Clinical-Genomic Models of Node-Positive Breast Cancer: Training, Testing, and Validation

Abstract Purpose There is no useful model for predicting the risk of recurrence in node-positive patients regardless of breast cancer subtype. We developed and validated two clinical-genomic models (recurrence index [RI]-local recurrence [LR]) and -distant recurrence (RI-DR) for stratifying these patients into low- and high-risk groups. Patients and methods The four datasets were: (1) training group (n = 112); (2) testing group (n = 46); (3) validation group (n = 388); and (4) E-MTAB-365 dataset (n = 426). Patients who had undergone mastectomy or breast-conserving surgery and mRNA microarray analysis of their primary tumor tissue with a pathological stage of I–III were enrolled in the training, testing, and validation groups. Using pre-set cut-offs obtained from the training group, the models were tested and validated in the three other independent groups. Results In the validation dataset, the RI-LR distinguished between low- and high-risk groups according to 10-year LR-free interval (LRFI; 100% vs. 93.0%, P = 0.015) and relapse-free survival (RFS; 85.0% vs. 76.9%, P = 0.032). The RI-DR distinguished the low- from the high-risk group according to RFS (85.7% vs. 77.4%, P = 0.025). RI-DR and -LR scores were independent prognostic factors in N1–N2 patients (hazard ratio [HR] = 3.3, 95% confidence interval [95% CI]: 1.1–10.2; and HR = 2.7, 95% CI: 1.1–6.7, respectively) by multivariate analysis. The RI-DR- and -LR genetic models were tested similarly using the E-MTAB dataset with HRs of 2.5 ( P = 0.0048) and 2.7 ( P = 0.0285), respectively, in node-positive patients. Conclusion Both RI-DR and -LR can partition N1–N2 patients into low- and high-risk groups for RFS; however, the latter is superior for predicting local/regional recurrence.