Abstract A016: Microenvironmental drivers of DCIS recurrence and progression to breast cancer

Background Studies show that only 20–30% of untreated ductal carcinoma in situ (DCIS) cases will evolve into invasive breast cancer. Thus, many patients are subject to overtreatment and its associated morbidities. The ecological context of the tumor microenvironment may impact how DCIS evolves. We applied landscape ecology statistics to hematoxylin and eosin (H&E) images from DCIS to test whether they predict recurrence of DCIS or progression to invasive disease. The ability to make such predictions is key for optimizing treatment decisions, reducing associated risks, and improving the efficiency of healthcare resource allocation. Methods We conducted a retrospective cohort study comprising 158 patients with a DCIS diagnosis, which included 57 cases of DCIS recurrence (2–228 months) and 45 cases of progression to invasive disease (12–180 months). Of these, 21 patients received lumpectomy, 92 had lumpectomy followed by radiation therapy, and 45 had total mastectomy. We generated a digital pathology library of whole-slide images (WSIs) from H&E tumor sections and performed machine learning guided nuclear segmentation and classification. A comprehensive set of spatial statistics was calculated for each WSI to characterize the spatial architecture and cellular interactions within the microenvironment. Cross-validation using a linear spline-based hazard regression model (HARE) was applied to a training subset of the data. The resulting survival model was validated on the test subset to predict disease-free survival (DFS) of recurrence and progression. Results We found that the type of DCIS treatment significantly stratified patients based on DFS (log-rank test p<0.0001). Further analysis of patients who received lumpectomy + radiation or mastectomy revealed that decreased mean nearest neighbor distance between tumor cells and lymphocytes, and between fibroblasts and lymphocytes, was significantly associated with improved recurrence-free survival (concordance=0.68). Stratification of samples based on the median risk score derived from our model was discriminatory with a p-value <0.015. Additionally, a decreased nearest neighbor distance between lymphocytes and fibroblasts showed a marginal association with an increased time to invasive disease (p<0.06). Conclusions Integrating spatial ecological statistics derived from digital pathology provides enhanced risk stratification for DCIS, beyond the established effects of initial treatment. Our findings underscore the importance of the tumor microenvironment's ecological dynamics in DCIS recurrence and progression. The predictive risk associated with specific spatial interactions, particularly involving lymphocytes and fibroblasts, suggests that previously unappreciated ecological mechanisms may play a role in the development of early micro-metastases, influencing outcomes even after standard therapies such as mastectomy or lumpectomy with radiation. These insights highlight novel targets for improved prognostic tools and more personalized therapeutic strategies. Citation Format: Luis H. Cisneros. Microenvironmental drivers of DCIS recurrence and progression to breast cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Cancer Evolution: The Dynamics of Progression and Persistence; 2025 Dec 4-6; Albuquerque, NM. Philadelphia (PA): AACR; Cancer Res 2025;85(23_Suppl):Abstract nr A016.