Peer-Reviewed Publication
NPJ Digit Med2026July 6, 2026Journal Article
Large language models are powerful electronic health record encoders.
Stefan Hegselmann1,2,3, Georg von Arnim4,5, Tillmann Rheude4,5, Noel Kronenberg4, David Sontag6,7, Gerhard Hindricks8, Roland Eils4,9, Benjamin Wild4,10
1Berlin Institute of Health at Charité - Universitätsmedizin Berlin, Center of Digital Health, Berlin, Germany. stefan.hegselmann@charite.de.
2Berlin Institute of Health at Charité - Universitätsmedizin Berlin, BIH Biomedical Innovation Academy, BIH Charité Digital Clinician Scientist Program, Berlin, Germany. stefan.hegselmann@charite.de.
3Deutsches Herzzentrum der Charité (DHZC) - Universitätsmedizin Berlin, Berlin, Germany. stefan.hegselmann@charite.de.
4Berlin Institute of Health at Charité - Universitätsmedizin Berlin, Center of Digital Health, Berlin, Germany.
5Freie Universität Berlin, Berlin, Germany.
6Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (MIT), Cambridge, MA, USA.
7Layer Health Inc., Boston, MA, USA.
8Deutsches Herzzentrum der Charité (DHZC) - Universitätsmedizin Berlin, Berlin, Germany.
9Intelligent Medicine Institute, Fudan University, Shanghai, China.
10Institute of Medical Informatics (IMI) - Universitätsmedizin Berlin, Berlin, Germany.
Abstract
Electronic health records (EHRs) offer considerable potential for clinical prediction, but their complexity and heterogeneity challenge traditional machine learning. Domain-specific electronic health record foundation models trained on unlabeled EHR data have shown improved predictive accuracy and generalization. However, their development is constrained by limited data access and site-specific vo…
Create a free account to keep reading
Free members get 10 full research views every month across publications, clinical trials, FDA clearances, adverse events, and NIH grants. No credit card required.
Want unlimited research access? See Pro plans