Large Language Models in Medicine

Jinwook Choi

doi:10.4258/hir.2025.31.2.111

The use of large language models (LLMs) has become widespread and is actively expanding across numerous fields. The medical field is no exception, as various types of LLMs have been introduced, demonstrating impressive results. Current research ranges from models capable of passing the United States Medical Licensing Examination to systems designed for summarizing complex medical documents.

To keep pace with these rapid developments, the Korean Society of Medical Informatics (KOSMI) has organized a special issue titled "LLMs in Medicine" to disseminate information regarding the use of LLMs in healthcare and foster the development of innovative and impactful research ideas.

For this special issue, we are honored to have the following guest editors: Namkug Kim (University of Ulsan, Korea), Edward Choi (KAIST, Korea), and Dukyong Yoon (Yonsei University, Korea). Their active engagement and dedicated efforts were instrumental in bringing this special issue to fruition. Without their valuable insights and hard work, the preparation of this issue would have been challenging. We extend our deepest gratitude to them.

1. Special Issue Topics

The following topics were selected for the special issue:

LLMs in clinical decision support systems: This topic explores how large language models assist healthcare professionals in real-time decision-making by utilizing patient data and medical knowledge.
Predictive modeling for disease diagnosis: This area covers the application of LLMs in developing models for early disease detection, risk stratification, and outcome prediction based on clinical data.
LLMs for medical text generation, summarization, and coding: This topic focuses on automating clinical documentation, text summarization, and medical coding, aiming to reduce workload and enhance data quality.
LLMs for patient interaction and data interpretation: This section examines the use of LLMs in interpreting patient-generated data, such as wearables and online inputs, and facilitating chatbot-based patient communication.
Natural language processing (NLP) in electronic health records: This topic describes NLP techniques powered by LLMs designed to extract and structure information from unstructured electronic health records.
Deploying LLMs in healthcare organizations: This area discusses the technical and operational challenges, including infrastructure, integration, and constraints encountered during the real-world deployment of LLMs.
Safety and reliability of LLMs in medical contexts: This topic assesses the trustworthiness, accuracy, and robustness of LLMs within clinical settings, particularly under conditions of uncertainty.
Ethical implications of artificial intelligence (AI) in healthcare: This area explores critical issues such as bias, privacy, transparency, and accountability when applying LLMs and AI within medical practice.
Adverse event detection and pharmacovigilance: This section covers the use of LLMs to identify, classify, and monitor drug-related adverse events within clinical datasets.
Training and fine-tuning of LLMs on medical data: This topic outlines approaches for adapting general-purpose LLMs for healthcare applications through fine-tuning, the use of domain-specific data, and prompt engineering.

2. Selected Papers

A total of 23 papers were submitted for this special issue, with ten selected for publication. These contributions cover a broad spectrum of topics, exploring various applications, challenges, and innovations related to the use of LLMs and AI in healthcare.

1) Large language models in medicine—clinical applications, technical challenges, and ethical considerations

This review explores the diverse applications of LLMs in healthcare, analyzing real-world use cases and offering insights into the future of their adoption in clinical practice.

2) Developing an explainable AI system for mobile-based diagnosis of febrile diseases

This paper proposes a mobile-based explainable AI (XAI) platform for diagnosing febrile illnesses. It integrates the LIME (local interpretable model-agnostic explanations) technique with GPT's explanatory capabilities to provide interpretable and trustworthy clinical decisions.

3) LLM-based response generation for Korean adolescents using a RAG model

This study presents a retrieval-augmented generation (RAG) system designed to deliver personalized and culturally relevant responses to questions from Korean adolescents. It also compares the response quality of RAG-based models with non-RAG alternatives, validating the system’s effectiveness.

4) Consensus on the potential of LLMs in healthcare: insights from a Korean delphi survey

This paper explores expert perceptions of the safety and risks associated with LLM usage in healthcare based on a systematic collection and analysis of opinions from Korean professionals. The findings provide insights into how to support the safe and effective deployment of LLMs in clinical settings.

5) Generative AI-based nursing diagnosis and documentation recommendation using virtual patient electronic nursing record data

As nursing documentation accounts for roughly 30% of nurses’ working hours, this study compares traditional documentation with a generative AI system. It evaluates the effectiveness of the AI system in reducing documentation time and ensuring the accuracy of nursing records using virtual patient data.

6) Automated pipeline for Korean medical preference dataset construction

This experimental paper proposes a methodology for creating Korean-language medical QA datasets. Research questions were generated from PubMed and matched with multiple answers, forming a structured dataset that supports LLM experimentation in Korean medical contexts.

7) Symptom and sentiment analysis of older people with cancer and caregivers: a text mining approach using Korean social media data

Although it does not use LLMs directly, this paper presents a text-mining approach for analyzing sentiments among older cancer patients and caregivers. Using latent Dirichlet allocation (LDA), the study extracts and evaluates key sentiment categories from blog data.

8) Generative pre-trained transformer: trends, applications, strengths and challenges in dentistry: a systematic review

This literature review investigates applications of LLMs in dentistry. From 704 screened papers, 16 were selected, revealing that most research centers on clinical decision-making tools and evaluations of LLM technologies.

9) Large language models for pre-mediation counseling in medical disputes: a comparative evaluation against human experts

Due to a lack of specialized medical knowledge, it is difficult for the general public to make a clear judgment about the possibility of medical malpractice. This paper presents the development and evaluation of a chain-of-thought based large language model chatbot for medical dispute counseling.

10) Multi-agent approach for sepsis management

This study introduces a multi-agent system integrating LLMs to manage early-stage sepsis. The agents include modules for guideline adherence, antibiotic recommendations, and decision support, demonstrating how LLMs enhance clinical workflows through collaborative intelligence.

This special issue not only offers valuable insights into future directions for LLM development in health informatics but also encourages reflection on important considerations and precautions for LLM use. Clearly, LLMs will increasingly integrate into all aspects of research activities, emphasizing the importance of wise collaboration and active knowledge sharing among researchers.