The DOI matches its Crossref record; 2 of the 2 references it cites were found in a registry or an index.
Elucidating Mechanisms of Demographic Bias in LLMs for Healthcare
Open accessAbstract
We know from prior work that LLMs encode social biases, and that this manifests in clinical tasks (Gerszberg, 2024;Zack et al., 2024;Zhang et al., 2020).In this work we adopt tools from mechanistic interpretability to unveil sociodemographic representations and biases within LLMs in the context of healthcare.Specifically, we ask: Can we identify activations within LLMs that encode sociodemographic information (e.g., gender, race)?We find that, in three open weight LLMs, gender information is highly localized in MLP layers and can be reliably manipulated at inference time via patching.Such interventions can surgically alter generated clinical vignettes for specific conditions, and also influence downstream clinical predictions which correlate with gender, e.g., patient risk of depression.We find that representation of patient race is somewhat more distributed, but can also be intervened upon, to a degree.To our knowledge, this is the first application of mechanistic interpretability methods to LLMs for healthcare 1 . 1 Our code is available at https://github.com/ hibaahsan/interp-healthcare-bias/ "Generate a clinical vignette of a patient with Multiple Sclerosis … "male" patch LLM ( ) "Is this patient at risk of depression?
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Yet these methods offer capabilities that generative approaches structurally cannot: training-data attribution [129], mechanistic localization and intervention on internal representations [114], and formal faithfulness guarantees from transparent surrogates [125].”
From Explainability of decoder-only clinical large language models: A scoping review (2026) · cites this paper as [114]
In the citing paper
“Notably, every study conducting mechanistic or internal analysis: influence functions [129], attention visualization [113,116, 128], activation patching [114], necessarily used open-source models.”
From Explainability of decoder-only clinical large language models: A scoping review (2026) · cites this paper as [114]
In the citing paper
“with only one study in our review [114] employing such techniques.”
From Explainability of decoder-only clinical large language models: A scoping review (2026) · cites this paper as [114]
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Cited by
Sorted by citations
Kaliosis, P., Ganesan, A. V., Kjell, O., Ringwald, W. R., Feltman, S. M., Carr, M. A., Samaras, D., Ruggero, C. J., Luft, B. J., Kotov, R., et al. (2025) · Research Square · doi:10.21203/rs.3.rs-8376581/v1 · Cited by 2
Explainability of decoder-only clinical large language models: A scoping review
Mishra, N., Abu-Hanna, A., & Calixto, I. (2026) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2026.103501
Stupnicki, A., Mendes, B., Reinstein, M., Ong, A. Y., Malem, A. H., Keane, P. A., & Thirunavukarasu, A. J. (2026) · Research Square · doi:10.21203/rs.3.rs-9383173/v1
Puzis, R., Cohen, I.-C., Elyashar, A., & Cohen, O. (2026) · doi:10.2196/preprints.107902
Metadata
- Authors
- Ahsan, H., Sen Sharma, A., Amir, S., Bau, D., & Wallace, B. C.
- Venue
- Findings of the Association for Computational Linguistics: EMNLP 2025
- Published
- 2025
- Pages
- 14614-14631
- OpenAlex
- W4416034639
- Cited by
- 1
- References
- 0
- Citations by year
- 2026: 1
- FWCI
- 0.95
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Funders
- Open Philanthropy Project
- National Institutes of Health
- U.S. National Library of Medicine (R01LM013772)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- demographic bias, large language models, health care, mechanistic interpretability, gender bias, racial bias, clinical vignettes, depression risk prediction, multiple sclerosis, clinical prediction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
