Validation of a novel prognostic staging system for breast cancer incorporating age at diagnosis

Abstract

Age at diagnosis is an independent prognostic variable for breast cancer-specific and overall survival (OS). We previously developed a novel prognostic staging system that incorporates age and demonstrated refined risk stratification compared with the current American Joint Committee on Cancer (AJCC) staging schema. We now aim to externally validate this staging system. The National Cancer Database was used to identify adult females diagnosed with invasive breast cancer from 2010-2015. Women with prior malignancy, unknown vital status, or unknown AJCC clinical prognostic stage (CPS) variables were excluded. Serial multivariable Cox’s proportional hazards models evaluated associations between OS and CPS +/- age +/- effect modification (CPS by age interaction). Models were evaluated using concordance index (C-index). Among 663,659 patients, the median age was 60 years (IQR 51-70) and the median follow-up was 91.3 months. At last follow-up, 133,273 patients (20.1%) were dead. The model that incorporated both age and effect modification had the highest C-index (0.7852, versus 0.6881 for the model excluding age and 0.7847 for the model without effect modification), indicative of the best predictive performance. Using this model to predict OS stratified by CPS showed differential survival across the age spectrum, consistent with the initial model. Within each stage group, women diagnosed at age 40 had the best survival, whereas women at the extremes of age had inferior survival. This analysis provides external validation of our novel prognostic staging system for breast cancer. Future editions of AJCC may consider incorporating age to achieve more accurate survival predictions.