Blinded Validation Study of Genomic Predictions for Survival Following Adjuvant Sequential Anthracycline-Docetaxel Chemotherapy with or Without Endocrine Therapy.

1030 Background: The adjuvant chemotherapy +/- endocrine survival (ACES) predictor is a microarray-based genomic algorithm that combines signatures for sensitivity to endocrine therapy (SET), chemo-resistance, and chemo-sensitivity to provide a binary prediction of sensitivity or insensitivity to adjuvant chemotherapy +/- endocrine therapy (Symmans et al, SABCS, 2010). METHODS Eligible patients received ≥ 5 cycles of sequential anthracyline-taxane chemotherapy for HER2-negative cancer diagnosed ≥ 3 years ago. Sites sent de-identified frozen primary tumor samples to MDACC for gene expression profiling (Affymetrix U133A) for predictions of ACES, pCR (DLDA30), genomic grade (GGI), and intrinsic subtype (PAM50). Clinical information (from sites) and prediction results (from MDACC) were matched and analyzed at Georgetown. Distant relapse-free survival at 3 years (3yr-DRFS), and absolute risk reduction (ARR), for patients predicted to be treatment (Rx)-sensitive (versus Rx-insensitive) were evaluated with 95% confidence interval (CI). Estimated sample size: 165 patients for 80% power to detect 3yr-DRFS >80% if predicted to be Rx-sensitive, at 5% one-sided significance level. RESULTS Follow up ≥ 3yr was known for 101 (95%) of 106 evaluable patients (3yr-DRFS 80%) with AJCC Stage I (24%), IIA (41%), IIB (24%), or III (12%). Chemotherapy was fluorouracil, epirubicin, cyclophosphamide (FEC), followed by docetaxel in 100, or paclitaxel in 6 patients. Patients with ACES prediction of Rx-sensitive (39%) had 3yr-DRFS 90% (CI 80-99) and absolute risk reduction (ARR) 17% (CI 2-30), compared to predicted Rx-insensitive. In contrast, patients predicted to be chemo-sensitive by DLDA30 (pCR, 19%), GGI (high, 59%), or PAM50 (basal/luminal B, 30%) had 3yr-DRFS of 47% (CI 25-70), 74% (CI 63-85), and 67% (CI 51-84); and ARR of -40% (CI -64 to -15), -15% (CI -28 to 2), and -18% (CI -36 to 1), respectively (negative ARR means increased risk). CONCLUSIONS ACES prediction of Rx-sensitive breast cancer was associated with excellent and significantly improved 3yr-DRFS, whereas three other published predictors of response had paradoxically worse DRFS.