摘要
Receiver operating characteristic (ROC) analysis is a tool used to describe the discrimination accuracy of a diagnostic test or prediction model. While sensitivity and specificity are the basic metrics of accuracy, they have many limitations when characterizing test accuracy, particularly when comparing the accuracies of competing tests. In this article we review the basic study design features of ROC studies, illustrate sample size calculations, present statistical methods for measuring and comparing accuracy, and highlight commonly used ROC software. We include descriptions of multi-reader ROC study design and analysis, address frequently seen problems of verification and location bias, discuss clustered data, and provide strategies for testing endpoints in ROC studies. The methods are illustrated with a study of transmission ultrasound for diagnosing breast lesions.
学科与关键词
- Consideration in Selection of the Control Group in Diagnostic TestsJianlan, Cui; Naiqing, ZhaoInternational Journal of Biological Sciences, 2012, 8(10): 1418-1419.被引 3
- Reader variability in mammography and its implications for expected utility over the population of readers and casesWagner, RE*; Beam, CA; Beiden, SVMedical Decision Making, 2004, 24(6): 561-572.被引 36
- Quasi-continuous and discrete confidence rating scales for observer performance studies: Effects on ROC analysisHadjiiski, Lubomir*; Chan, Heang-Ping; Sahiner, Berkman; Helvie, Mark A.; Roubidoux, Marilyn A.ACADEMIC RADIOLOGY, 2007, 14(1): 38-48.被引 19
