Development and validation of a machine learning-based model for predicting sentinel lymph node metastasis in patients with clinically node-negative breast cancer following neoadjuvant chemotherapy.

[BACKGROUND] Step-down therapy is becoming a critical approach in the treatment of breast cancer in a localized setting. The applicability of sentinel lymph node biopsy (SLNB) in patients with clinically node-negative (cN0) breast cancer following neoadjuvant chemotherapy (NACT) is still a topic of controversy. The objective of this study was to compare a variety of machine learning algorithms to determine the most effective model for predicting sentinel lymph node (SLN) metastasis in cN0 breast cancer following NACT.

[METHODS] A total of 221 patients with cN0 breast cancer who underwent standardized NACT combined with SLNB at the Affiliated Hospital of Southwest Medical University from January 2017 to January 2025 were included in this retrospective study. Predictive models for the risk of SLN metastasis were created using four machine learning algorithms. The clinical net benefit was compared through decision curve analysis (DCA), and the diagnostic performance was evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. SHapley Additive exPlanations (SHAP) was also implemented to evaluate the model that demonstrated the most optimal performance.

[RESULTS] The SLN-negative group contained 186 cases (84.2%), while the SLN-positive group contained 35 cases (15.8%). The logistic regression (LR) model demonstrated superior performance in comparison to the other models in the testing set. The sensitivity, specificity, accuracy, F1 values, and AUC were, respectively, 0.448, 0.947, 0.864, 0.508, and 0.889 [95% confidence interval (CI): 0.886-0.892]. It yielded the greatest net benefit across the majority of threshold ranges in the testing set. The main predictors identified by SHAP analysis were radiological complete response (rCR), lymphovascular invasion, and axillary nodes on ultrasonography.

[CONCLUSIONS] The LR model developed in this study demonstrates high specificity and can reliably identify patients without SLN metastasis, thereby supporting the exemption of SLNB in low-risk patients with cN0 breast cancer following NACT.