Development and Validation of a Predictive Model for Sentinel Lymph Node Biopsy Exemption in Ductal Carcinoma in situ Patients.

Abstract

Introduction: There is no uniform standard on whether total mastectomy for ductal carcinoma in situ (DCIS) can exempt sentinel lymph node biopsy (SLNB). This study attempts to find the risk factors for the underestimation of DCIS pathology and establish the corresponding prediction model to screen suitable DCIS patients for exemption from SLNB.

Methods: A total of 826 patients with DCIS met the inclusion criteria. Logistic regression identified lesion size, Ki67, estrogen receptor (ER) status, human epidermal growth factor receptor 2 (HER2) status, histological grade, and diagnostic method as independent predictors of pathological underestimation (p < 0.05). Based on these variables, a predictive model was developed: p = 0.354 × lesion size + 0.017 × Ki67 + 1.186 × ER - 2.501 × diagnosis method (1) - 1.575 × diagnosis method (2) - 0.050 × HER2 (1) - 1.578 × HER2 (2) + 1.160 × grade (1) + 1.497 × grade (2) - 2.418 (if age <50) - 0.156 × 1 (if age >50). The model showed good performance with a sensitivity of 79.2%, specificity of 73.8%, and overall accuracy of 76.2%. The area under the ROC curve (AUC) was 0.856 (95% confidence interval: 0.831-0.881, p < 0.001). Subgroup analyses indicated that age, presence of mass, ER, HER2, tumor grade, and histological grade significantly affected model performance (AUC = 0.787; sensitivity = 0.695; specificity = 0.753). Stratified analysis showed higher sensitivity in patients <50 years (0.840 vs. 0.656) and higher AUC in ER-positive cases (0.865). In HER2-based analysis, only the presence of a mass remained significant. Mass-based analysis revealed all variables except age were significant, with a higher AUC in patients without a mass (0.784 vs. 0.727).

Conclusion: This study developed a predictive model based on lesion size, Ki67, ER status, HER2 status, histological grade, and diagnostic method to assess the risk of pathological underestimation in DCIS. The model demonstrated good predictive performance (AUC = 0.856) with high sensitivity and specificity, indicating its potential clinical utility. Subgroup analyses revealed that factors such as age, presence of a mass, and ER status influenced model performance, with particularly better accuracy observed in patients under 50 and those with ER-positive tumors. This model may serve as a useful tool to support clinical decision-making, especially in preoperative evaluation of invasive potential in DCIS patients.