We hypothesized that amplification or overexpression of HER-2 (c-erbB-2), the Ki-67 antigen (Mib1), cyclin D-1 (CD1), interleukin-6 (IL-6), or the transforming growth factor beta II receptor, (TGFbetaRII), would predict relapse in women with early stage, estrogen (ER) and/or progesterone receptor (PR) positive breast cancer treated with tamoxifen. Conditional logistic regression models and a new novel analytic method - support vector machines (SVM) were used to assess the effect of multiple variables on treatment outcome. All patients had stage I-IIIa breast cancer (AJCC version 5). We paired 63 patients who were disease-free on or after tamoxifen with 63 patients who had relapsed (total 126); both disease-free and relapsed patients were matched by duration of tamoxifen therapy and time to recurrence. These 126 patients also served as the training set for SVM analysis and 18 other patients used as a validation set for SVM. In a multivariate analysis, larger tumor size, increasing extent of lymph node involvement, and poorer tumor grade were significant predictors of relapse. When HER-2 or CD1 were added to the model both were borderline significant predictors of relapse. The SVM model, after including all of the clinical and marker variables in the 126 patients as a training set, correctly predicted relapse in 78% of the 18 patient validation samples. In this trial, HER-2 and CD1 proved of borderline significance as predictive factors for recurrence on tamoxifen. An SVM model that included all clinical and biologic variables correctly predicted relapse in >75% of patients.