Comparison of three mathematical models for prediction of non-sentinel lymph node metastasis in patients with breast cancer and positive sentinel lymph node

Abstract

Objective: It was to compare a “Brazilian Nomogram” developed for this study, the “Memorial Sloan-Kettering Cancer Center Nomogram” and the “Stanford Online Calculator” for predicting the presence of metastasis in non-sentinel lymph nodes in Brazilian patients with breast cancer and positive sentinel lymph node. Methods: We included 261 patients with positive sentinel lymph node biopsy and submitted to complete axillary lymph node dissection in two Brazilian centers for treatment of breast cancer; 92 (35%) of these women had metastasis in non-sentinel lymph nodes. The “Brazilian Nomogram”, the “Nomogram from Memorial Sloan Kettering Cancer Center” and the “Stanford Online Calculator” were applied to calculate the probability of metastasis in non-sentinel lymph nodes. The accuracy of each model was calculated by the area under the ROC curve (AUC). Results: The “Brazilian Nomogram”, the “Nomogram from Memorial Sloan Kettering Cancer Center” and the “Stanford Online Calculator” showed AUC values of 0.694 (95%CI 0.629–0.760, p<0.0001), 0.626 (95%CI 0.555–0.697, p=0.001) and 0.619 (95%CI 0.548–0.689, p=0.002), respectively. Conclusions: All three models showed a reasonable discrimination in predicting metastasis in non-sentinel lymph nodes in patients with breast cancer in this study. The “Brazilian Nomogram” had the best accuracy among the models tested, but needs to be validated in population groups different from those used in its construction.