Gene expression profiles predict pathologic complete response to preoperative chemotherapy with gemcitabine, epirubicin and docetaxel in primary breast cancer
Andreas Schneeweiß, O. Thuerigen, Grischa Toedt et al. · Journal of Clinical Oncology · 2005
2001 Background: While primary systemic therapy (PST) has gained considerable attention, it is impossible to predict sensitivity of primary breast cancer (PBC) to specific drugs. This study was designed to discover a gene expression profile that predicts pathologic complete response (pCR) to treatment with gemcitabine (G), epirubicin (E) and docetaxel (Doc) administered either as triple therapy (GEDoc) or sequentially (GE→Doc) for a total of 18 weeks as PST. Patients and Methods: Patients (pts) with PBC T2–4 N0–2 M0 underwent pretreatment core needle biopsy to collect snap frozen tissue for microarray analysis. Small amounts of isolated breast tumor RNA were amplified with a newly developed amplification protocol. Comprehensive RNA expression profiles were generated using DNA microarrays carrying 21.329 unique gene specific 70mer oligonucleotide probes. Analysis was started with a training data set of 22 pts (pCR n=5; non-pCR n=17). Several supervised learning methods were tested to identify the best class prediction algorithm. Results were validated on an independent data set of 48 PBCs (pCR n=9; non-pCR n=39). Results: A support vector machine (SVM) using linear kernel yielded the best estimated accuracy. Classification performance of the linear SVM was assessed by leave-one out validation on the training data set for the purpose of model selection. Subgroups of genes were defined by their discriminatory power and their classification error rates were validated by a permutation test. The permutation test showed that the achieved classification error rates were significantly better as what could be expected by chance. By applying our classifier to the independent test set, prediction of therapy response was achieved with an overall accuracy of 71%, a sensitivity of 88% and a specificity of 67%. Thus, 8 of the 9 pts with pCR were correctly predicted. Conclusions: Our results suggest that transcriptional profiling can identify a gene expression profile in PBC that may serve as a clinically useful predictor of pCR to PST containing G, E and Doc. Thus, a highly efficient selection of pts possibly benefitting from these drugs could be achieved. Author Disclosure Employment or Leadership Consultant or Advisory Role Stock Ownership Honoraria Research Funding Expert Testimony Other Remuneration Aventis, Lilly Oncology Lilly Oncology
Cited by 1 works. doi:10.1200/jco.2005.23.16_suppl.2001