比例优势模型实现ROC分析的方法及其应用前景分析  

The Proportional Odds Model and its Prospect Applied to ROC Analysis

作　　者：宇传华[1] 余松林[1] 徐勇勇[2] 

机构地区：[1]华中科技大学同济医学院公共卫生学院流行病与卫生统计学系,430030 [2]第四军医大学预防医学系卫生统计学教研室

出　　处：《中国卫生统计》2005年第5期293-296,共4页Chinese Journal of Health Statistics

基　　金：国家自然科学基金资助项目(30371254)

摘　　要：目的探讨比例优势模型在ROC分析中的应用前景。方法比较比例优势模型与双正态模型等经典方法所计算的ROC曲线下面积及其标准误;采用灵敏度残差平方和与决定系数两个指标评价参数模型的拟合优度。结果在一般情况下,由比例优势模型所得到的ROC曲线指标结果与经典方法很接近;对于有序分类资料和连续型资料,该模型的拟合效果均较好;但由于该模型获得的ROC曲线形状单一,有些情况下该模型的拟合不理想。结论与经典的方法相比,比例优势模型有其自身的特点,实际应用时应慎重做出选择。Objective Explore the role of proportional odds model in ROC analysis. Methods Compare the area under ROC curve Of proportional odds model and its standard errors with of common methods, such as binormal model;evaluate the goodness of fit for the parametric models using the sum of square of residuals of sensitivity and determination coefficient between actual sensitivity and estimated sensitivity. Results The values of the measures of ROC analysis from the proportional odds model are similar with of common methods; whether ordinal category data or continuous data, the effect of fit for the model are generally good. Conclusion Comparing with common methods of ROC analysis, the proportional odds model have itself characteristic. It is an alternative method for ROC analysis in practice.

关 键 词：ROC曲线 比例优势模型 模型评价 双正态模型 ROC分析 比例 ROC曲线下面积 有序分类资料 经典方法 拟合优度 

分 类 号：R195[医药卫生—卫生统计学]