Random genes lists have disparate prognostic frequencies across breast cancer gene expression data sets. Each graph is a histogram showing the percentage of random gene lists (70 genes/list) that were significantly predictive of metastatic recurrence latencies in 4 gene expression data sets. A. The bars represent analyses based on 1,000 random lists of 70 genes derived from all genes within a data set. B. Cox regression analyses based on 5,000 random gene lists (70 genes/list) selected from 11,318 common unigenes IDs for either a univariate or multivariable model adjusting for expression of the ER-α gene. C. Analyses of tumors stratified by ER status based on 5,000 random gene lists (70 genes/list) selected from 11,318 genes that had a unique unigene identifier common to all 4 data sets. The same 5,000 lists were separately evaluated in either ER positive (ER+) or ER negative (ER-) tumors within each data set.