ROC Curves in medical decision

Abstract The accurate medical diagnostic of a disease condition is fundamental for a correct medical decision. Disease screening programs are based, in general, in diagnostic tests which provide a binary response: a subject is classified as positive, if the test result is above a given threshold, and negative, otherwise. Therefore, false positive and false negative classifications can be generated. The performance of test can be evaluated by ROC curves which defined, for a given threshold, the compro-mise between Sensitivity and Specificity, i.e., the True and False Positive fractions. In this work, we address the problem of comparing two diagnostic systems where the corresponding ROC curves cross each other. A methodology is developed pro-viding a graphical display that identifies the regions where one curve is superior to the other, with the corresponding Sensitivity and Specificity regions.