[Discriminative ability, reclassification and net benefit: the predictive value of biomarkers]

More and more biomarkers are becoming available to predict the risk of disease. Several performance measures are used to express the predictive value of a biomarker. In this article we use as an example the Framingham risk score with and without the HDL cholesterol concentration as a risk factor for the prediction of the occurrence of cardiovascular disease. We consider 3 performance measures for diagnostic or prognostic quality: (a) the area under the Receiver Operating Characteristic (ROC) curve (AUC), a traditional measure to indicate discriminative ability; (b) the Net Reclassification Improvement (NRI), a popular measure based on the improvements in the high- and low-risk classification of patients by adding the HDL value as a risk factor; (c) Net Benefit, a measure based on decision sciences. Net benefit is a better measure for the value of a biomarker, since it takes more of the clinical context into account and is easier to interpret than the AUC and the NRI.