Refining breast cancer prognosis by incorporating age at diagnosis into clinical prognostic staging: introduction of a novel online calculator.

Abstract

American Joint Committee on Cancer (AJCC) clinical staging is used to estimate breast cancer prognosis, but individual patient survival within each stage varies considerably by age at diagnosis. We hypothesized that the addition of age at diagnosis to the staging schema will enable more refined risk stratification. We performed a retrospective population analysis of adult women diagnosed with invasive breast cancer between 2010 and 2015 registered in SEER. Multivariable Cox hazards models were used to evaluate the association of AJCC 8th edition clinical prognostic stage (CPS) and age with risk of overall mortality. Separate hierarchical models were fit to the data: Model 1: CPS alone; Model 2: CPS + age + age2; and Model 3: CPS + age + age2 + CPS x age + CPS x age2. Models were compared by the Akaike information criterion (AIC), the c-statistic for time-dependent receiver operator characteristic curves, and category-free net reclassification improvement (NRI). Internal validation was performed using bootstrapping samples. Among 86,637 women, the median follow-up was 36months and 3-year overall survival was 91.9% ± 0.1%. Age significantly modified the effect of CPS on survival (p < 0.0001). Model 3 was the most precise, with the lowest AIC (126,619.63), the highest c-statistic (0.8212, standard error 0.0187), and superior NRI indices. Age at diagnosis is a highly prognostic variable that warrants consideration for inclusion in future editions of the AJCC Breast Cancer Staging Manual.