90.7%,and the cut-off value was 0.629 according to the Yoden index.The internal validation showed that the model had a good predictive ability.The scores showed that the risk factors affecting the failure of air enema in children with intussusception were in turn the thickness of the sheath,bloody stool or not,and the onset time.Conclusions The random forest algorithm model based on clinical data and ultrasonic image characteristics has certain clinical application value in predicting the air enema reduction failure of children with intussusception,and is worthy of clinical promotion.The random forest algorithm scores show that the risk factors affecting the air enema failure of children with intussusception are in turn the thickness of the sheath,bloody stool or not,and the onset time.

关 键 词：小儿肠套叠 随机森林算法 空气灌肠复位 

分 类 号：R445.1[医药卫生—影像医学与核医学]