Assessment and Integration of Large Language Models for Automated Electronic Health Record Documentation in Emergency Medical Services.

Journal: Journal of medical systems
PMID:

Abstract

Automating Electronic Health Records (EHR) documentation can significantly reduce the burden on care providers, particularly in emergency care settings where rapid and accurate record-keeping is crucial. A critical aspect of this automation involves using natural language processing (NLP) techniques to convert transcribed conversations into structured EHR fields. For instance, extracting temperature values like "102.4 Fahrenheit" from the transcribed text "His temperature is 39.1, which is 102.4 Fahrenheit." However, traditional rule-based and single-model NLP approaches often struggle with domain-specific medical terminology, contextual ambiguity, and numerical extraction errors. This study investigates the potential of integrating multiple Large Language Models (LLMs) to enhance EMS documentation accuracy. We developed an LLM integration framework and evaluated four state-of-the-art LLMs-Claude 3.5, GPT-4, Gemini, and Mistral-on a dataset comprising transcribed conversations from 40 EMS training simulations. The evaluation focused on precision, recall, and F1 score across zero-shot and few-shot learning scenarios. Results showed that the integrated LLM framework outperformed individual models, achieving overall F1 scores of 0.78 (zero-shot) and 0.81 (few-shot). In addition to quantitative evaluation, a preliminary user study was conducted with domain experts to assess the perceived usefulness and challenges of the integrated framework. The findings suggest that this approach has the potential to reduce documentation effort compared to traditional manual documentation. However, challenges such as misinterpretation of medical context and occasional omissions were noted, highlighting areas for further refinement and future work. This research is the first to systematically explore and evaluate the use of LLMs for real-time EMS EHR documentation. By addressing key challenges in automated transcription and structured data extraction, our work lays a foundation for real-world implementation, improving efficiency and accuracy in emergency medical documentation.

Authors

  • Enze Bai
    School of Computer Science and Information Systems, Pace University, New York City, NY, USA.
  • Xiao Luo
    Department of Spine Surgery, The Third Hospital of Mianyang, Sichuan Mental Health Center, Mianyang, China.
  • Zhan Zhang
    School of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001, China.
  • Kathleen Adelgais
    Department of Pediatrics, University of Colorado School of Medicine, Aurora, Colorado.
  • Humaira Ali
    Maimonides Medical Center, New York City, NY, USA.
  • Jack Finkelstein
    Interfaith Medical Center, New York City, NY, USA.
  • Jared Kutzin
    Mount Sinai Hospital, New York City, NY, USA.