Fig. 2

Performance measures for event-free survival. Each of the 18 models are assessed using the testing dataset. Four measures of performance are considered: the adjusted root mean squared prediction error (RMSE); the logrank test statistic from using the predicted survival time as a classifier on high-risk patients, thresholded at 2 years (LPS2) and 5 years (LPS5); and Harrell’s c-index. 95% confidence intervals are obtained by bootstraping on the testing dataset. This is done by resampling observations with replacement and recomputing each measure. The process is repeated for B=1000 times, and the middle 95% of measures are used for the confidence interval. Note, the upper limit of RMSE for T-2 is not visible in the plot