Figure 2.

Sample breakdown. A total of 858 samples passed all filtering steps including 487 samples with 10-year follow-up data (213 relapse; 274 no relapse). The remaining 371 samples had insufficient follow-up for 10-year classification analysis but were retained for use in survival analysis. The 858 samples were broken into two-thirds training and one-third testing sets resulting in: a training set of 572 samples for use in survival analysis and 325 samples with 10-year follow-up (143 relapse; 182 no relapse) for classification analysis; and a testing set of 286 samples for use in survival analysis and 162 samples with 10-year follow-up (70 relapse; 92 no relapse) for classification analysis.