Validation and comparison of models to predict non-sentinel lymph node metastasis in breast cancer patients

Several models for predicting the risk of non-sentinel lymph node (NSLN) metastasis in breast cancer patients with positive sentinel lymph nodes (SLNs) have been developed. The purpose of this study was to validate and compare these models in Chinese patients. A total of 159 breast cancer patients with positive SLNs treated at our institution were included. Among them, 81 (50.9%) patients had at least one NSLN involvement. The Cambridge, Mou, Mayo, Tenon, MDA, Memorial Sloan-Kettering Cancer Center (MSKCC), Ljubljana, SNUH, Turkish, Louisville, Stanford, and Saidi models were evaluated and compared using receiver operating characteristic (ROC) curves, calibration plots, and false negative (FN) rates. The Cambridge and Mou models outperformed the others, both with area under the ROC curves (AUCs) of 0.73. The Mayo, Tenon, MDA, MSKCC, Turkish, Ljubljana, SNUH, and Louisville models had AUCs of 0.68, 0.66, 0.66, 0.64, 0.63, 0.62, 0.61, and 0.60, respectively. The Stanford and Saidi models did not present any discriminative capabilities, with AUCs of 0.54 and 0.50, respectively. The Cambridge, MSKCC, and Mayo models were well calibrated. With adjusted thresholds, the Mayo model outperformed the others by classifying the highest proportion of patients (20%) into the low-risk group. Our study revealed that the Cambridge and Mou models performed well in Chinese patients. The ROC curves, calibration plots, and FN rates should be used together for the accurate evaluation of prediction models. Selection of these models should be based on the clinicopathological features of the targeted population. The models specifically designed for patients with micrometastases or macrometastases of SLNs are needed in the future.