An interpretable weighted ensemble based on routinely collected clinical data for the accurate prediction of axillary lymph node metastasis.

Background: Axillary lymph node (ALN) status is a primary prognostic indicator in breast cancer, yet conventional surgical staging for determining ALN status is invasive. We aimed to develop an interpretable, noninvasive weighted ensemble model for ALN metastasis prediction using only routine, universally accessible clinicopathological data. Methods: We analyzed a retrospective cohort of 915 patients (training set: n=732; test set: n=183). Twelve routine clinicopathological variables, including age, tumor diameter, histological grade, and biomarkers [estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki-67], served as predictors. A two-stage weighted ensemble was developed through the integration of logistic regression (LR) and extreme gradient boosting (XGBoost) via Python version 3.9. Model performance was evaluated with the held-out test set according to the area under the receiver operating characteristic curve (AUC), area under the precision-recall curve (AUPRC), and sensitivity. Model interpretability was achieved through Shapley additive explanations (SHAP). Results: The weighted ensemble model achieved a superior AUC of 0.762 on the test set, outperforming optimized XGBoost (AUC =0.752) and tuned LR (AUC =0.741). The model demonstrated a robust AUPRC of 0.575 and achieved a high sensitivity of 0.800. SHAP analysis revealed that model predictions were primarily driven by tumor diameter, invasive ductal carcinoma pathology type, and plateau time-intensity curve patterns. Conclusions: The interpretable weighted ensemble model, based only on standard tabular clinicopathological data, provides accurate and transparent ALN risk stratification. Its high sensitivity supports its use as a valuable triage tool for identifying low-risk patients who may safely forego invasive axillary surgery.