P5-11-05: Measurement of Neoadjuvant Chemotherapy Tumor Response in Locally Advanced Breast Cancer by Three Methodologies. Correlation with Overall Survival.

Abstract

Background Measurement of residual disease following neoadjuvant chemotherapy that accurate predicts long-term survival in locally advanced breast cancer (LABC) is an essential requirement for new drugs efficacy evaluation. Several methods to assess neoadjuvant chemotherapy tumor response have been described. However, to our knowledge, agreement between methods and correlation with survival in independent prospective cohorts has not been reported. Patients & Methods: In this study we report neoadjuvant chemotherapy tumor response and survival in 151 consecutive LABC patients, included in a neoadjuvant clinical trial (http://www.clinicaltrials.gov; NCT00123929). Patients were randomized to either neoadjuvant docetaxel 100 mg/m2 every 21 days or neoadjuvant doxorubicin 75mg/m2, every 21 days, for 4 cycles. Following surgery, response was established according to three methodologies: the measurement of residual breast cancer burden (RBC) as described by Symman's (Symmans WF et al. J Clin Oncol. 2007;25:4414–22), Miller and Payne classification (Ogston KN et al. Breast. 2003;12:320–7) and RECIST criteria. Regarding to Symmans classification we have evaluate both RBC index, as a continuous variable, and RBC classes as a categorical variable (RBC-0,I,II,III). Kappa Cohen's coefficient (K) was used to test agreement between methods. We assessed the correlation between treatment outcome and overall survival (OS) by calculating the Harrell's C- statistic. Results: Median of follow up was 51.9 months. All three methods showed a moderate capacity to classify patients according to OS. The C-statistic to predict OS was 0.76 (IC: 0.67- 0.84) for RBC index and 0.71 (IC: 0.64−0.78) for RBC classes, 0.68 (IC: 0.58−0.78) for RECIST criteria and 0.69 (0.60−0.78) for Miller and Payne classification. Interesting, we did not encountered any death events within RCB-0 class. No significant differences were found between C-statistic when patients were stratified according to therapy. In order to assess the agreement between techniques, we grouped categories 1 and 2 of Miller and Payne classification in 1 category. The agreement between Miller and Payne classification and Symmans method was very high (K=0.87). In contrast, we found a moderate-fair agreement between Miller and Payne classification and RECIST criteria (K=0.46) and Symmans method and RECIST criteria (K=0.27). Conclusion: All three methods predicted fairly well OS. RCB-0 identified the best outcome group. The agreement between methodologies based in pathology analyses was very high. However, the agreement falls off when these methodologies were compared with RECIST criteria. Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P5-11-05.