Analysis of cut-off criteria in ROC curve for endarterectomy decision making

A method to compute the atherosclerotic plaque activity index (AI) was recently proposed in the literature to support the decision of endarterectomy. The dependency of this index on objective measures, extracted from Ultrasound and Clinical Data, aims at the increasing of the diagnosis accuracy by removing part of the subjective component associated with the human decision. The cut-off used to decide the surgical intervention for plaque removal depends on several factors such as population specificities, activity index computation algorithm and on the optimality criterion used for the selection of that threshold. A common naive approach to select the optimal cut-off value assumes the same cost on the reduction of False Positives Rate (FPR) and increasing of the True Positive Rate (TPR) which means finding the intersection of the ROC curve with the descending diagonal. In the Youden index, another widespread ROC based criterion, the optimal cut-off is chosen in order to maximize the difference between TPR and FPR. In this paper a preliminary statistical analysis of these naive approaches is presented, in order to suggest more sophisticated and complex criteria that would take into account other factors.