Abstract
BACKGROUND: Invasive micropapillary carcinoma (IMPC) is a rare subtype of breast cancer characterized by a high risk of lymph node metastasis (LNM). The study aimed to identify predictors of LNM and to develop a machine learning (ML)-based risk prediction model for patients with breast IMPC. METHODS: We retrospectively analyzed a cohort of 229 patients diagnosed with breast IMPC between 2019 and 2021. Patients were randomly assigned to training and test sets in a 7:3 ratio. Independent risk factors for LNM were identified using univariable and multivariable logistic regression analyses. Thirteen ML algorithms were trained and compared to determine the optimal model. Model performance was evaluated using the area under the curve (AUC), calibration plots, and decision curve analysis. Internal validation was performed using 100 iterations of tenfold cross-validation. RESULTS: LNM was present in 158 patients (69%). Tumor size, histological grade, progesterone receptor staining intensity, and lymphovascular invasion were identified as independent predictors of LNM (all p < 0.05). Among the 13 ML models, logistic regression (LR) demonstrated the best performance, achieving an AUC of 0.88 in the test set. A nomogram based on the LR model was constructed to facilitate clinical application, showing excellent calibration, clinical utility, and a classification accuracy of 76% (95% confidence interval: 70%-82%). The median AUC across cross-validation iterations was 0.83 (interquartile range: 0.76-0.91). CONCLUSIONS: This study identified key predictors of LNM in breast IMPC and developed a well-calibrated nomogram to support individualized treatment decision-making.