M3: Conversational LLMs simplify secure clinical data access, understanding, and analysis
PLOS Digital Health · 17.09.2026 · 5 (9) · e0001671
DOI: 10.1371/journal.pdig.0001671 ↗Kurz gefasst
Fachartikel in der Zeitschrift PLOS Digital Health, online erschienen am 17.09.2026. Verfasst von Rafi Al Attrach und 5 weiteren Autor:innen. Beteiligte Einrichtung: Laboratory for Computational Physiology, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NIH HHS, FIC NIH HHS, Korea Health Industry Development Institute.
Abstract (Originalsprache)
Large-scale clinical databases offer opportunities for medical research, but their complexity creates barriers to effective use. The Medical Information Mart for Intensive Care (MIMIC-IV), one of the world’s largest open-source electronic health record databases, traditionally requires both SQL proficiency and clinical domain expertise. We introduce M3, a system that enables natural language querying of MIMIC-IV data through the Model Context Protocol. With a single command, M3 retrieves MIMIC-IV from PhysioNet, launches a local SQLite instance or connects to hosted BigQuery, and allows researchers to pose clinical questions in plain English. We evaluated M3 using samples from the EHRSQL 2024 benchmark with two language models. On one hundred answerable questions, the proprietary Claude Sonnet 4 achieved 94% accuracy and the open-weights gpt-oss-20B (deployable locally on consumer hardware) achieved 93%; on a matched sample of one hundred unanswerable questions, where correct behavior is to abstain rather than produce SQL, gpt-oss-20B correctly abstained on 69%. Both models translate natural language into SQL, execute queries against MIMIC-IV, and return structured results alongside the underlying query for verification. Error analysis revealed that most failures stemmed from complex temporal reasoning or ambiguous question phrasing rather than fundamental architectural limitations. The comparable performance of a smaller open-weights model demonstrates that privacy-preserving local deployment is viable for sensitive clinical data analysis. M3 lowers technical barriers to critical care data analysis and is designed with security measures including OAuth2 authentication, query validation, and audit logging.
Autor:innen und Institutionen
- Laboratory for Computational Physiology, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America.
- Department of Computer Science, Technical University of Munich, Munich, Germany.
- Department of Engineering, Universitat Pompeu Fabra, Barcelona, Spain.
- Office of Data Science, St. Jude Children's Research Hospital, Memphis, Tennessee, United States of America.
- Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States of America.
- Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, United States of America.
ORCID-Profile: Rafi Al Attrach, Pedro Moreira, Rajna Fani, Renato Umeton, Amelia Fiske, Leo Anthony Celi
Förderung
- NIH HHS
- FIC NIH HHS
- Korea Health Industry Development Institute
Zitieren
APA
Al Attrach, R., Moreira, P., Fani, R., Umeton, R., Fiske, A., & Celi, L. A. (2026). M3: Conversational LLMs simplify secure clinical data access, understanding, and analysis. PLOS Digital Health, 5(9), e0001671. https://doi.org/10.1371/journal.pdig.0001671
Vancouver
Al Attrach R, Moreira P, Fani R, Umeton R, Fiske A, Celi LA. M3: Conversational LLMs simplify secure clinical data access, understanding, and analysis. PLOS Digital Health. 2026;5(9):e0001671. doi:10.1371/journal.pdig.0001671
BibTeX
@article{AlAttrach2026,
author = {Al Attrach, Rafi and Moreira, Pedro and Fani, Rajna and Umeton, Renato and Fiske, Amelia and Celi, Leo Anthony},
title = {M3: Conversational LLMs simplify secure clinical data access, understanding, and analysis},
journal = {PLOS Digital Health},
year = {2026},
volume = {5},
number = {9},
pages = {e0001671},
doi = {10.1371/journal.pdig.0001671}
}
EUVIMED