Abstract
Background: Breast lesions of uncertain malignant potential (B3 lesions) represent a heterogeneous group with variable risks of malignant upgrade, making the choice between surveillance and surgical excision challenging. This study aimed to develop and internally validate a practical multiparametric risk model for predicting malignant upgrade in patients with B3 breast lesions. Materials and Methods: In this single-center retrospective cohort study, 356 B3 lesions diagnosed by image-guided 16-gauge core needle biopsy between 2011 and 2025 were evaluated. Of these, 112 were managed with surveillance and 244 underwent surgical excision. Malignant upgrade, defined as ductal carcinoma in situ or invasive carcinoma in the biopsied (index) lesion on the excision specimen, was assessed among excised lesions; carcinomas in a separate focus were analysed separately. Multivariable logistic regression identified independent predictors of upgrade, and a simplified risk score was derived from the final model. Model performance was assessed using ROC analysis, bootstrap resampling, and 10-fold cross-validation, with variable selection repeated within each resample and fold. Results: Index-lesion-related malignant upgrade occurred in 46 of 244 excised lesions (18.9%; 13 DCIS and 33 invasive carcinomas); five further carcinomas in a focus separate from the biopsied lesion (2.0%) were analysed separately. Increasing age, high BI-RADS category (4B–5), microcalcifications, atypia on biopsy, and hormone replacement therapy use were independently associated with upgrade; lesion size > 10 mm was not. The full model showed high discrimination (AUC, 0.948; optimism-corrected AUC, 0.930; 10-fold cross-validation AUC, 0.918). The derived 0–16-point score achieved an AUC of 0.945 (optimism-corrected, 0.926) and stratified lesions into low- (0–7 points; 2.8%), intermediate- (8–10 points; 50.0%), and high-risk (11–16 points; 86.2%) groups. The bootstrap-corrected Brier score was 0.082, with a bootstrap-corrected calibration slope of 0.79. Conclusions: A simple risk score integrating routinely available clinical, radiological, and histopathological variables demonstrated high discriminatory performance for malignant upgrade in B3 breast lesions in this selected retrospective cohort. Safe omission of surgery and a reduction in unnecessary procedures have not yet been demonstrated; external and prospective validation is required before this tool can support individualized decision-making.