Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: May 24, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

De-Identification of Free-Text Medical Records Using Large Language Models.

Richard Noll1, Maximilian Englisch1, Elias Hofmann1

  • 1Goethe University Frankfurt, University Medicine, Institute of Medical Informatics (IMI), Frankfurt am Main, Germany.

Studies in Health Technology and Informatics
|May 23, 2026
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Prediction of Postoperative Complications Following Mandibular Fractures Using Machine Learning.

Studies in health technology and informatics·2026
Same author

Evaluation of a user-oriented inter-physician exchange portal within a decision support system for primary care - a qualitative study.

International journal of medical informatics·2026
Same author

Experiences With Integrating Medical Terminologies Into User Interfaces for a Decision Support System for Primary Care: Conceptual and Development Study.

JMIR medical informatics·2026
Same author

SHARE: towards usable, trustworthy and interoperable synthetic health data for rare diseases.

BMJ health & care informatics·2026
Same author

Retrieval-Augmented Generation for Medical Decision-Making in Emergency Care.

Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference·2025
Same author

Co-creation process of an app for people with rare diseases - a citizen science approach.

Orphanet journal of rare diseases·2025

Large language models (LLMs) offer efficient, automated de-identification of medical records, achieving high accuracy comparable to human annotators. Localized, GPU-free inference is possible with quantized models, enhancing privacy and scalability.

Area of Science:

  • Medical Informatics
  • Artificial Intelligence
  • Natural Language Processing

Background:

  • Automated de-identification of electronic health records (EHRs) is crucial for privacy.
  • Large Language Models (LLMs) show potential for complex text processing tasks.

Purpose of the Study:

  • To evaluate LLMs for automated de-identification of free-text medical records.
  • To compare the performance of a full LLM against a quantized variant for de-identification.
  • To assess the efficiency and privacy implications of LLM-based de-identification.

Main Methods:

  • Utilized 150 synthetically generated, personal health information (PHI)-enriched doctor's letters.
  • Compared the full LLaMA-3.1-8B model (BF16, GPU) with a quantized variant (Q8, CPU).
Keywords:
Artificial IntelligenceElectronic Health RecordsLarge Language ModelsPersonally Identifiable Information

Related Experiment Videos

Last Updated: May 24, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

  • Employed zero-shot and few-shot prompting strategies for evaluation.

    • Main Results:

    • Few-shot prompting improved performance from macro F1 of 0.989 to 0.996.
    • The quantized Q8 model achieved a macro F1 of 0.992 with GPU-free inference.
    • LLMs de-identified complete letters in seconds, significantly faster than human annotation.

    Conclusions:

    • LLMs demonstrate high-precision, GDPR-compliant de-identification capabilities for medical records.
    • Quantized LLMs enable efficient, local inference without requiring GPUs.
    • While effective, false negatives for sensitive PHI like names warrant further attention.