摘要
目的探讨随机森林模型和决策树模型在肝硬化并发上消化道出血患者预后评估中的应用。方法利用logistic回归方法从肝硬化住院患者病历资料中筛选出与并发上消化道出血有关联的变量;以筛选出的变量作为输入变量,以是否出血作为结局变量,分别建立随机森林、决策树和传统logistic回归模型,通过受试者工作特征曲线(ROC)来评价三种模型的性能。结果 logistic回归、决策树和随机森林模型在测试集中的准确率分别为81.5%、75.1%和88.9%,三种模型的ROC曲线下面积(AUC)分别为0.854、0.720和0.909;在随机森林模型的变量重要性评分中,血糖、胆固醇、血清钾、总蛋白、碱性磷酸酶、尿素氮等指标得分较高,提示这些指标或有临床意义。结论随机森林模型在肝硬化上消化道出血患者的预后评估中具有较高的应用价值。
Objective To explore the application of random forests and decision trees in the prognosis of upper gastrointestinal bleeding in patients with liver cirrhosis.Methods The logistic regression model was used to screen out the variables associated with upper gastrointestinal bleeding from the medical records of hospitalized patients with cirrhosis.Random forests,decision trees and logistic regression models were established with screened variables as the input variables and whether bleeding was used as the outcome variable.The performance of the models was evaluated by the receiver operating characteristic curve(ROC).Results The accuracy rates of logistic regression,decision trees and random forests models in the test set were 81.5%,75.1% and 88.9% respectively,The areas under the ROC curve(AUC) of the three models were 0.854,0.720 and 0.909,respectively;In the random forest model,blood glucose,cholesterol,serum potassium,activated partial thromboplastin time,edema,total protein,alkaline phosphatase,and urea nitrogen had higher importance scores,which indicated high clinical significance.Conclusion The random forest model had better performance in predicting upper gastrointestinal bleeding of patients with liver cirrhosis.
作者
于大海
李金
罗艳虹
刘近春
张岩波
Yu Dahai;Li Jin;Luo Yanhong(Shanxi Medical University(030001),Taiyuan)
出处
《中国卫生统计》
CSCD
北大核心
2019年第2期162-166,共5页
Chinese Journal of Health Statistics
基金
国家青年科学基金项目资助(81502897)
关键词
随机森林
决策树
肝硬化并发上消化道出血
疾病预后
Random forest
Decision tree
Liver cirrhosis
Upper gastrointestinal bleeding
Disease prognosis
