基于术前检测指标构建Logistic回归模型在预测壶腹周围癌病理类型中的价值  

Prediction value of Logistic regression model based on preoperative detection indexes in pathological types of periampullary carcinoma

作　　者：王琦琦 龚梦元 冯正源 韩亮[1] 王铮[1] 马清涌[1] 仵正[1] Wang Qiqi;Gong Mengyuan;Feng Zhengyuan;Han Liang;Wang Zheng;Ma Qingyong;Wu Zheng(Department of Hepatobiliary Surgery,the First Affiliated Hospital of Xi'an Jiaotong University,Xi'an 710061,China)

机构地区：[1]西安交通大学第一附属医院肝胆外科,710061

出　　处：《中华肝脏外科手术学电子杂志》2023年第2期196-200,共5页Chinese Journal of Hepatic Surgery(Electronic Edition)

基　　金：基金资助:陕西省自然科学基础研究计划(2020JQ-510);西安交通大学第一附属医院科研发展基金(2021QN-24)。

摘　　要：目的探讨基于术前检测指标构建Logistic回归模型在预测壶腹周围癌病理类型中的价值。方法本研究对象为2015年1月至2018年12月在西安交通大学第一附属医院行根治性胰十二指肠切除术的256例壶腹周围癌患者。其中男159例,女97例;年龄39~88岁,中位年龄64岁。患者均签署知情同意书,符合医学伦理学规定。收集患者术前数据和术后病理类型,建立预测壶腹周围癌病理类型的Logistic回归模型、决策树模型和随机森林模型。随机选取192例作为训练集,64例为测试集,采用ROC曲线下面积(AUC)用于交互验证模型预测价值。结果所有患者病理类型中,肠型99例,胰胆管型157例。基于肿瘤直径、Hb变量构建多元Logistic回归模型,信息准则(AIC)值达到最低24。根据AdaBoost算法结果,CEA、CA125、ALB、肿瘤直径、CA19-9等是影响决策树的重要因素,分别设置随机种子为91和62得到相应决策树模型和随机森林模型。测试集验证Logistic回归、决策树、随机森林模型的准确度依次为0.810、0.766、0.766,敏感度依次为0.848、0.775、0.956,特异度依次为0.706、0.750、0.316,AUC依次为0.764、0.762、0.698。结论基于肿瘤直径、Hb建立的Logistic回归模型可在术前较为准确判别壶腹周围癌的病理类型,指导术前辅助化疗。Objective To evaluate the prediction value of constructing Logistic regression model based on the preoperative detection indexes in the pathological types of periampullary carcinoma.Methods 256 patients with periampullary carcinoma who underwent radical pancreaticoduodenectomy in the First Affiliated Hospital of Xi'an Jiaotong University from January 2015 to December 2018 were recruited in this clinical trial.Among them,159 patients were male and 97 female,aged from 39 to 88 years,with a median age of 64 years.The informed consents of all patients were obtained and the local ethical committee approval was received.Preoperative data and postoperative pathological types of all patients were collected,and Logistic regression model,decision tree model and random forest model were established to predict the pathological types of periampullary carcinoma.192 cases were randomly selected into the training set and 64 cases into the test set.The predictive value of this model was assessed by the area under ROC curve(AUC).Results Among all patients,99 patients were classified as the intestinal type and 157 cases of pancreaticobiliary type.Based on the tumor diameter and Hb variables,a multivariate Logistic regression model was constructed,and the Akaike information criterion(AIC)value reached the lowest 24.According to the results of AdaBoost algorithm,CEA,CA125,ALB,tumor diameter and CA19-9 were important factors affecting the decision tree.The corresponding decision tree model and random forest model were obtained by setting random seeds as 91 and 62,respectively.In the test set,the accuracy of Logistic regression,decision tree and random forest models was 0.810,0.766 and 0.766,the sensitivity was 0.848,0.775 and 0.956,the specificity was 0.706,0.750 and 0.316,and the AUC was 0.764,0.762 and 0.698,respectively.Conclusions Logistic regression model based on tumor diameter and Hb can accurately identify the pathological types of periampullary carcinoma before surgery,and provide guidance for the preoperative adjuvant chemot

关 键 词：壶腹周围癌 病理类型 LOGISTIC回归模型 决策树模型 随机森林模型 

分 类 号：R735.0[医药卫生—肿瘤]