3种统计模型在预测肺癌术后并发症中的比较  

A comparison between three statistical models in predicting post-operative complication for lung cancer patients

作　　者：宋健[1] 苏虹[1] 周洋洋[1] 郭亮亮[1] 王保龙[2] 

机构地区：[1]安徽医科大学公共卫生学院流行病与卫生统计学系,合肥230032 [2]安徽医科大学附属省立医院检验科,合肥230001

出　　处：《安徽医科大学学报》2014年第4期472-475,共4页Acta Universitatis Medicinalis Anhui

基　　金：国家自然科学基金(编号:81172172)

摘　　要：目的探讨BP神经网络模型在预测肺癌术后并发症中的应用价值。方法调查肺癌患者术后并发症发生情况。分别应用Logistic回归、BP神经网络模型和经Logistic回归筛选变量后的BP神经网络模型3种办法建立预测模型,并比较3种模型的预测准确度。结果 Logistic回归、BP神经网络模型和经Logistic回归筛选变量后的BP神经网络模型的预测一致率分别为81.6%、89.7%、90.8%。3种模型受试者工作特征曲线(ROC曲线)下面积(AUC)分别为0.636、0.801、0.808。Logistic模型的AUC与两种BP神经网络模型的差异有统计学意义(P<0.05)。结论 BP神经网络对肺癌术后并发症预测的效果优于Logistic回归模型。Objective To explore the application value of BP neural network in predicting post-operative complication for lung cancer patients. Methods We applied Logistic regression, BP neural network model and BP neural network model screening variables by Logistic regression to establish prediction models and evaluate the practical application of each model in the prediction accuracy. Results The prediction accuracy of Logistic regression, BP neural network model and BP neural network model screening variables by Logistic regression were 81. 6% , 89. 7% ,90. 8% and the AVe of Roe in the three models were 0. 636,0. 801,0.808, respectively. There were significant differences of the AVe of Roe between Logistic regression and two BP neural network models. Conclusion The discrimination performance of BP neural network models is better than Logistic regression in the prediction of post-operative complication for lung cancer patients.

关 键 词：LOGISTIC模型 BP神经网络 肺癌 并发症 

分 类 号：R195.1[医药卫生—卫生统计学] R734.2[医药卫生—卫生事业管理]