长住院出院小结由 LLM 生成的临床医生评估:事实准确性更高但遗漏少
Clinician-centered evaluation of large language model-generated discharge summaries for longer hospitalizations
一项发表于npj Digital Medicine的研究对60个住院时长7-21天的内科病例进行了评估。与医生手写小结相比,临床医生在95%的病例中更偏好LLM生成版本,且在事实准确性、可读性和完整性方面评分更高。在LLM识别的31个影像学附带发现中,93.5%被评为事实准确,87.1%适合上报,未发现比手写小结更高的潜在伤害可能性。
- Article
- Open access
- Published: 02 October 2026
- Tyler Osborne1,
- Tauhid Mahmud2,
- Xia Zheng3,4,
- Sai Jampala1,
- Sadia Abbasi5,
- Stephanie Hong5,
- Kimberly Kranz5,
- Susan Lee5,
- Patricia Ng5,
- Kerim Odekon5,
- Lindsey Schachter5,
- Robert Sexton5,
- Tracey Spinnato5,
- Mathew Tharakan2,5,
- Zhi Jian Wu5,
- Fusheng Wang ORCID: orcid.org/0000-0002-9369-93611,2 &
- …
- Rachel Wong2,5
npj Digital Medicine (2026) Cite this article
We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.
Abstract
Although large language models (LLMs) have shown promise for discharge summary generation, their performance in longer hospitalizations remains less well studied. Prior evaluations of LLM-generated discharge summaries have largely involved shorter stays and have rarely examined receiving-clinician priorities or incidental finding reporting. We compared LLM-generated and human-authored discharge summaries for 60 Internal Medicine hospitalizations lasting 7 to 21 days, with paired assessment by hospitalists and primary care physicians (PCPs). Clinician reviewers preferred LLM-generated summaries for 95% of encounters and rated them higher for quality, readability, factuality and completeness. PCPs rated LLM-generated summaries as better for understanding and communicating hospital care to patients and providing follow-up care. LLM-generated summaries had fewer clinician-reviewer annotations of either factual inaccuracies or clinically relevant omissions, primarily due to fewer omissions, with no significant differences in estimated harm potential or likelihood compared with human-authored summaries. PCPs annotated more omissions and assigned higher likelihood of harm than hospitalists. Among 31 LLM-identified radiology incidental findings, 93.5% were rated factually correct and 87.1% appropriate for reporting. In this retrospective clinician evaluation, LLM-generated summaries for longer, complex hospitalizations were rated more favorably and had fewer annotated omissions than human-authored summaries, with no significant differences in estimated harm potential or likelihood.
Acknowledgements
First, we would like to acknowledge Fang Wang, a Clinical Research Data Analyst/Architect with the Research Computing & Informatics (RCI) team at Stony Brook University, for her efforts in extracting the raw EHR dataset used in this study. Second, we would like to acknowledge Erich Bremer, MSc, the Director of Applied Informatics, and Joe Balsamo, a Supervising Programmer Analyst, from the Stony Brook Medicine Department of Biomedical Informatics, for their efforts in setting up the servers and access necessary for us to develop the HIPAA-compliant full-stack web application to collect review data from the clinicians.
Author information
Authors and Affiliations
Department of Computer Science, Stony Brook University, Stony Brook, NY, USA
Tyler Osborne, Sai Jampala & Fusheng Wang
Department of Biomedical Informatics, Stony Brook University, Stony Brook, NY, USA
Tauhid Mahmud, Mathew Tharakan, Fusheng Wang & Rachel Wong
School of Communication and Journalism, Stony Brook University, Stony Brook, NY, USA
Xia Zheng
The Alan Alda Center for Communicating Science, Stony Brook, NY, USA
Xia Zheng
Department of Internal Medicine, Stony Brook Medicine, Stony Brook, NY, USA
Sadia Abbasi, Stephanie Hong, Kimberly Kranz, Susan Lee, Patricia Ng, Kerim Odekon, Lindsey Schachter, Robert Sexton, Tracey Spinnato, Mathew Tharakan, Zhi Jian Wu & Rachel Wong
Authors
- Tyler Osborne
- Tauhid Mahmud
- Xia Zheng
- Sai Jampala
- Sadia Abbasi
- Stephanie Hong
- Kimberly Kranz
- Susan Lee
- Patricia Ng
- Kerim Odekon
- Lindsey Schachter
- Robert Sexton
- Tracey Spinnato
- Mathew Tharakan
- Zhi Jian Wu
- Fusheng Wang
- Rachel Wong
Corresponding authors
Correspondence to Fusheng Wang or Rachel Wong.
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Osborne, T., Mahmud, T., Zheng, X. et al. Clinician-centered evaluation of large language model-generated discharge summaries for longer hospitalizations. npj Digit. Med. (2026). https://doi.org/10.1038/s41746-026-03320-y
Received: 03 June 2026
Accepted: 16 September 2026
Published: 02 October 2026
DOI: https://doi.org/10.1038/s41746-026-03320-y
来源:Nature 健康护理主题 · nature.com