Evaluating the Impact of AI on Breast Cancer Screening Quality and Effectiveness: A Prospective RCT

Mammography screening is limited by imperfect human interpretation. Computer-aided detection (CAD) tools were introduced decades ago to improve health services, healthcare quality, and healthcare effectiveness in 2D mammography but were FDA-cleared without prospective randomized controlled trials (RCTs). These early CAD tools ultimately lowered accuracy, led to adverse patient outcomes, and increased costs. Over the past decade, 3D mammography, which contains >50-fold more image data within each exam, has become the dominant screening modality. Concurrently, newer artificial intelligence (AI)-based CAD tools, FDA-cleared based on small reader studies in artificial settings, are rapidly expanding in clinical use on these 3D exams. Given historical CAD failures, broad AI adoption should require evidence of effectiveness in improving patient outcomes. This study will conduct the first real-world RCT of an FDA-cleared AI tool for 3D mammography in U.S. clinical practice. We will assess the comparative effectiveness of mammography interpretation with vs. without AI, generating evidence on healthcare quality, health services effectiveness, and patient outcomes. Data collection across 16 distinct breast imaging facilities in two geographic regions will capture effectiveness at the patient-, exam-, and radiologist-level, linking to state cancer registries for gold-standard outcomes. Aim 1. Conduct a two-year pragmatic RCT of > 154,000 3D screening exams randomized to be interpreted either with vs. without AI. We will assess short-term performance measures and outcomes of the screening and diagnostic evaluation cascade (primary study outcomes are cancer detection rate and recall rate; secondary outcomes include biopsy rate, biopsy yield, short-interval follow-up recommendations). Aim 2. Assess 1-year breast cancer outcome performance measures (e.g., sensitivity, specificity) and clinical outcomes (e.g., interval cancer rate, tumor molecular subtypes) for 3D screening mammograms randomized to be interpreted with vs. without AI. Aim 3. Investigate the human-AI interface in screening mammography decision making by examining the heterogeneity in radiologists’ performance with vs. without AI based on key characteristics (e.g., clinical experience in mammography, interpretive volume, tolerance of ambiguity, trust in AI, experience with AI). Additionally, we will examine automation bias, assess the learning curve associated with adopting AI, and evaluate potential carry-over effect of AI use. This innovative RCT will provide the highest level of evidence on AI effectiveness in 3D mammography, filling critical gaps on its impact on decision-making and healthcare delivery in community-based settings. Project Number: 1R01CA300103-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Cancer Institute (NCI) | Principal Investigator: JOANN ELMORE (+2 co-PIs) | Institution: UNIVERSITY OF CALIFORNIA LOS ANGELES, LOS ANGELES, CA | Award Amount: $761,691 | Activity Code: R01 | Study Section: Special Emphasis Panel[ZRG1 HSS-D (90)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11296142