支持向量机模型和Logistic回归模型在肝硬化食管静脉曲张中的预测价值  

Predictive value of support vector machine model and Logistic regression model in esophageal varices with cirrhosis

作　　者：徐晶 张霞[2] 封爽 郑珊 冯慧芬[1] XU Jing;ZHANG Xia;FENG Shuang;ZHENG Shan;FENG Huifen(Department of Gastroenterology,the Fifth Affiliated Hospital,Zhengzhou University,Zhengzhou 450052;Zhengzhou University Press,Zhengzhou 450052)

机构地区：[1]郑州大学第五附属医院消化内科,郑州450052 [2]郑州大学出版社,郑州450052

出　　处：《郑州大学学报（医学版）》2022年第4期524-527,共4页Journal of Zhengzhou University(Medical Sciences)

基　　金：中国肝炎防治基金会王宝恩肝纤维化研究基金项目(2020006)。

摘　　要：目的:比较支持向量机模型和Logistic回归模型在肝硬化食管静脉曲张中的预测价值。方法:收集在郑州大学第五附属医院住院治疗的肝硬化患者305例,以电子胃镜检查结果作为“金标准”,将食管静脉曲张分为无或轻度曲张150例,中重度曲张155例,构建支持向量机、Logistic回归模型进行肝硬化食管静脉曲张风险预测。结果:构建的支持向量机模型和Logistic回归模型的预测正确率分别为86.8%和83.5%,二者的AUC、敏感度、特异度、阳性预测值、阴性预测值分别为0.931、0.896,84.8%、82.6%,88.9%、84.4%,88.6%、84.4%,85.1%、82.6%。支持向量机输出预测变量重要性居前4位的依次为肝硬度值、门静脉直径、血红蛋白、血小板计数/脾脏厚度,与Logistic回归模型一致。结论:支持向量机构建的肝硬化食管静脉曲张预测模型有较好的应用价值,较传统的Logistic回归模型表现更佳。Aim:To compare the predictive value of support vector machine model and Logistic regression model in esophageal varices with cirrhosis.Methods:A total of 305 patients with cirrhosis who were hospitalized in the Fifth Affiliated Hospital of Zhengzhou University were collected.With the results of electronic gastroscopy as the"gold standard",they were allocated into no or mild varicosis(n=150)and moderate or severe varicosis(n=155).Support vector machine model and Logistic regression model were established to predict the risk of esophageal varicosis in patients with cirrhosis.Results:The predictive accuracy of support vector machine model and Logistic regression model were 86.8%and 83.5%,respectively.The AUC,sensitivity,specificity,positive predictive value and negative predictive value of the 2 models were 0.931 and 0.896,84.8%and 82.6%,88.9%and 84.4%,88.6%and 84.4%,85.1%and 82.6%,respectively.The 4 most important predictive variables of support vector machine model output were liver stiffness,portal vein diameter,hemoglobin,platelet count/spleen thickness,which were consistent with Logistic regression model.Conclusion:The support vector machine model has greater application value in predicting esophageal varices in patients with cirrhosis,which has better performance than the traditional Logistic regression model.

关 键 词：食管静脉曲张 肝硬化 支持向量机模型 LOGISTIC回归模型 预测 

分 类 号：R575.2[医药卫生—消化系统]