CHEOS Work in Progress Seminar | From ROC to RIC: New metrics for documenting the clinical utility of (bio)markers and risk prediction scores

Mohsen Sadatsafavi, University of BC

Wednesday, Nov 21, 2018
Hurlburt Auditorium, St Paul’s Hospital

The Precision Medicine movement and arrival of vast “omics” data have accelerated the development of (bio)markers and clinical prediction models. There are many tools for communicat­ing the capacity of a marker in risk concentration, including the Receiver Operating Characteristic (ROC) curve, Predictiveness Curve, and Lorenz Curve. However, these metrics largely focus on studying ‘risk’. It is time to move on from the estimates of concentration risk to estimates of ‘concentration of benefit’ in marker development. We introduce a new graphical method: the Relative Impact Characteristic (RIC) curve. RIC visually rep­resents the concept of concentration of benefit when the marker is used to inform a specific treatment decision. The proposed­ tool juxtaposes two fundamental quantities associated with marker implementation: the population-level impact of marker-informed treatment and the relative size of the treated popula­tion. We establish analogies between the ROC and the RIC curves around the interpretations of shape, slopes, and area under the curve. A case study in COPD shows how the data from a clinical trial can be used to construct the RIC curve and evaluate the clinical utility of a prediction model for exacerbations.

More information