A medically grounded LLM agent–based tool to detect patient safety events in medical records
Ein medizinisch fundiertes LLM-Agenten-Tool zum Erkennen von Patientensicherheitsereignissen in medizinischen Akten
PLOS Digital Health · 17.09.2026 · 5 (9) · e0001174
DOI: 10.1371/journal.pdig.0001174 ↗Auf einen Blick
SAFE-AI erkannte in 300 EMS-Chartakten epinephrinbezogene Sicherheitsereignisse mit hoher Genauigkeit und schnitt besser ab als Baseline-LLMs. Die Trefferquote lag bei 97,9 % für Überdosierungen und 91,6 % für Verzögerungen.
Fragestellung
Kann SAFE-AI epinephrinbezogene Sicherheitsereignisse in medizinischen Akten zuverlässig erkennen?
Methode
Test von SAFE-AI an 18.402 Zeilen aus 300 EMS-Clinical-Charts; Vergleich mit unabhängiger Doppelbeurteilung durch 2 Expert:innen für epinephrinbezogene ASEs.
Ergebnisse
300 EMS-Charts, 18.402 Zeilen. Übereinstimmung der Ärzt:innen: 96 %. SAFE-AI: 97,9 % Genauigkeit bei Überdosierungen und 91,6 % bei Verzögerungen der Epinephrin-Gabe; besser als Baseline-LLMs.
Schlussfolgerung der Autor:innen
SAFE-AI stellt einen neuen Ansatz für klinische KI-Anwendungen dar und adressiert laut Autor:innen zentrale Schwächen aktueller Machine-Learning-Methoden.
Kernaussagen
- 300 EMS-Charts wurden auf epinephrinbezogene ASEs geprüft
- SAFE-AI erreichte 97,9 % bzw. 91,6 % Genauigkeit
- Basismodell-LLMs schnitten schlechter ab als SAFE-AI
Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.
Kurz gefasst
Fachartikel in der Zeitschrift PLOS Digital Health, online erschienen am 17.09.2026. Verfasst von Diego Trujillo und 11 weiteren Autor:innen. Beteiligte Einrichtung: Center for Learning Health Care Delivery, Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States of America. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NHLBI NIH HHS, National Heart, Lung, and Blood Institute.
Themen: Patientensicherheit
Abstract (Originalsprache)
Large language models (LLMs) have shown incredible promise in medicine. While LLMs may be particularly useful in areas requiring extensive review of clinical records, their use remains limited due to their tendency to hallucinate and fabricate information. Hallucination issues, as well as their consequences, are exacerbated in low–probability, high–stakes scenarios such as rare adverse safety events or medical errors. We present SAFE–AI (Structured and Automated Framework for Explainable AI), a novel method for clinical decision making that combines the strengths of clinical expert knowledge with LLMs in an ontology–driven model that minimizes hallucinations using strict rules. We test this method to identify medication errors in medical charts. We collected a sample of 18,402 lines of clinical information from 300 EMS clinical charts that were independently dually reviewed by two expert physicians for epinephrine adverse safety events (ASEs), with 96% inter-rater agreement. We tested SAFE–AI against these labels, achieving similar performance to human experts in detecting epinephrine overdoses with 97.9% accuracy, and 91.6% accuracy in identifying delays in epinephrine administration, greatly outperforming baseline LLMs models. Notably, some disagreements between clinicians and the model were found to be justifiable differences in judgment rather than errors. SAFE-AI presents a novel approach for clinical AI applications that addresses two key limitations of current machine learning methods: 1) over-reliance on probabilistic pattern recognition instead of established medical knowledge, and 2) perpetuation of biases present in training data. This framework is easily adaptable to a range of clinical applications, paving the way for provable and trustworthy AI in medicine.
Autor:innen und Institutionen
- Center for Learning Health Care Delivery, Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States of America.
- Department of Obstetrics Gynecology and Reproductive Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts, United States of America.
- McWilliams School of Biomedical Informatics, UTHealth, Houston, Texas, United States of America.
- Department of Emergency Medicine, Oregon Health and Science University, Portland, Oregon, United States of America.
- Department of Biomedical Informatics, Harvard Medical School, Boston, Massachusetts, United States of America.
- Department of Pediatrics and Emergency Medicine, University of British Columbia, Vancouver, Canada.
ORCID-Profile: Tina Yi-Jin Hsieh, Byeongyeon Cho, Xiaoqian Jiang, Jeanne-Marie Guise
Förderung
- NHLBI NIH HHS
- National Heart, Lung, and Blood Institute
Zitieren
APA
Trujillo, D., Wang, D., Bahr, N., Yi-Jin Hsieh, T., Cho, B., Meckler, G., Hansen, M., Eriksson, C., Seo Kim, K., Bedrick, S., Jiang, X., & Guise, J. M. (2026). A medically grounded LLM agent–based tool to detect patient safety events in medical records. PLOS Digital Health, 5(9), e0001174. https://doi.org/10.1371/journal.pdig.0001174
Vancouver
Trujillo D, Wang D, Bahr N, Yi-Jin Hsieh T, Cho B, Meckler G, et al. A medically grounded LLM agent–based tool to detect patient safety events in medical records. PLOS Digital Health. 2026;5(9):e0001174. doi:10.1371/journal.pdig.0001174
BibTeX
@article{Trujillo2026,
author = {Trujillo, Diego and Wang, Dulin and Bahr, Nathan and Yi-Jin Hsieh, Tina and Cho, Byeongyeon and Meckler, Garth and Hansen, Matthew and Eriksson, Carl and Seo Kim, Kyu and Bedrick, Steven and Jiang, Xiaoqian and Guise, Jeanne-Marie},
title = {A medically grounded LLM agent–based tool to detect patient safety events in medical records},
journal = {PLOS Digital Health},
year = {2026},
volume = {5},
number = {9},
pages = {e0001174},
doi = {10.1371/journal.pdig.0001174}
}
EUVIMED