SEP: score for expression profile-a novel method for predicting clinical outcome in breast cancer

Several studies have attempted to use expression profile technique as a prognostic indicator of breast cancer. However, the candidate gene lists derived by these studies have low consistency with each other. By re-analyzing published data utilizing a new method, we tested the hypothesis that when using expression profile to predict breast cancer outcome, other clinical factors should be considered. Our results demonstrate that expression profiles are not independent of tumor size, histological grade, or estrogen receptor (ER) and progesterone receptor (PR) status. Therefore, we applied a partial correlation analysis to control out ER status from expression data and derived a list of 124 genes as having prognostic value. Expression profile based on this gene list correctly predicted the outcomes for 83% patients without any optimization procedure. Further investigation of this group of genes found that many belonged to the spindle checkpoint pathway regulating cell cycle and these genes may play an important role in breast cancer recurrence. We conclude that to correctly evaluate the correlation between gene and breast cancer outcome, clinical scenario maybe important. Therefore, the combination of expression data with other prognostic factors could achieve better prognostic predictions.