Estimated likelihood of relapse at 10 years for any RFRS value. The likelihood of relapse was calculated in the training dataset (n = 505) for 50 RFRS intervals (from 0 to 1). A smooth curve was fitted using a loess function and 95% confidence intervals plotted to represent the error in the fit. Short vertical marks just above the x axis, one for each patient, represent the distribution of RFRS values observed in the training data. Thresholds for risk groups are indicated. The plot shows a linear relationship between RFRS and likelihood of relapse at 10 years with the likelihood ranging from approximately 0 to 40%.