PS4-05-18: The predictive value of tumor-to-gland density ratio in pathological complete response to Neoadjuvant chemotherapy in breast cancer

Abstract Objective: Identifying effective predictors of response to neoadjuvant chemotherapy (NAC) is essential for personalized breast cancer treatment. Our previous observations suggested that a higher tumor-to-gland density ratio (TGDR) on mammography may predict poorer therapeutic response. This study evaluated the predictive value of TGDR in NAC and developed models based on various pathological complete response (pCR) criteria. Methods: This retrospective study included 303 breast cancer patients who received NAC at the First Affiliated Hospital of China Medical University from January 2020 to July 2024. pCR was defined by four criteria: pCR1 (ypT0/is), pCR2 (ypT0/is ypN0), pCR3 (ypT0), and pCR4 (ypT0 ypN0). Patients were randomly divided into training and test sets (7:3). Logistic regression analyses identified independent predictors for each pCR criterion, and multiple logistic regression (LR) models were constructed. Model performance was assessed using AUC, F1 score, DeLong test, DCA, NRI, and IDI. Four machine learning algorithms (SVM, K-nearest neighbor, XGBoost, and LightGBM) were applied to compare predictive performance. Results: For pCR1, 140 patients (46.2%) achieved response. TGDR, ADC, ER, HER2, histological grade, and molecular subtype were independent predictors. Combined-variable models (AUCs: 0.840-0.852) outperformed single-variable models and demonstrated good clinical utility. Model 3 was selected for its simplicity and MRI independence; the LR algorithm performed best. For pCR2 (n=127), similar predictors were identified. Combined models (AUCs: 0.847-0.859) showed strong performance; SVM performed best. For pCR3 (n=118), model 2 (TGDR + pathological features) outperformed TGDR alone (AUC: 0.848 vs. 0.705); XGBoost performed best. For pCR4 (n=109), model 2 (TGDR, ER, HER2) outperformed model 1 (TGDR alone) in both training and test sets; the LR model performed best. Conclusion: TGDR is a novel imaging biomarker with predictive performance comparable to ADC and is an independent predictor of NAC response. Across all pCR criteria, TGDR-based models combined with clinicopathological features consistently outperformed TGDR alone. TGDR offers a practical alternative for NAC response prediction, particularly in settings with limited MRI access. Citation Format: W. Shijing, W. Hongrui, X. Mao. The predictive value of tumor-to-gland density ratio in pathological complete response to Neoadjuvant chemotherapy in breast cancer abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS4-05-18.