ROC曲线下面积的ML估计与假设检验

Maximum likelihood estimation and hypothesis test of areas under receiver operating characteristic(ROC) curves

摘要:
目的探讨诊断试验中配对设计资料的ROC分析方法。
方法在双正态模型基础下应用ML估计方法计算ROC曲线下面积, 正态近似法估计面积的可信区间及假设检验。
结果由迭代法进行参数估计, 得到ROC曲线下的面积、面积的标准误及置信区间, 可计算出面积比较的U检验统计量。
结论可用于配对设计的诊断试验的比较和评价, 包括对连续性和等级分类资料的处理。

Abstract:
Objective To explore receiver operating characteristic(ROC)analysis method with paired datasets in medical diagnostic tests.
Methods On the basis of binormal model, maximum likelihood estimation was applied to areas calculation, and approximately normal method to confidence interval estimation and hypothesis test.
Results By the iterative procedure, ML estimation and standard errors of ROC curve parameters a and b, ROC-area index Az were obtained and hypothesis test statistic U value could be caculated.
Conclusion ROC analysis can be applied to analyse and evaluate the diagnostic test with paired data, including continuous data and rank-ordered data.