Auditing sex/gender disparities in emergency triage with LLM-based paired comparisons
Abstract
We present a domain-agnostic paired-comparison approach that uses Large Language Models (LLMs) to quantify sex/gender-related asymmetries in documented clinical decision-making. The method trains an LLM to emulate observed decisions, then evaluates sex-swapped pairs in which only sex is flipped, holding documented clinical content constant. We apply it to emergency triage, analyzing more than 140,000 Bordeaux University Hospital (France) admissions and testing methodological portability on MIMIC-IV, spanning a different language, population, and healthcare system. Fine-tuning Mistral NeMo 12B for triage prediction and using Mistral Small 24B for pair generation, we find otherwise identical presentations were more likely to receive a lower-severity predicted score as female than male: 1.1% (95% CI 0.9–1.3) in the French cohort, 2.2% (1.7–2.7) in MIMIC-IV. Predictions are sensitive to both tabular sex markers and textual gender cues, with the asymmetry emerging primarily in the combined bimodal setting. A model retrained on sex-neutralized inputs eliminated the between-sex prediction gap, indicating the asymmetry is mediated by explicit sex markers. Patterns vary with nurse-patient sex concordance, suggesting the model captures stable features of the recorded data rather than random artifacts. These effects are small and documentation-level. We therefore present this as a methodological feasibility study: LLMs can serve as scalable probes of documented decisions, generating hypotheses rather than establishing bedside clinician behavior or clinically meaningful undertriage, which would require clinician-anchored validation. Beyond emergency care, the approach supports bias audits in other domains.
Related
- Illuminating gendered ageism with the WHO ageism experiences scale: a psychometric analysis and gender comparisons in a multi-country dataset.
- Sample overlap in genome-wide association meta-analysis: practical comparisons across significance thresholds
- Comparisons among Yanomami populations reveal transitional gut microbiome configurations under urban influence
- A ratiometric ER calcium sensor for quantitative comparisons across cell types and subcellular regions
- Autonomous generalizability auditing of scientific foundation models
BibTeX
@article{adames2026auditing,
title = {{Auditing sex/gender disparities in emergency triage with LLM-based paired comparisons}},
author = {Ariel Guerra Adames and Marta Avalos-Fernandez and Océane Dorémus and Leo Anthony Celi and Cédric Gil-Jardiné and Emmanuel Lagarde},
journal = {npj Digital Medicine},
year = {2026},
month = sep,
doi = {10.1038/s41746-026-03090-7},
eprint = {2511.17124},
archivePrefix = {arXiv},
url = {https://hal.science/hal-05779198v1},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Auditing sex/gender disparities in emergency triage with LLM-based paired comparisons
AU - Ariel Guerra Adames
AU - Marta Avalos-Fernandez
AU - Océane Dorémus
AU - Leo Anthony Celi
AU - Cédric Gil-Jardiné
AU - Emmanuel Lagarde
PY - 2026
DA - 2026/09/16
JO - npj Digital Medicine
DO - 10.1038/s41746-026-03090-7
UR - https://hal.science/hal-05779198v1
AB - We present a domain-agnostic paired-comparison approach that uses Large Language Models (LLMs) to quantify sex/gender-related asymmetries in documented clinical decision-making. The method trains an LLM to emulate observed decisions, then evaluates sex-swapped pairs in which only sex is flipped, holding documented clinical content constant. We apply it to emergency triage, analyzing more than 140,000 Bordeaux University Hospital (France) admissions and testing methodological portability on MIMIC-IV, spanning a different language, population, and healthcare system. Fine-tuning Mistral NeMo 12B for triage prediction and using Mistral Small 24B for pair generation, we find otherwise identical presentations were more likely to receive a lower-severity predicted score as female than male: 1.1% (95% CI 0.9–1.3) in the French cohort, 2.2% (1.7–2.7) in MIMIC-IV. Predictions are sensitive to both tabular sex markers and textual gender cues, with the asymmetry emerging primarily in the combined bimodal setting. A model retrained on sex-neutralized inputs eliminated the between-sex prediction gap, indicating the asymmetry is mediated by explicit sex markers. Patterns vary with nurse-patient sex concordance, suggesting the model captures stable features of the recorded data rather than random artifacts. These effects are small and documentation-level. We therefore present this as a methodological feasibility study: LLMs can serve as scalable probes of documented decisions, generating hypotheses rather than establishing bedside clinician behavior or clinically meaningful undertriage, which would require clinician-anchored validation. Beyond emergency care, the approach supports bias audits in other domains.
SN - 2398-6352
ER -