Evaluating Large Language Model Performance in Generating Clinically Relevant Intensive Care Unit Discharge Summaries
LLMs can generate fluent and readable ICU discharge summaries but may overlook critical clinical details and lack depth compared to human-authored summaries. Further ICU-specific fine-tuning and incorporation of domain-specific knowledge are needed to improve LLM alignment with human expertise.
Authors
- Jai Madhok (ORCID: https://orcid.org/0000-0002-3567-4465)
- Samantha Regala
- Onkar Litake (ORCID: https://orcid.org/0000-0002-5484-5290)
- Ronald G. Pearl (ORCID: https://orcid.org/0000-0002-3998-7637)
- Seshadri C. Mudumbai (ORCID: https://orcid.org/0000-0001-8486-9966)
- A. Evans (ORCID: https://orcid.org/0000-0002-3142-8953)
- Philip Chung (ORCID: https://orcid.org/0000-0002-1194-7510)
- Rodney A. Gabriel (ORCID: https://orcid.org/0000-0003-4443-0021)
- Ji‐Qing Chen (ORCID: https://orcid.org/0000-0002-1598-8811)
- Martín Krause
Institutions
- Duke University (US)
- VA Palo Alto Health Care System (US)
- University of California San Diego (US)
- Stanford University (US)
Publication Details
- Journal
- A&A Practice
- Published
- 2025-09-01
- DOI
- https://doi.org/10.1213/xaa.0000000000002057
- Primary Topic
- Machine Learning in Healthcare
- Type
- article
- Field-Weighted Citation Impact
- 0.00