RESUMO
BACKGROUND: Breast cancer subtypes can be identified by genomic testing or pathology-based approximations. However, these classifications are not equivalent and the clinical relevance of both classifications needs to be fully explored. METHODS: Ninety-four patients were randomized to neoadjuvant single agent doxorubicin or docetaxel. Tumor subtype was assessed by pathology-based classification and by gene expression using the PAM50 plus the claudin-low predictor (CLP). Kappa Cohen's coefficient (κ) was used to test the agreement between methods. Multivariate Cox proportional hazards analyses were used to determine the significance of each methodology in the prediction of prognosis. Likelihood ratio statistics of both classifications were evaluated. RESULTS: The agreement between pathology-based classification and PAM50 was moderate [κ = 0.551, 95 % confidence interval (95 % CI) 0.467-0.641]. Tumor subtype assessed by both classifications were prognostic for overall survival (OS) and relapse-free survival (P < 0.05). However, PAM50 + CLP provided more prognostic information, in terms of OS, than the pathology-based classification (P < 0.05). Patients with triple negative tumors as well as basal-like tumors had worse OS when first treated with doxorubicin (HR = 5.98, 95 % CI 1.25-28.67, and HR = 5.02, 95 % CI 0.96-26.38, respectively). However, claudin-low tumors did not show significant differences in OS according to neoadjuvant treatment branch. Indeed, we found that claudin-low tumors treated with pre-operative doxorubicin had significantly better OS than basal-like tumors treated with neoadjuvant doxorubicin (adjusted HR = 0.16, 95 % CI 0.04-0.69, P = 0.014). CONCLUSIONS: The assignment of tumor subtype can differ depending on the methodology, which might have implications on patient's management and therapy selection.