Evaluation of an AI-based digital pathology tool for breast cancer recurrence risk

Hormone receptor-positive (HR+) breast cancer is the most common type among women in the United States. Multigene assays such as Oncotype DX (ODX) assess recurrence risk and chemotherapy benefit but are costly and time-consuming. This exploratory study evaluates the AI-powered PreciseBreast (PDxBR) and compares its risk scoring with established tools. This pilot study was conducted in a community health system. Female patients with newly diagnosed early-stage HR+ invasive ductal or mixed ductal/lobular carcinoma eligible for ODX testing were included. Pathology slides were analyzed using PDxBR. Results were compared with ODX and RSClin scores using Cohen’s kappa, Gwet’s AC2, and symmetry testing. Turnaround times were compared using a paired t-test. Results did not influence clinical management. Fifty patients were enrolled (mean age 61); 72% had stage IA disease. Most had invasive ductal carcinoma (94%), 6% had mixed histology, and 18% were node-positive. Agreement between PDxBR and ODX was fair (κ = 0.25, p = 0.29), with most patients classified as low risk. Agreement with RSClin was slight (κ = 0.06–0.084, p < 0.05), with no agreement beyond chance by Gwet’s AC2. Mean turnaround time was shorter for PDxBR (1.9 vs. 6.9 days, p < 0.0001). Cost was lower ($1500 vs. $4620). PDxBR confirmed the expected workflow advantages of AI-based pathology, including shorter turnaround time and lower estimated cost, but had limited concordance with ODX and RSClin. AI-based digital pathology may complement genomic assays, but larger studies with clinical outcome data are required before its role as an alternative can be established.