Figure 4.

Validation of pathway-based breast cancer classifiers constructed from the optimal significant genes. To find the optimal number of genes as a signature, ROC analyses, with 5-year DMFS as defining point, with an increasing number of genes were conducted in the training set of ER-positive tumors or ER-negative tumors. For ER-positive tumors, in the "apoptosis" pathway, 24 genes (reaching an AUC of 0.784) were considered optimal (Table 3). For the "regulation of cell cycle pathway" in ER-positive tumors, 17 genes (AUC of 0.777) were considered optimal (Table 4). For ER-negative tumors, the optimal number of genes was 7 (AUC of 0.790) for the "regulation for cell growth" pathway (Table 5), and 5 (AUC of 0.788) for the "regulation of G-protein coupled receptor signaling" pathway (Table 6), respectively. The selected genes for the top 2 pathways for ER-positive and ER-negative tumors were subsequently used to construct prognostic gene signatures separately for the 2 ER-subgroups of tumors. The 152-patient test set [23] consisted of 125 ER-positive tumors and 27 ER-negative tumors based on the expression level of ER gene on the chip. (A) ROC (Left) and Kaplan-Meier (Right) analysis of the 38-gene signature for ER-positive tumors. Thirteen patients with less than 5-year follow-up were excluded from ROC analysis. (B) ROC (Left) and Kaplan-Meier (Right) analysis of the 12-gene signature for ER-negative tumors. One patient with less than 5-year follow-up was excluded from ROC analysis. (C) ROC (Left) and Kaplan-Meier (Right) analysis of a combined 50-gene signature for ER-positive and ER-negative tumors. Fourteen patients with less than 5-year follow-up were excluded from ROC analysis.