EP15 Dr. Andrew Taylor, MD MHS
ECG Foundation Models, Human-in-the-Loop Triage, and the Road to Equitable AI in the ED
Guest: Andrew Taylor, MD, MHS — Professor of Emergency Medicine & Vice Chair for Research and Innovation, University of Virginia
Release Date: 9/02/25
Episode Summary:
In this episode of the STAT AI Podcast, Drs. DJ Apakama and Ethan Abbott sit down with Dr. Andrew Taylor, one of the nation’s leading voices in clinical informatics and emergency medicine innovation. The conversation spans two of the field’s most exciting frontiers: ECG foundation models and human-in-the-loop triage AI.
DJ and Ethan kick things off by unpacking a landmark study on a 10.7-million-record ECG foundation model that generalizes across both 12-lead and single-lead waveforms. They discuss how lead-augmentation and positive–unlabeled learning offer earlier risk signals and external validation across multiple health systems—while raising critical questions about representation and fairness when training data under-represent certain populations.
Dr. Taylor then brings the discussion to the bedside, sharing results from a multi-site implementation of AI-assisted triage with nurse override. His team’s work redistributed patients out of the “ESI-3 blob,” improved throughput, and showed early evidence that AI can reduce disparities when thoughtfully deployed. Together, they explore the nuances of trust, automation bias, clinical override (“never let testicular pain be triaged away”), and what a practical AI partnership could look like in 2030—ambient intake, policy retrieval chatbots, and wearables streaming clinically useful signals.
The episode closes with Taylor reflecting on how he uses modern code-assist tools to prototype apps in minutes, why outcomes—not accuracy scores—must drive adoption, and how we should be teaching residents to leverage AI critically without falling into blind trust.
In This Episode, We Cover:
⚡ How ECG foundation models enable new risk prediction across 12-lead and single-lead data
🫀 Bias and representation challenges in national-scale cardiology datasets
👩⚕️ AI-assisted triage with nurse override—and the lessons learned from “the ESI-3 blob”
⚖️ Automation bias, liability gray zones, and when to override the algorithm
🛠️ Using modern coding copilots to accelerate clinical innovation
📊 Why outcomes matter more than AUROC in real-world deployment
🧑🎓 Teaching residents to partner with AI while keeping clinical judgment sharp
Papers Discussed This Episode:
Li J, Aguirre A, Moura J, et al. An electrocardiogram foundation model built on over 10 million recordings with external evaluation across multiple domains. arXiv. Preprint posted online October 5, 2024. doi:10.48550/arXiv.2410.04133
Hinson JS, Levin SR, Steinhart BD, et al. Enhancing emergency department triage equity with artificial intelligence: outcomes from a multisite implementation. Ann Emerg Med. 2025;85(3):288-290. doi:10.1016/j.annemergmed.2024.10.014
Connect with the Guest:
University of Virginia Department of Emergency Medicine
Andrew Taylor, MD, MHS (University Profile & Linkedin)
Follow STAT AI:
🌐 Website: stataipodcast.com
🐦 Twitter/X: @STATaipodcast
🎧 Listen: Apple Podcasts | Spotify | Prime Music
Tags: #AIinHealthcare #EmergencyMedicine #ECGAI #ClinicalDecisionSupport #FoundationModels #EDWorkflow #DigitalHealth #EquityInAI #HumanInTheLoop #FutureOfEM
