Abstract PS3-06-09: Her2-ladder: a spatially interpretable artificial intelligence-driven model for tailored dual-targeted therapy in HER2-positive breast cancer

Abstract Background Neoadjuvant dual HER2 blockade with trastuzumab and pertuzumab plus chemotherapy represents the current standard of care for HER2-positive breast cancer. However, marked heterogeneity in patient response has intensified discussions on treatment de-escalation for low-risk individuals and alternative therapies for resistant cases. Presently, clinically robust precision oncology tools facilitating such individualized therapeutic strategies remain unavailable. Methods We developed HER2-LADDER (Layered AI-based Dual-targeteD anti-HER2 Recommendation), an AI-driven decision-support framework that integrates clinicopathological information with spatial features derived from routine hematoxylin and eosin (H&E) and HER2 immunohistochemistry (IHC) whole-slide images. The predictive model was trained and tested using patients from the PCbHP arm of the phase II FASCINATE-N trial (NCT05582499) and subsequently validated externally in multiple real-world cohorts (total N = 381). Spatial in situ profiling techniques were employed to elucidate the biological mechanisms underlying the model’s predictive capacity. Results HER2-LADDER achieved outstanding predictive performance, demonstrated by an area under the curve (AUC) of 0.986 in the training cohort and 0.939 in the testing cohort, with consistent external validation results. Based on individualized HER2-LADDER scores, patients were stratified into Low-, Medium-, and High-score groups. These classifications effectively identified suitable candidates for treatment de-escalation (THP or PCbH/TCbH), standard dual-targeted therapy (PCbHP/TCbHP), or alternative regimens, including next-generation anti-HER2 antibody-drug conjugates (ADCs) and tyrosine kinase inhibitors (TKIs), respectively. Spatial analyses revealed key biological determinants underpinning HER2-LADDER predictions, specifically highlighting homogeneous HER2-enriched tumor cell patterns and neutrophil-helper T cell interactions. Conclusions HER2-LADDER offers clinicians a powerful, clinically actionable tool to personalize neoadjuvant HER2-targeted treatments. By accurately predicting individual responses and delineating optimal therapeutic pathways, it substantially advances precision medicine for patients with HER2-positive breast cancer. Citation Format: X. Wu, D. Ma, H. Lv, S. Zhao, W. Yang, Y. Jiang, G. Di, Z. Shao. Her2-ladder: a spatially interpretable artificial intelligence-driven model for tailored dual-targeted therapy in HER2-positive breast cancer [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 PS3-06-09.