Prognosis of breast cancer : a survival analysis of 1184 patients with 4-10 years follow-up, illustrating the relative importance of estrogen receptors, axillary nodes, clinical stage and tumor necrosis

Prognostic indicators, measured at diagnosis, are important in breast cancer. They help clinicians select optimal treatment, provide rational bases for stratification of treatment trials and assist analysis of response to treatment. Univariate statistical survival curves have identified many such indicators. However, they do not explain why some patients, classified as favoured by one or other factor(s), experience early treatment failure, nor why a substantial number with unfavourable signs remain recurrence-free many years later. This study was undertaken to identify independent prognostic factors with the use of multivariate regression. A Cox proportional hazards model of disease-specific survival was based on 1184 primary breast cancer patients referred to the Cancer Control Agency of B.C. between 1975 and 1981 (median follow-up 60 months). Significant univariate associations with overall survival were found for estrogen receptor concentration ([ER]), axillary nodal status (NO, Nl-3, N4+), clinical stage (TNM I, II, III, IV), histologic differentiation and confluent tumor necrosis (minimal, marked). These factors were assessed at primary diagnosis. A subset of 859 patients with complete data on these variables and also histologic type, menopausal status, age, tumor size and treatment was used to fit the multivariate model. Nodal status was the most important independent factor but three others, TNM stage, [ER] and tumor necrosis, were needed to make adequate predictions. A derived Hazard Index defined risk groups with 8-fold variation in survival. Five-year predicted survival ranged from 36% (N4+, loge[ER]=0, marked necrosis) to 96% (NO, loge[ER]=6, no necrosis) with TNM I and 0% to 70% for the same categories in TNM IV. This wide variation occurred across all stages. Study of post-recurrence survival (369 patients) yielded a model with only three independent predictors: [ER], nodal status and tumor necrosis. Survival - overall, recurrence-free and post-recurrent - is predictable by modelling a few factors measureable at diagnosis. Use of ER concentration, rather than the more common ER status (+ or -), greatly strengthens the model. Presence of ER was also shown to be increasingly important as 'protective', attenuating the effect of other factors, as risk of mortality increases.Medicine, Faculty ofPathology and Laboratory Medicine, Department ofGraduat