Abstract PS2-02-14: Improving Identification and Enrollment in HR+/HER2− Breast Cancer Trials Using AI Clinical Trial Patient Matching Tool

Abstract Background: Significant enrollment gaps persist in breast cancer clinical trials, particularly among patients with high-risk hormone receptor-positive (HR+), HER2-negative early breast cancer (EBC). Traditional manual methods are challenged by high patient volume, unstructured clinical notes, and limited research staffing. These barriers hinder recruitment efficiency and contribute to limited inclusion of all patient populations. To address these gaps, we deployed a Clinical Trial Patient Matching (CTPM) system powered by artificial intelligence (AI) and natural language processing (NLP) to enhance identification of eligible patients for two multi-site interventional trials: EMBER-4 (NCT05514054) and DARE (NCT04567420). Methods: The CTPM algorithm was deployed across 13 breast oncology sites within the Yale New Haven Health (YNHH) Smilow Cancer Hospital network. From September 2022 to May 2024, the system performed automated weekly pre-screening of all breast cancer patients seen across the network. CTPM extracted both structured and unstructured data from the electronic health record (EHR), applying trial-specific eligibility criteria for two multi-site interventional trials: EMBER-4 and DARE. Identified patient matches underwent manual chart review by a clinical research coordinator (CRC) to confirm eligibility prior to outreach and enrollment. Results: Between September 2022 and May 2024, the CTPM system pre-screened 45,380 unique breast cancer patients across YNHH. Based on trial-specific criteria, the system identified 429 patients as potentially eligible for EMBER-4 and 353 for DARE. Following manual chart review by CRCs, 263 patients were confirmed as eligible for EMBER-4 and 140 for DARE. For the EMBER-4 trial, CTPM identified 263 eligible patients, of whom 2.28% were aged 18-39, 26.6% were 70 years, 26.6% were racial/ethnic minorities, 6.1% NES, 22.8% Medicaid-insured, and 6.1% rural residents. Thirty-seven patients (13.81%) enrolled, with higher representation among racial/ethnic minorities (27.0%), patients 70 years (13.5%), and Medicaid-insured patients (16.2%). For DARE, the 140 eligible patients identified by CTPM included 5.7% were aged 18-39, 22.1% were 70 years, 25.7% were racial/ethnic minorities, 8.6% NES, 24.3% Medicaid-insured, and 7.1% rural residents. Thirty-one patients (22.1%) enrolled in DARE, with notable representation among patients 70 years (22.6%), racial/ethnic minorities (19.4%), and Medicaid-insured (19.35%) patients. As a result of this enhanced identification and screening process, YNHH was a top enrollment site for patients nationally in both trials. Conclusions: This study presents the first prospective application of an AI-driven trial-matching system for high-risk HR+/HER2- EBC patients. Integration with EHRs enhanced identification of eligible patients, particularly among historically underrepresented groups. The subset of patients deemed ineligible after manual review reflects the difference between the CTPM system’s design— prioritizing sensitivity using a limited set of high-priority eligibility criteria from structured EHR data— and CRCs’ comprehensive reviews including unstructured clinical data not readily extractable through AI/NLP methods. While relaxed algorithm criteria minimized false negatives, we found that a two-tiered approach of broad AI-driven prescreening followed by expert validation supported efficient, scalable trial matching while preserving protocol fidelity. Yet, improved identification did not translate to proportional increases in consent or enrollment rates. To maximize clinical trial participation, AI tools like CTPM must be combined with expert review and targeted interventions that address downstream barriers to patient consent and enrollment. Citation Format: J. Liu, G. Gong, S. Pandya, J. Xie, J. Parikh, N. Fischbach, P. Kunz, L. Pusztai, M. Lustberg. Improving Identification and Enrollment in HR+/HER2− Breast Cancer Trials Using AI Clinical Trial Patient Matching Tool 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 PS2-02-14.