Implementing Large Language Models in Health Care: Clinician-Focused Review With Interactive Guideline.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Natural Language Processing

Background:

• Large language models (LLMs) offer potential for clinical applications like medical question answering and report generation.
• The rapid advancement of LLMs presents challenges for clinicians in selecting suitable AI tools.
• Guidance is needed to facilitate the effective integration of LLMs into healthcare settings.

Purpose of the Study:

• To provide systematic guidance for clinicians in selecting appropriate LLMs for their specific needs.
• To facilitate the integration of LLMs into clinical workflows.
• To offer a practical reference for applying LLMs in healthcare settings.

Main Methods:

• Conducted a systematic literature search of clinical LLM applications from January 2022 to March 2025 across major databases (PubMed, ScienceDirect, Scopus, IEEE Xplore) and arXiv.
• Included 270 studies focusing on clinical applications of innovative multimodal LLMs.
• Collected data on 330 LLMs, their application frequency, and performance in clinical tasks.

Main Results:

• LLMs show utility across clinical tasks, particularly in stages 2, 3, and 4 of a 5-stage workflow.
• GPT-3.5 and GPT-4 demonstrated versatility, covering a significant percentage of clinical subtasks.
• General-purpose LLMs often require fine-tuning for specialized clinical areas; multimodal LLMs frequently lack transparency and pose data privacy concerns.

Conclusions:

• LLMs can assist clinicians, but a lack of generalist models applicable across diverse clinical tasks poses deployment challenges.
• A proposed interactive online guideline aims to help clinicians select suitable LLMs based on specific clinical tasks.
• The guideline is designed for clinical use, avoiding technical jargon to ensure accessibility and successful LLM application.