Relative impact characteristic curve: a graphical tool to visualize and quantify the clinical utility and population-level consequences of implementing markers.

Purpose: Receiver operating characteristic (ROC) curve analysis is a popular method for evaluating the performance of (bio)markers. However, the standard ROC curve does not directly connect marker performance to patient-related outcomes. Our aim was to fill this gap by proposing a conceptually similar graphical tool that carries information about the clinical uitility of markers.

Methods: We propose a novel graphical tool, the relative impact characteristic (RIC) curve, that depicts the trade-off between the population-level impact of treatment as a function of the size of the treated population for a given marker positivity rule (e.g., a threshold). We establish analogies between the ROC and the RIC curves around the interpretations of shape, slopes, and area under the curve and discuss parametric inference on RIC.

Results: As a case study, we used data from a clinical trial on preventive therapy for exacerbations of chronic obstructive pulmonary disease. We illustrate how the RIC curve can be constructed for a predication score and be interpreted in terms of a marker's ability toward concentrating treatment benefit in the population. We discuss hoe the RIC curve can be used to identify a threshold on the risk score based on the maximal acceptable number-needed-to-treat.

Conclusions: The RIC curve enables evaluation of markers in terms of their treatment-related clinical utility. Its analogies with the standard ROC analysis can facilitate its interpretation, bringing a population-based perspective to the activities of diverse marker development and evaluation teams.