Skip to content

S5 — Fairness and Bias in Biomedical AI/ML

Full session title: Fairness and Bias in Biomedical AI/ML: Defining Goals and Putting Them Into Practice Papers: 5 (pp. 524–595) Session introduction: Martinez-Martin N. et al., pp. 520–523 · PDF · PMID 41758165

Organizers — all four from the Center for Biomedical Ethics, Stanford University School of Medicine:

  • Nicole Martinez-Martin
  • Mildred K. Cho
  • Abdoul Jalil Djiberou Mahamadou
  • Madalena Ng (rendered "Magdalena Ng" in the session-intro byline)

Invited talk: Katie Shilton (Wed Jan 7, 8:50 a.m.) — affiliation not printed in the schedule.

Reading this session

The subtitle carries the argument: defining goals and putting them into practice. The session is deliberately weighted toward operational work rather than critique — three of the five papers are tools or protocols.

Auditing at scale. Ansari et al. use LLMs to audit healthcare model biases — turning the model class that creates the problem into part of the detection apparatus, which is either elegant or circular depending on how the evaluation is set up. Worth reading closely.

Detection to mitigation. Mottez et al. work on bias in deep learning for chest X-ray diagnosis. This pairs directly with S4's Automated Chest X-ray Report Generation Remains Unsolved — same substrate, two different failure analyses, in adjacent sessions at the same meeting.

Empirical disparity. Coggan et al. examine how demographic factors influence the treatment of pediatric patients in the emergency department — the measurement paper the rest of the session is arguing about.

Study-design ethics. Sun et al. document barriers to designing inclusive ecological-momentary-assessment and wearable protocols for AI-driven substance-use monitoring in Hawai'i — a study sited where the meeting was held. Foti et al. deliver a practical tool for identifying key decision points in building fair and trustworthy biomedical AI.

Papers

Title First author Pages PDF PMID
Using Large Language Models to Audit Model Healthcare Biases Zara N. Ansari 524–537 ansari.pdf 41758166
From Detection to Mitigation: Addressing Bias in Deep Learning Models for Chest X-Ray Diagnosis Clemence Mottez 538–550 mottez.pdf 41758167
Deciphering the Influence of Demographic Factors on the Treatment of Pediatric Patients in the Emergency Department Helena Coggan 551–565 coggan.pdf 41758168
Barriers to Designing Inclusive Ecological Momentary Assessment and Wearable Data Collection Protocols for AI-Driven Substance Use Monitoring in Hawai'i Yinan Sun 566–579 sun.pdf 41758169
Building Fair and Trustworthy Biomedical AI: A Tool for Identifying Key Decision Points Nicole Foti 580–595 foti.pdf 41758170

Connections to the corpus

  • Chest X-ray bias ↔ chest X-ray benchmark failure. Read mottez.pdf next to zhang_x.pdf in S4; together they are a compact case that imaging-AI evaluation is under-specified in two independent directions.
  • Equity in precision medicine is a thread the corpus already touches at AACR — see the "Precision for All" session in aacr-2026/sessions/.
  • This session has no direct oncology content, but the audit and decision-point tooling applies to any clinical-biomarker deployment the corpus tracks.

Sources