Abstract PS3-06-02: Prospective Evaluation of “PreciseBreast” AI Tool in Early-Stage Invasive Breast Cancer Risk Stratification

Abstract Background: Hormone receptor-positive breast cancer is the most common type of breast cancer among women in the United States. Multigene assays such as Oncotype DX (ODX) assess recurrence risk and the potential benefit of chemotherapy, guiding treatment decisions. These assays have limitations, including high cost and lengthy processing time. This study evaluates the novel AI-powered PreciseBreast (PDxBR) risk score, developed by PreciseDx, which uses digital pathology and machine learning to provide a rapid, phenotype-based assessment of recurrence risk. The primary objective is to evaluate PDxBR as a reliable tool for guiding treatment based on recurrence risk. Methods: This prospective correlational study was conducted within a community health system. Eligible participants included female patients with newly diagnosed early-stage, hormone receptor-positive invasive ductal carcinoma who qualified for ODX testing. Pathology slides were analyzed using the PDxBR assay, which incorporates morphologic features such as nuclear pleomorphism, mitotic activity, tubule formation, and tumor-infiltrating lymphocytes. ODX defines low risk of distance recurrence as a score of 0-25, and high risk as ≥26. The RSClin tool's low risk is defined as ≤ 3%, intermediate risk as 3-5%, and high risk as ≥5% in validation studies. PDxBR and ODX scores were compared using McNemar’s test, and PDxBR and RSClin scores were compared using a symmetry test. Turnaround time was compared using a paired t-test. The results of this study were for research purposes and did not affect patient management. Results: A total of 32 patients were enrolled. The mean age was 63 years, ranging from 40 to 82 years. Twenty-six patients (81. 3%) had stage IA disease, 4 patients (12. 5%) had stage IIA, and 2 patients (6. 3%) had stage IIB disease. Three patients (9. 4%) had carcinoma with mixed invasive ductal and lobular features, while the remaining 29 patients (90. 6%) had pure invasive ductal carcinoma. Among the seven premenopausal patients, ODX scores ranged from 6 to 16, with one patient scoring 18. Among patients classified as ODX low risk (N=6), 100% were deemed high risk by PDxBR. Only one patient had a high-risk ODX score, and that patient was classified as low risk by PDxBR (p = 0. 059). There was a significant association between RSClin and PDxBR classifications: 66. 7% of PDxBR high risk patients were also high risk by RSClin, while most RSClin intermediate and low risk patients were classified as low risk by PDxBR (p 0. 0001). The mean turnaround time for ODX was 6. 3 days (SD 3. 4) versus 2. 0 days (SD 1. 5) for PDxBR (p 0. 0001). The list price for PDxBR was 1, 500 compared to 4, 620 for ODX. Conclusion: The PDxBR assay offers significantly faster turnaround time and lower cost compared to ODX. The PDxBR risk score demonstrates strong concordance with the RSClin tool, and while there is discordance with ODX, the small sample size limits definitive conclusions. A larger study is needed to further evaluate concordance, non-inferiority, and clinical impact. Additional investigation is warranted to explore the basis of discordant results. Ultimately, a combined approach incorporating gene expression profiling, clinical features, and tumor morphology may provide the most accurate recurrence risk assessment. AI-powered digital pathology tools like PDxBR may complement or, in certain settings, serve as cost-effective alternatives to multigene assays, especially in resource-limited settings. This study is ongoing, and continued enrollment will allow for a more robust evaluation. Citation Format: T. Telvizian, A. P. Maity, M. Weiss, O. Agyei, R. Ciocca, C. Carruthers, J. Sabol, S. Kjelstrom, A. Ghaneie. Prospective Evaluation of “PreciseBreast” AI Tool in Early-Stage Invasive Breast Cancer Risk Stratification 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 PS3-06-02.