Figure 4

Improved biomarker performance by accounting for classification robustness. (a-d) Marker performance improved dramatically when differentiating patients with identical classifications across all pre-processing schemes from the patients with ambiguous classifications for both the three-gene biomarker (a versus b) and the six-gene biomarker (c versus d) in the Director's Challenge dataset [21]. For all Kaplan-Meier curves; good prognosis patients are indicated by blue curves and poor prognosis patients by red curves. Hazard ratios and P-values are from stage-adjusted Cox proportional hazard ratio modeling followed by the Wald test.