1. ¹CNR-IBIM, Clinical Epidemiology and Physiopathology of Renal Diseases and Hypertension of Reggio Calabria, Reggio Calabria, Italy
2. ²ERA–EDTA Registry, Department of Medical Informatics, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands
3. ³Department of Clinical Epidemiology, Leiden University Medical Centre, Leiden, The Netherlands

Correspondence: Giovanni Tripepi, CNR-IBIM, Istituto di Biomedicina, Epidemiologia Clinica e Fisiopatologia, delle Malattie Renali e dell'Ipertensione Arteriosa, c/o Euroline di Ascrizzi Vincenzo, Via Vallone Petrara n. 55/57, 89124 Reggio Calabria, Italy. E-mail: gtripepi@ibim.cnr.it

Received 21 February 2009; Revised 7 April 2009; Accepted 15 April 2009; Published online 20 May 2009.

Abstract

Clinical practice commonly demands 'yes or no' decisions; and for this reason a clinician frequently needs to convert a continuous diagnostic test into a dichotomous test. Receiver operating characteristic (ROC) curve analysis is an important test for assessing the diagnostic accuracy (or discrimination performance) of quantitative tests throughout the whole range of their possible values, and it helps to identify the optimal cutoff value. The value of this analysis is not confined to diagnosis in that it may also be applied to assess the prognostic value of biomarkers and to compare their predictive value. ROC curve analysis may also serve to estimate the accuracy of multivariate risk scores aimed at categorizing individuals as affected/unaffected by a given disease/condition. However, one should be aware that, when applied to prognostic questions, ROC curves don't consider time to event and right censoring, and may therefore produce results that differ from those provided by classical survival analysis techniques like Kaplan–Meier or Cox regression analyses.