This study was designed to identify predictive signatures of pathological complete response (pCR) in breast cancer treated by taxane-based regimen, using clinicopathological variables and transcriptomic data (Affymetrix Hgu133 Plus 2.0 devices). The REMAGUS 02 trial (n = 153,training set) and the publicly available M.D. Anderson data set (n = 133, validation set) were used. A re-sampling method was applied. All predictive models were defined using logistic regression and their classification performances were tested through Area Under the Curve (AUC) estimation. A stable set of 42 probesets (31 genes) differentiate pCR or no pCR samples. Single-or 2-probesets signatures, mainly related to ER pathway, were equally predictive of pCR with AUC greater then 0.80. Models including probesets associated with ESR1, MAPT, CA12 or PIGH presented good classification performances. When clinical variables were entered into the model, only CA12 and PIGH, remained informative (p = 0.05 and p = 0.005) showing that a combination of a few genes provided robust and reliable prediction of pCR.
Keywords: Breast cancer; Neoadjuvant setting; Pathological complete response; Prediction; Transcriptome.
Copyright © 2013 Elsevier Ltd. All rights reserved.