Orchestra-ICU: A Multi-Agent Platform for Augmenting Diagnostic Reasoning During ICU Rounds, Majid Afshar

Speaker: Majid Afshar, MD, MS | Tenured Associate Professor, School of Medicine and Public Health, UW–Madison

Date: Monday, June 15th, 2025

Time: 2:00 PM Central Time

Location:  WIMR2409 (Zoom available)

Title: “Orchestra-ICU: A Multi-Agent Platform for Augmenting Diagnostic Reasoning During ICU Rounds”

Abstract: Diagnostic errors remain a leading cause of preventable harm in the ICU, where clinicians must synthesize vast amounts of evolving patient data under intense time pressure. While large language models offer promise for augmenting clinical reasoning, single-model approaches struggle with the breadth, specialization, and verification demands of real ICU workflows. Our prior work building specialized agents for information retrieval, evaluation of hallucinations and omissions, AI scribes, and diagnostic generation has converged into Orchestra-ICU, a modular multi-agent platform designed to augment physician diagnostic reasoning in real time during ICU rounds. Early prototype work demonstrates direct integration into clinical workflows and provides a path into studying Human–AI teaming at the bedside. In this talk, we will describe how Orchestra-ICU is being developed and validated as a diagnostic decision support system for critically ill patients, the rationale behind a multi-agent rather than monolithic approach, and what we are learning about evaluating LLM-driven augmentation in high-acuity clinical environments.

   

Bio: Majid Afshar, MD, MS is a tenured Associate Professor at the University of Wisconsin–Madison and co-leads an AI research lab. He also serves as Director of the Learning Health System and Associate Director of Data Science for the School of Medicine’s Institute for Clinical and Translational Research (ICTR), focusing on AI-driven interventions in healthcare.Dr. Afshar has helped build a program to evaluate AI at the bedside, integrating clinical trials with health operations. He guided one of the first real-time large language model pipelines for clinical decision support, Ambient AI scribes for healthcare practitioners, and is leading initiatives to responsibly deploy generative AI technologies in pragmatic trials.As a physician-scientist, his research centers on early disease detection using natural language processing and predictive analytics with electronic health record data. He has organized national data challenges to advance diagnostic decision support, serves as a chartered NIH study section member for AI grants, and has co-chaired national informatics conferences.Dr. Afshar has mentored numerous trainees, co-authored the TRIPOD-LLM guidelines for reporting LLM-based research, and has over 180 publications including Nature Medicine and NEJM AI.