Figure 4

Kaplan-Meier plots illustrating separation among the 1000 test data sets. Predictions of low or high risk were based on Cox regression models fitted to each training set, either using gene expression (GE) only ( A = overall survival (OS), D = disease free survival (DFS)), clinical data only ( B = OS, E = DFS), or both gene expression and clinical data ( C = OS, F = DFS). Clinical data included for both outcomes was patient stage of disease at diagnosis (1, 2, and 3 or 4), ER status (+/-), and PR status (+/-). Genes included in both the OS and DFS models were PGR, GABRP, TBC1D9, SLC39A6 and LRBA, while NAT1 was also included in the model for DFS.