Prediction Models for Sentinel Lymph Node Metastasis in Clinically Node-Negative Breast Cancer: Validation of Existing Nomograms, Model Development, and Ensemble Evaluation.

backgroundAxillary management in early breast cancer (EBC) is increasingly moving toward surgical de-escalation. While sentinel lymph node biopsy (SLNB) has reduced morbidity compared with axillary lymph node dissection, it still carries risks and may be unnecessary in selected patients with very low nodal risk. An accurate estimate of the risk of sentinel lymph node (SLN) metastasis may support risk-adapted omission strategies. This study aimed to validate existing prediction models, develop a cohort-derived model, and evaluate whether ensemble modeling improves the prediction of SLN macrometastasis in clinically node-negative (cN0) EBCs.

methodsWe conducted a retrospective cohort study of 1080 women with cN0 EBCs treated at a tertiary center between 2012 and 2020. Existing prediction tools (Memorial Sloan Kettering Cancer Center Nomogram (MSKCCN) and MD Anderson Cancer Center Nomogram (MDACCN) were externally validated. A cohort-derived generalized linear model (GLM) was developed using prespecified clinicopathological predictors. Two ensemble approaches combining model predictions were also evaluated. Model performance was assessed using discrimination (area under the receiver operating characteristic curve, AUC), calibration, and Brier score. Threshold-based clinical performance was evaluated at predicted risk cut-offs of ≤ 5% and ≥ 15%.

resultsAmong 1080 patients, 187 (17.3%) had nodal macrometastases. The MDACCN achieved an AUC of 0.78 (95% CI 0.74-0.82). The cohort-derived GLM demonstrated improved discrimination with an AUC of 0.83 (95% CI 0.71-0.93) and a Brier score of 0.108. A logistic ensemble model achieved an AUC of 0.82 (95% CI 0.69-0.93) and the lowest Brier score (0.103). At a ≤ 5% predicted risk threshold, the GLM classified 24% of patients as low risk (observed macrometastasis rate 3.8%), while the ensemble classified 37% as low risk (observed rate 2.5%) with a false-negative rate of 5.6%.

conclusionsPrediction models using routinely available clinicopathological variables can identify a subgroup of patients with a very low risk of SLN macrometastasis. Ensemble modeling modestly improved classification performance and identified a larger low-risk group. Risk-adapted prediction tools may support selective axillary de-escalation in cN0 EBCs.