Nomograms for predicting the likelihood of non-sentinel lymph node metastases in breast cancer patients with a positive sentinel node biopsy

BACKGROUND: Breast cancer patients with sentinel lymph node (SLN) metastases may have a low risk of non-SLN metastases. Accurate estimates of the likelihood of additional disease in the non-SLN metastases can avoid many complications mentioned the axillary lymph node dissection (ALND). This study aims to develop a new model based on Chinese real-world patients to ascertain the likelihood of non-SLN metastases in a breast cancer patient with disease-positive SLN, enabling the surgeons to make a better choice of surgical procedures. METHODS: Out of the 470 patients from CSCO Breast Cancer Database collaborated Group, a proportion of 3 (347 cases): 1 (123 cases) was considered for assigning patients to training and validation groups, respectively. Two training models were created to predict the likelihood of having additional, non-SLN metastases in an individual patient. Training model 1 was created with pathological size of the tumor, pathological type, lymphovascular invasion, the number of positive SLNs/number of total SLNs ratio, and the Her-2 status based on multivariable logistic regression (P < .05). Training model 2 was based on the variables in model 1 and age, estrogen receptor status, progesterone receptor status, Ki-67 count, menopause status. RESULTS: The area under the receiver operating characteristic (ROC) curve of the training model 1 was 0.754, while the area of training model 2 was 0.766. There was no difference between model 1 and model 2 regarding the ROC curve, P = .243. Next, the validation cohort (n = 123) was developed to confirm the model 1's performance and the ROC curve was 0.703. The nomogram achieved good concordance indexes of 0.754 (95% CI, 0.702–0.807) and 0.703 (95% CI, 0.609–0.796) in predicting the non-SLN metastases in the training and validation cohorts, respectively, with well-fitted calibration curves. The positive and negative predictive values of the nomogram were calculated, resulting in positive values of 59.3% and 48.6% and negative predictive values of 79.7% and 83.0% for the training and validation cohorts, respectively. CONCLUSION: We developed 2 models that used information commonly available to the surgeon to calculate the likelihood of having non-SLN metastases in an individual patient. The numbers of variables in model 1 were less than in model 2, while model 1 had similar results as model 2 in calculating the likelihood of having non-SLN metastases in an individual patient. Model 1 was more user-friendly nomogram than model 2. Using model 1, the risk for an individual patient having ALND could be determined, which would lead to a rational therapeutic choice.