A systematic review of clinical translation of large language models using electronic health records
Abstract We mapped clinical translation of large language models (LLMs) evaluated using electronic health record data. PubMed, Embase, Scopus and IEEE Xplore were searched for peer-reviewed empirical reports published between 1 January 2022 and 1 June 2026. LLM agent-assisted screening and profiling, human validation and targeted full-text review produced a descriptive map of 1,870 reports. Primary classifications came from full texts for 181 reports and titles and abstracts for the remainder. Most evaluations were retrospective or offline (92.3%); 56.2% used a single dataset or setting. Integrated, operational, interventional or outcome-impact evaluation occurred in 4.6%; 3.1% combined these evaluations with validation across multiple datasets or settings. Evidence connecting broader validation with clinical use remains limited. Translation into practice requires transportability, accountable human oversight, prospective evaluation in real-world workflows and clinically meaningful outcomes.
Authors
- Taiyu Zhu (ORCID: https://orcid.org/0000-0002-9782-3470)
- Linglong Qian (ORCID: https://orcid.org/0000-0003-3930-045X)
- Richard Dobson
- Mandy Miresmaeili
- Christopher Tomlinson
- Zina Ibrahim
- James Teo
Institutions
- King's College London (GB)
- King's College Hospital NHS Foundation Trust (GB)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-17
- DOI
- https://doi.org/10.21203/rs.3.rs-11651581/v1
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- preprint
Funders
- King's College London