The discrimination of a risk prediction model measures that model's ability to distinguish between subjects with and without events. The area under the receiver operating characteristic curve (AUC) is a popular measure of discrimination. However, the AUC has recently been criticized for its insensitivity in model comparisons in which the baseline model has performed well. Thus, 2 other measures have been proposed to capture improvement in discrimination for nested models: the integrated discrimination improvement and the continuous net reclassification improvement. In the present study, the authors use mathematical relations and numerical simulations to quantify the improvement in discrimination offered by candidate markers of different strengths as measured by their effect sizes. They demonstrate that the increase in the AUC depends on the strength of the baseline model, which is true to a lesser degree for the integrated discrimination improvement. On the other hand, the continuous net reclassification improvement depends only on the effect size of the candidate variable and its correlation with other predictors. These measures are illustrated using the Framingham model for incident atrial fibrillation. The authors conclude that the increase in the AUC, integrated discrimination improvement, and net reclassification improvement offer complementary information and thus recommend reporting all 3 alongside measures characterizing the performance of the final model.

While the authors agree with John Ioannidis that "most research findings are false," here they show that replication of research findings enhances the positive predictive value of research findings being true.

No abstract

The area under the receiver operating characteristic (ROC) curve (AUC) is commonly used for assessing the discriminative ability of prediction models even though the measure is criticized for being clinically irrelevant and lacking an intuitive interpretation. Every tutorial explains how the coordinates of the ROC curve are obtained from the risk distributions of diseased and non-diseased individuals, but it has not become common sense that therewith the ROC plot is just another way of presenting these risk distributions. We show how the ROC curve is an alternative way to present risk distributions of diseased and non-diseased individuals and how the shape of the ROC curve informs about the overlap of the risk distributions. For example, ROC curves are rounded when the prediction model included variables with similar effect on disease risk and have an angle when, for example, one binary risk factor has a stronger effect; and ROC curves are stepped rather than smooth when the sample size or incidence is low, when the prediction model is based on a relatively small set of categorical predictors. This alternative perspective on the ROC plot invalidates most purported limitations of the AUC and attributes others to the underlying risk distributions. AUC is a measure of the discriminative ability of prediction models. The assessment of prediction models should be supplemented with other metrics to assess their clinical utility.