Building Trust in Clinical LLMs: Bias Analysis and Dataset Transparency

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Maslenkova, Svetlana, Christophe, Clement, Pimentel, Marco AF, Raha, Tathagata, Salman, Muhammad Umar, Mahrooqi, Ahmed Al, Gupta, Avani, Khan, Shadab, Rajan, Ronnie, Kanithi, Praveenkumar
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914105265225728
author Maslenkova, Svetlana
Christophe, Clement
Pimentel, Marco AF
Raha, Tathagata
Salman, Muhammad Umar
Mahrooqi, Ahmed Al
Gupta, Avani
Khan, Shadab
Rajan, Ronnie
Kanithi, Praveenkumar
author_facet Maslenkova, Svetlana
Christophe, Clement
Pimentel, Marco AF
Raha, Tathagata
Salman, Muhammad Umar
Mahrooqi, Ahmed Al
Gupta, Avani
Khan, Shadab
Rajan, Ronnie
Kanithi, Praveenkumar
contents Large language models offer transformative potential for healthcare, yet their responsible and equitable development depends critically on a deeper understanding of how training data characteristics influence model behavior, including the potential for bias. Current practices in dataset curation and bias assessment often lack the necessary transparency, creating an urgent need for comprehensive evaluation frameworks to foster trust and guide improvements. In this study, we present an in-depth analysis of potential downstream biases in clinical language models, with a focus on differential opioid prescription tendencies across diverse demographic groups, such as ethnicity, gender, and age. As part of this investigation, we introduce HC4: Healthcare Comprehensive Commons Corpus, a novel and extensively curated pretraining dataset exceeding 89 billion tokens. Our evaluation leverages both established general benchmarks and a novel, healthcare-specific methodology, offering crucial insights to support fairness and safety in clinical AI applications.
format Preprint
id arxiv_https___arxiv_org_abs_2510_18556
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Building Trust in Clinical LLMs: Bias Analysis and Dataset Transparency
Maslenkova, Svetlana
Christophe, Clement
Pimentel, Marco AF
Raha, Tathagata
Salman, Muhammad Umar
Mahrooqi, Ahmed Al
Gupta, Avani
Khan, Shadab
Rajan, Ronnie
Kanithi, Praveenkumar
Computation and Language
Large language models offer transformative potential for healthcare, yet their responsible and equitable development depends critically on a deeper understanding of how training data characteristics influence model behavior, including the potential for bias. Current practices in dataset curation and bias assessment often lack the necessary transparency, creating an urgent need for comprehensive evaluation frameworks to foster trust and guide improvements. In this study, we present an in-depth analysis of potential downstream biases in clinical language models, with a focus on differential opioid prescription tendencies across diverse demographic groups, such as ethnicity, gender, and age. As part of this investigation, we introduce HC4: Healthcare Comprehensive Commons Corpus, a novel and extensively curated pretraining dataset exceeding 89 billion tokens. Our evaluation leverages both established general benchmarks and a novel, healthcare-specific methodology, offering crucial insights to support fairness and safety in clinical AI applications.
title Building Trust in Clinical LLMs: Bias Analysis and Dataset Transparency
topic Computation and Language
url https://arxiv.org/abs/2510.18556