Cultural Influences on the Compliance of Large Language Models with Neuroradiology Guidelines
Overview
This study investigates the potential geographic bias of large language models (LLMs) in adhering to neuroradiology guidelines, particularly focusing on U.S.-centric recommendations.
Background
The complexity of medical imaging and the frequent updates to clinical guidelines create challenges for clinicians in ordering appropriate studies. Variations in guideline adherence have been documented, with significant discrepancies observed between different countries. Understanding how LLMs align with these guidelines is crucial.
Data Highlights
No numerical data or trial results were provided in the source material.
Key Findings
['Geographic bias in LLMs may misalign clinical workflows with local practice standards.', 'LLMs have shown a tendency to reproduce U.S.-centric values in medical decision support.', 'Conflicting recommendations exist between U.S. and non-U.S. neuroradiology guidelines.', 'Biases in LLMs can perpetuate health disparities due to training data imbalances.', 'Thirty fictitious clinical vignettes were created to assess guideline adherence by LLMs.']
Clinical Implications
Awareness of these biases is essential for clinicians using LLMs in decision-making.
Conclusion
Further investigation into geographic bias in LLMs is needed to ensure alignment with diverse clinical guidelines and practices.
Researchers found misinformation rates ranged from 0% to 57% across platforms and topics, with higher rates reported in several TikTok studies compared with YouTube.
