Abstract PS3-06-03: Refining adjuvant CDK4/6 inhibitor eligibility in early HR+ HER2- breast cancer with artificial intelligence

Background: CDK4/6 inhibitors have demonstrated efficacy in reducing disease recurrence among HR+ HER2- breast cancer patients meeting high risk of recurrence (RoR) criteria. These criteria, defined in trials such as NATALEE, consist of factors including Ki67>20%, G3, or Oncotype DX>26. However, these criteria are suboptimal predictors of recurrence risk and leave room for improvement with respect to only providing CDK4/6i to truly high RoR patients. We evaluated Ataraxis Breast (ATX), an artificial intelligence tool which integrates morphological features extracted from H&E-stained slides with clinical information, to (1) establish the RoR by ATX broadly, and (2) assess ATX-based RoR performance within risk groups defined by NATALEE criteria. Methods: We generated ATX scores in data from 2,228 HR+ HER2- patients from five institutions with available H&E slides and complete clinical information. Results were pooled for analysis. Patients received standard of care treatment which did not include CDK4/6i as they were not available at the time of collection. Clinical High-Risk and Low-Risk groups were defined according to NATALEE trial CDK4/6i eligibility criteria. T2N0 patients without available Ki-67 or Oncotype scores were excluded. The ATX model generated 5-year RoR scores, with a primary cut-off at 10% to classify patients into ATX-High (greater or equal than 10%) and ATX-Low (less than 10%) risk groups. The primary endpoint was disease-free interval, evaluated using hazard ratios (HR per 0.1 increase in ATX) and C-index. ATX was not trained with any data used in this analysis. Results: Out of 2,228 HR+ HER2- patients, 918 (41%) were Clinical High-Risk, i.e., eligible for adjuvant CDK4/6i according to NATALEE criteria, and 1,310 were Clinical Low-Risk, i.e., not eligible for adjuvant CDK4/6i by NATALEE. In the Clinical High-Risk group, ATX was prognostic, with a HR of 1.81 (1.58-2.06) and C-index of 0.71 (0.68-0.74). In the Clinical Low-Risk group, ATX was also prognostic with a HR of 2.53 (1.57-4.06) and C-index of 0.72 (0.69-0.75). Among all Clinical Low-Risk patients, the ATX-High subgroup (N=90) had a 5-year recurrence rate of 16%, comparable to 17% recurrence rate in the Clinical High-Risk cohort, suggesting that ATX might identify high-risk CDK4/6i-eligible patients not captured by the current clinical criteria. Conversely, among Clinical High-Risk patients with ATX-Low scores (N=376), the 5-year recurrence rate was 10%, similar to the 6% rate seen across Clinical Low-Risk patients with primary T1N0 disease, suggesting that ATX might identify Clinical High-Risk patients who might not be appropriate candidates for CDK4/6i based on clinical factors alone. Conclusions: ATX score enhances risk stratification in early-stage HR+/HER2− breast cancer patients and could potentially redefine which patients are eligible for adjuvant CDK4/6i relative to the current criteria. Citation Format: N. P. McAndrew, E. Chiru, E. Senkus-Konefka, A. Bardia, J. Cappadona, K. G. Zeng, K. J. Geras, J. Witowski. Refining adjuvant CDK4/6 inhibitor eligibility in early HR+ HER2- breast cancer with artificial intelligence [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 PS3-06-03.