Understanding oncologists' use of artificial intelligence in the clinical setting: A field medical study.
Abstract
685 Background: Oncologists increasingly use clinical artificial intelligence (AI) tools in their decision-making. Understanding that use is central to ensuring AI only maintains or improves patient care quality. Over 1,450 AI-enabled medical devices are FDA-authorized, and most US health systems report AI deployment, yet AI sits between clinicians and the evidence base with little structured insight into how it shapes decisions or how far it's trusted. We piloted a field medical method to understand oncologists' AI use in decision making as recurring, longitudinal insight, not a one-time survey. Methods: US Oncology Field Medical Colleagues ran a standardized three-item single-select questionnaire with oncology healthcare professionals, paired with open-text verbatim capture on AI use in clinical decision making. The structured framework was used during routine scientific exchange. Items assessed individual AI use, organizational EHR-integrated AI, and trust in AI output. Affirmative answers prompted free-text detail on tools, uses, and trust drivers. Verbatims were coded to one primary use case. Initial capture for this interim analysis ran 26 May–05 June 2026. The full effort with updated data continues through 2026. Counts are directional, not statistical. Results: Eleven Field Medical Colleagues captured 54 insights, with 19–20 responses per item and 26 coded verbatims (Table). Individual AI use in clinical decision support was 84%. Organizational EHR-integrated AI was 53% and skewed towards academic versus community cancer centers. High trust in AI output was 60%, with 40% cautious or skeptical. Treatment decision support led use cases ahead of trial matching; one AI reference tool was named in ~38% of signals, more than all others. Two modes emerged: a point-of-care reference mode oncologists used to confirm decisions, and an enterprise trial-matching mode, promising but limited by errors and data gaps. Conclusions: Oncologists are already using AI to inform treatment decisions with the majority indicating high trust in the AI-generated output. For patient care, this raises the importance of grounding AI-informed decisions in accurate, validated evidence. Healthcare organizations and stakeholders can help as a scientific partner, reinforcing evidence accuracy and collaborating on tool reliability for better patient care. Capture is ongoing and longitudinal; directional findings should be paired with further structured research. Domain Result By setting Individual AI use (Q1, n=19) 84% yes (16/19) All community-based answered Yes; Nos from academic EHR-integrated AI (Q2, n=15) 53% yes (8/15) 7/8 yes from academic Trust in AI output (Q3, n=20) 60% high, 35% medium, 5% low Split evenly Coded verbatims (n=26) Decision support, trial matching lead Adoption skews academic Trust stance (n=26) 5 skeptic, 10 exploring, 11 all-in Middle is movable