Methods to ensure factual accuracy of text generated by large language models (LLM) in clinical medicine are lacking. VeriFact is an artificial intelligence system that combines retrieval-augmented generation and LLM-as-a-Judge to verify whether LLM-generated text is factually supported by a patient's medical history based on their electronic health record (EHR). To evaluate this system, we introduce VeriFact-BHC, a new dataset that decomposes Brief Hospital Course narratives from discharge summaries into a set of simple statements with clinician annotations for whether each statement is supported by the patient's EHR clinical notes. Whereas highest agreement between clinicians was 88.5%, VeriFact achieves up to 92.7% agreement when compared to a denoised and adjudicated average human clinican ground truth, suggesting that VeriFact exceeds the average clinician's ability to fact-check text against a patient's medical record. VeriFact may accelerate the development of LLM-based EHR applications by removing current evaluation bottlenecks.
@article{arxiv.2501.16672,
title = {VeriFact: Verifying Facts in LLM-Generated Clinical Text with Electronic Health Records},
author = {Philip Chung and Akshay Swaminathan and Alex J. Goodell and Yeasul Kim and S. Momsen Reincke and Lichy Han and Ben Deverett and Mohammad Amin Sadeghi and Abdel-Badih Ariss and Marc Ghanem and David Seong and Andrew A. Lee and Caitlin E. Coombes and Brad Bradshaw and Mahir A. Sufian and Hyo Jung Hong and Teresa P. Nguyen and Mohammad R. Rasouli and Komal Kamra and Mark A. Burbridge and James C. McAvoy and Roya Saffary and Stephen P. Ma and Dev Dash and James Xie and Ellen Y. Wang and Clifford A. Schmiesing and Nigam Shah and Nima Aghaeepour},
journal= {arXiv preprint arXiv:2501.16672},
year = {2025}
}