Regiões de incerteza para a curva ROC em testes diagnósticos

Abstract

Diagnostic tests are methods capable of indicating the presence or absence of a disease, with a probability of error. The performance of a diagnostic test can be verified by some indicator, as: the specificity, the sensitivity and the ROC curve. A graph of the specificity complement versus sensitivity is called as ROC curve. The ROC curve demonstrates the test s ability to discriminate the different disease diagnosis, therefore it is a graphical tool that is used to assess the performance of a test. We define three types of confidence regions around the ROC curve: the punctual, the regional and the global. In some instances, depending on the clinical needs, the decision is taken under an specific region of the ROC curve. We review some procedures for estimating confidence region for the ROC curve and we propose two new methods (optimized averages and averages thresholds optimized) to estimating that region. We use the bootstrap method to search for a confidence region around the ROC curve. Using numerical examples, we apply the methods an compare their performance.