The area under the ROC curve (AUC) is commonly used as a measure for the discriminative performance of NTCP models. Here, we demonstrate that for typical patient cohorts, the AUC is an unsuitable measure for that purpose since it is typically limited to values below 0.8 and it exhibits large statistical variation.
Keywords: AUC; Model validation; NTCP; Normal tissue complications; ROC curve.
Copyright © 2019 Elsevier B.V. All rights reserved.