A Nomogram-Based Malignancy Risk Assessment Model for Breast Masses Upgraded from BI-RADS 1-3 in Mammography to Category 4 in Supplemental Ultrasound for Women with Dense Breasts

Background: Mammography, the primary breast cancer screening method, exhibits high false-negative rates in women with dense breasts. Supplemental ultrasound increases cancer detection sensitivity but also raises the number of unnecessary biopsies due to its low positive predictive value (PPV). Objective This study aims to develop a predictive model for assessing the malignancy risk of breast masses initially categorized as BI-RADS 1–3 in mammography but upgraded to BI-RADS 4 in supplemental ultrasound in women with dense breasts. The aim was to enhance the PPV of supplemental ultrasound, thereby reducing unnecessary biopsies. Methods A retrospective analysis was conducted to identify breast masses that met specific inclusion and exclusion criteria. These masses were then randomly divided into training and validation sets. Pathological, radiological, and clinical data of the breast masses were systematically collected. Using the LASSO algorithm, key variables were identified in the training set, which facilitated the development of a logistic regression model, along with a corresponding nomogram. The model’s efficacy was assessed in both the training and validation sets, using metrics such as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), ROC curves, calibration curves, and decision curve analysis. Results Of the 11,051 women screened, 425 masses met the study criteria. The LASSO algorithm identified ultrasonic characteristics, such as mass size, shape, margin, calcifications, and vascularity as significant predictors. The model demonstrated high discriminative power, with AUCs of 0.908 and 0.882 for the training and validation sets, respectively, and robust calibration. The PPV in the training set increased significantly from 0.178 to 0.556 and in the validation set from 0.213 to 0.561. The model also showed high NPVs, sensitivity, and specificity. Conclusion The study successfully developed a predictive model and corresponding nomogram for evaluating malignancy risk in a specific subset of breast masses. Validated for accuracy and clinical utility, the model significantly improves the PPV of supplemental ultrasound, indicating a potential reduction in unnecessary biopsies. It represents a promising advance in personalized breast cancer screening for women with dense breasts.