The impact of AI on clinical decision making in healthcare teams: A qualitative study of a randomized controlled study.

Abstract

749 Background: Despite growing enthusiasm for healthcare artificial intelligence (AI), no studies have examined how AI reshapes clinical team dynamics and decision-making within a randomized controlled trial (RCT). This leaves a critical gap in understanding how human–AI interaction redistributes authority and influences team-based cancer care. Methods: This qualitative study was conducted in two waves (2024 and 2026) in an academic radiation oncology department following the SHIELD-RT RCT (NCT04277650), in which an electronic health record–based machine learning algorithm identified patients at high risk for acute care during radiotherapy and significantly reduced acute care utilization. We conducted 30 semi-structured interviews with 29 clinicians (10 radiation oncologists, 4 advanced practice providers, 5 nurses, and 10 radiation therapists), analyzed using abductive analysis combining grounded theory and thematic analysis. Results: SHIELD-RT filled "authority gaps"—mismatches between staff expertise and their structural ability to act on it. For nurses, it enhanced predictive capability, enabling proactive care within their existing clinical authority; for radiation therapists, it validated observational knowledge and lent legitimacy to escalate patient concerns. These mechanisms expanded frontline clinician contributions to decision-making while preserving physician oversight. Notably, these improvements were sustained after trial completion. Effects were contingent on the algorithm's scientific credibility, universal accessibility across roles, seamless workflow integration, and organizational leadership endorsement. Conclusions: Clinical AI designed with attention to professional roles and decision-making structures can expand frontline clinician contributions to cancer care without disrupting physician oversight. Evaluation frameworks should extend beyond quantitative endpoints to capture how AI reshapes clinical relationships, redistributes professional authority, and influences team-based care delivery. How SHIELD-RT completed and activated frontline staff authority. Dimension Nurses Radiation Therapists Theoretical point Component held Affordances (formal standing to act and direct care) Expertise (tacit predictive knowledge from daily contact) Groups hold different single components, not uniformly low authority Component lacking Predictive expertise (practice was reactive) Affordances (no standing to compel action) Authority is partial in systematically different ways Mechanism Expertise augmentation Affordance extension One technology completes authority via opposite mechanisms Result Acts predictively through existing standing Observations gain legitimacy and compel action Expanded participation; physician jurisdiction intact (non-zero-sum)