Validation of the newly proposed American Joint Committee on Cancer (AJCC) breast cancer prognostic staging group and proposing a new staging system using the National Cancer Database.

Abstract

The eighth edition of AJCC cancer staging manual incorporated biomarker status into the prognostic staging group (PSG). We used data from National Cancer Database (NCDB) to validate and improve the PSG. All patients had surgery and at least some systemic treatment (endocrine therapy, chemotherapy or HER2 targeted therapy). Information from 420,520 patients was assessed for potential predictors of overall survival (OS), including age at diagnosis (age), tumor grade (G), hormonal receptor and HER2 status, and presence of lymph vascular invasion (LVI), stratified by stage or sub-stages. Based on the multivariate Cox analyses, we built different point systems to predict OS and evaluated the different point systems by Akaike's information criterion (AIC), Harrell's concordance index (C-index), and Uno's concordance index. Age, G, hormonal receptor and HER2 status, LVI and being TNBC were significantly associated with OS (all P < 0.0001). Three staging systems were correlated with OS: system 1 was the conventional anatomic TNM staging; system 2 included TNM, age, G, hormonal receptor, HER2, and LVI; system 3 included TNM, age, G, TNBC versus non-TNBC, and LVI. System 3 (C-index; 0.7316; AIC: 488138.91) achieved the best balance between predictive performance and goodness-of-fit to the NCDB data as compared to system 2 (C-index: 0.7325; AIC: 498087.73) and system 1 (C-index: 0.716; AIC: 688536.49). The new PSG is a better staging system than the conventional anatomic TNM system. Grouping breast cancer into TNBC versus non-TNBC may be simpler while retaining similar accuracy as using ER/PR/HER2 status to predict OS.