Abstract Background: Adjuvant trials have established the benefit of taxane and dose dense chemotherapy (CT) in high risk breast cancer. Digital pathology can provide biologic insights into CT sensitivity and better personalize treatment using data from completed trials. Methods: Digital H 0.5 vs true signature) were applied to predict distant recurrence in CALGB 9344. The most accurate signature as measured by concordance index (C-index) was subsequently tested in other cohorts. The primary outcome was distant recurrence free interval (DRFI); breast cancer specific survival (BCSS) was reported in ACS as DRFI was not recorded. Cox models controlling for age, nodal status, tumor size, and estrogen / progesterone receptor status were used to estimate interaction of signature and CT benefit. Cox models incorporating these same clinical factors and signature predictions were fit in CALGB 9344 and subsequently applied to all cohorts to better identify groups with low / high recurrence risk. Results: This study includes 2,355 cases from CALGB 9344, 1,460 cases from CALGB 9741, 1,122 cases from ACS, and 2,292 cases from Chicago hospitals. CALGB 9344 and 9741 included more hormone receptor negative (∼32%) and higher nodal burden (mean 5 involved) cases than the Chicago / ACS cohorts. A histology model for a "bronchioid" gene signature (originally designed to identify a favorable lung adenocarcinoma subtype but also prognostic in breast cancer), best predicted DRFI in CALGB 9344 (C-index 0.60) and performed well in CALGB 9741 (C-index 0.63), Chicago (C-index 0.69), and ACS cohorts (BCSS C-index 0.63). This histology signature outperformed OncotypeDX when available in the Chicago cohort (n = 983, DRFI C-index 0.69 vs 0.65). An explainability analysis demonstrated that tubule formation and low nuclear pleomorphism were associated with bronchioid signature prediction. Patients in the lowest risk tertile for the bronchioid signature did not benefit from the addition of taxane to adjuvant AC in CALGB 9344 (taxane HR in low risk tertile 1.13, 95% CI 0.81 - 1.59, HR in higher tertiles 0.74, 95% CI 0.62 - 0.88, interaction p = 0.04). Significant interaction also seen for distant recurrence free survival (p = 0.002) and overall survival (p = 0.01) in CALGB 9344. Similarly, higher signature tertiles predicted benefit from adjuvant CT (vs no CT) in the Chicago cohort (interaction p = 0.01 for DRFI). Dose dense CT (vs standard schedule) only improved DRFI in the higher signature risk tertiles (HR 0.73, 95% CI 0.53 - 1.00) in CALGB 9741, but interaction testing was not significant (p = 0.46). Cox models incorporating both clinical factors and bronchioid signature improved recurrence prediction (C-index = 0.65, 0.73, and 0.78 for DRFI in CALGB 9344, CALGB 9741, and Chicago, and 0.71 for BCSS in ACS). Low / high risk tertiles from this Cox model (with cutoffs defined from CALGB 9344) consistently identify patients with low (12.8%, 6.2%, 2.9%, 2.4% in CALGB 9344, 9741, Chicago, and ACS) and high (43.4%, 33.7%, 23.9%, 6.7%) 5-year distant recurrence rates (breast cancer survival events for ACS). Conclusion: Digital histology signatures predict adjuvant taxane / CT benefit and recurrence risk in trial and real-world cohorts and could be applied to personalize treatment. Digital pathology can provide biologic insights in datasets with limited tissue availability for genomic testing. Citation Format: F. M. Howard, S. Kochanny, A. Li, S. Hassan, J. Dolezal, E. M. Flores, R. Medenwald, C. Fan, M. Watson, L. McCart, L. R. Teras, C. Bodelon, A. V. Patel, W. F. Symmans, D. Stover, C. Perou, M. Sullivan, K. Yao, A. T. Pearson, D. Huo. Digital Histology Models Predict Chemotherapy Benefit in Randomized Adjuvant Trials and Real-World Cohorts 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 PS2-07-29.
Howard et al. (Tue,) studied this question.