The Bias is in the Code: How AI and Algorithms Shape Emergency Care—for Better or Worse (Plenary VII)

Wednesday, April 15, 2026

9:45 AM - 10:30 AM PT

Location: Grand III - 4th Floor

CE: 0.75 CME Credit

Speaker(s)

Cortlyn Brown, MD, MCSO, FAAEM (she/her/hers)

Vice Director for DEI
Carolinas
Charlotte, North Carolina

Disclosure(s): No financial relationships to disclose

Co-Speaker(s)

Italo M. Brown, MD MPH FAAEM

Assistant Professor, Emergency Medicine
Stanford School of Medicine
Hayward, California

Disclosure(s): Onviv: Consultant/Advisory Board (Terminated, February 1, 2025)
Kristyn J. Smith, DO, FAAEM (she/her/hers)

Assistant Professor of Clinical Emergency Medicine
Hospital of the University of Pennsylvania
Philadelphia, Pennsylvania

Disclosure(s): No financial relationships to disclose

Session Moderator(s)

Zachary Repanshek, MD, FAAEM

Professor and Associate Program Director
Lewis Katz School of Medicine at Temple University
Philadelphia, Pennsylvania

Disclosure(s): No financial relationships to disclose
Phillip A. Dixon, MD MBA FAAEM CHCQM-PHYADV

Medical Director, Utilization Management
The Ohio State University WMC
Delaware, Ohio

Disclosure(s): No financial relationships to disclose

Disclosure(s):

Cortlyn Brown, MD, MCSO, FAAEM: No financial relationships to disclose

Italo M. Brown, MD MPH FAAEM: Onviv: Consultant/Advisory Board (Terminated, February 1, 2025)

Kristyn J. Smith, DO, FAAEM: No financial relationships to disclose

A 45-year-old Black male presents to the ED with shortness of breath. Despite abnormal vitals, the triage algorithm assigns him a lower acuity score than a white patient with identical findings. The sepsis alert fails to fire. He is evaluated an hour later—now in shock.

What went wrong?

A widely used commercial risk prediction algorithm, trained using prior healthcare spending as a proxy for illness severity, underestimates his risk—a classic case of historical bias. The Epic Sepsis Model, trained on data not representative of diverse populations, misses him due to data bias. His triage nurse assigns a lower ESI score, and that score becomes part of the model’s training data—label bias. Because he was flagged as low risk, he received fewer labs and less documentation. This incomplete data feeds the next model iteration, further reducing the risk scores for future Black patients—a feedback loop bias.

As emergency departments adopt AI tools for triage, risk stratification, and decision support, we must recognize: AI is not neutral. These systems reflect the inequities embedded in the data and workflows from which they’re built.

In this session, we will:
• Review real-world examples of bias types in AI, including Epic’s Sepsis Model
• Explain how bias enters the system—from data collection to deployment
• Share strategies to audit, mitigate, and monitor bias in your department
• Offer key questions emergency physicians should ask before adopting AI tools

This talk will equip emergency medicine providers with the tools to be thoughtful stewards of AI—leveraging innovation while protecting equity, safety, and patient trust.

Learning Objectives:

Identify sources and mechanisms of bias in AI models
Analyze real case examples where biased algorithms affected emergency care outcomes
Advocate for equity-focused oversight and transparency when adopting new AI tools in emergency departments