R. Alonso Sanz, C. T. Nakas, M. D. C. Pardo Llorente
The problem of comparing alternate diagnostic markers for diagnosing a disease can be reduced to one of comparing ROC curves. The most commonly used approach to comparing ROC curves is to test the equality of the area under ROC (DeLong et al. (1988)) but in, some cases, an approach based on two-sample test statistics is a better alternative (Nakas et al. (2003)). Several approaches based on to use of the difference of two individual nonparametric statistics associated with the two diagnostic markers are proposed. Their distributions are obtained using bootstrap and their performance are studied through Monte Carlo simulation. One of our approaches is very powerful when ROC curves cross and other is comparable to DeLong test for non-crossing ROC curves. A real data set is used to illustrate our proposal.
Palabras clave: Receiver Operating Characteristic Curves (ROC); two-sample test; area under ROC curve; bootstrapProgramado
L08.3 Sesión de la Red Nacional de Bioestadística
5 de septiembre de 2016 15:40