Abstract
In current clinical practice for breast cancer patients, estrogen (ER) and progesterone receptor (PgR) concentrations, quantified by the dextran-coated charcoal assay, are categorized by an arbitrary cutoff into a negative or positive status. However, although the results obtained with this approach are easy to interpret, such a representation could oversimplify the relationship between ER and PgR content and patient outcome and imply an assumption of monotonicity, which is generally expected but rarely proven. We evaluated the relationship between ER and PgR content (considered on a continuous scale) and clinical outcome, using a flexible statistical model, in a group of postmenopausal patients with N-positive operable tumors who were submitted to surgery and different adjuvant treatments (tamoxifen or CMF). Univariate analysis indicated that in the tamoxifen-treated group, ER level, number of metastatic nodes (pN) and age, but not PgR, were significant indicators of clinical outcome (p=0.032, p=0.021 and p=0.029, respectively). Multivariate analysis indicated that in this group of patients there was no interaction between variables, and in the final model for disease-free survival (DFS) only ER and pN were retained with an overall predictive ability of the regression model of 0.723, as evaluated by Harrell's c. However, pN markedly contributed to the predictive ability of the model with respect to ER, since a marked decrease in Harrell's c statistic (c=0.582) was observed when pN was removed from the model. In the CMF-treated group, only pN affected clinical outcome. When the estimated DFS curves obtained from the final Cox regression models were plotted according to four values of ER (in the tamoxifen-treated group) or three values of pN (in the CMF-treated group) we observed that in the tamoxifen-treated group patients with an ER concentration equal to 0 fmol/mg cytosol protein had the worst prognosis, whereas a marked improvement of the expected DFS was observed for patients with a low but detectable ER level (generally classified as ER-negative because falling below the conventional cutoff value of 10 fmol/mg cytosol protein). Our results seem to suggest that the use of steroid receptor concentrations on a continuous scale, instead of dichotomous “status”, is to be preferred in the choice of adequate therapeutic strategies.