EP 22 Dr. Moira E. Smith, MD, MPH, University of Virginia
Supporting Clinical Judgment Without Adding to the Noise
Guest: Moira E. Smith, MD, MPH | Released 05/28/26
What would useful AI support look like in an emergency department already full of interruptions, incomplete information, and competing demands? Dr. Moira Smith joins DJ and Ethan to explore that question through her work in emergency medicine and clinical informatics at the University of Virginia. Drawing on a review she coauthored about AI and diagnostic error, she considers where technology might reduce cognitive burden and where another alert or an overconfident answer could make the work harder.
Dr. Smith starts with the conditions under which clinicians make decisions. A crowded department, repeated interruptions, and a complicated record can make it difficult to assemble a patient's story. She imagines tools that help reconstruct that story, surface conflicting information, and identify something that deserves another look. Those possibilities are prospective: she distinguishes the information-retrieval capabilities she would like to have from tools available in her own department at the time of recording. The goal is to support a clinician's assessment while keeping uncertainty visible.
Triage offers a practical example. A patient's condition can change after the first assessment, while new vital signs and laboratory results arrive over time. Smith and the hosts discuss whether AI could help teams notice when reassessment is needed. They also emphasize the knowledge of experienced triage nurses and the importance of including them in design. A useful system would need to fit that clinical work, communicate at an appropriate moment, and leave room for a team member to raise a concern that the model has missed.
Clinical decision support does not have to mean another pop-up. Smith discusses presenting the right information to the right person in a useful format and at the right point in care. A visual cue, a clearer summary, or information placed where a clinician is already working may be more useful than an interruption. The conversation also examines the need to evaluate tools over time, including changes in performance and differences across patient populations. Bedside explanations and deeper technical review serve different purposes, and both need an audience in mind.
The same attention to people shapes the discussion of quality improvement and education. Feedback about diagnostic decisions should create opportunities to learn, with psychological safety and room for reflection. In training, AI assistance brings questions about which skills learners need to develop independently and how they will evaluate generated content. Smith describes UVA's gradual introduction of ambient documentation across groups of clinicians. The group also considers future support for handoffs and discharge communication, without assuming those possibilities have already become routine practice.
Before the interview, DJ and Ethan discuss Utah's AI prescription-renewal pilot and a New England Journal of Medicine perspective on it. Their questions concern evidence, oversight, access, and what might be lost when a clinical interaction is automated. The interview carries those questions into everyday emergency care. Smith's approach centers on a tool's actual purpose: helping clinicians find, interpret, and act on information while preserving the judgment and relationships on which good care depends.
What we cover
• 🧠 Cognitive overload and the conditions behind diagnostic error
• 🗂️ Reconstructing the patient's story from a complex record
• 🚦 Triage reassessment and the expertise of emergency nurses
• 🔔 Decision support beyond another interruptive pop-up
• 🔎 Ongoing evaluation, bias, and explanations for different users
• 🎓 Psychological safety, feedback, and medical education
• 🎙️ Ambient documentation and future support for handoffs
Papers and resources discussed
• Leveraging artificial intelligence to reduce diagnostic errors in emergency medicine: Challenges, opportunities, and future directions, coauthored by Smith. Academic Emergency Medicine, 2025;32:327–339. Full text.
• Utah's Prescription-Renewal Pilot Program: Autonomous AI Managing Patient Care. New England Journal of Medicine, 2026. Discussed in the hosts' opening segment.
Connect with Dr. Smith
University of Virginia faculty profile
