Framework for bias evaluation in large language models in healthcare settings
Система оценки предвзятости больших языковых моделей в условиях здравоохранения
2025-07-07
SCID: 54.1/evkbyvej
Discuss with AI
audit frameworkbias evaluationclinical decision supporthealthcare settingslarge language models
Figures from the paper
Abstract (AI)
A critical gap in the adoption of large language models for AI-assisted clinical decisions is the lack of a standardized audit framework to evaluate models for accuracy and bias. Our framework introduces a five-step framework that guides practitioners through stakeholder engagement, model calibration to specific patient populations, and rigorous testing through clinically relevant scenarios. We provide open-access tools for stakeholder engagement and an example of an audit. As the regulation of models becomes more critical, we believe adoption of an audit framework that tests model outputs, rather than regulating specific hyperparameters or inputs, will encourage the responsible use of AI in clinical settings.
Key Findings
1
It introduces a five-step audit framework covering stakeholder engagement, calibration to specific patient populations, and testing with clinically relevant scenarios.
2
The authors argue that regulating and auditing model outputs, rather than prescribing specific hyperparameters or inputs, better supports responsible clinical AI use.
3
The framework is positioned to address increasing regulatory needs while facilitating safer adoption of large language models in healthcare.
4
The framework provides open-access stakeholder-engagement tools and an illustrative audit example for practical adoption.
5
The paper identifies a critical gap: no standardized framework adequately evaluates large language models for both accuracy and bias in clinical decision support.
Research Object
large language models used for AI-assisted clinical decision-making in healthcare settings
Research Subject
accuracy and bias in model outputs, evaluated through a standardized clinical audit framework across specific patient populations and clinically relevant scenarios
Publication Details
Publication Date
2025-07-07
Journal
Publisher
ISSN
Cited by
61
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai4
Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models2023
Revolutionizing healthcare: the role of artificial intelligence in clinical practice2023
Fairness in Machine Learning: A Survey2023
Large language models could change the future of behavioral healthcare: a proposal for responsible development and evaluation2024