Challenges of Implementing LLMs in Clinical Practice: Perspectives.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Healthcare Technology

Background:

• Large language models (LLMs) show promise for revolutionizing healthcare tasks like documentation, diagnosis, and patient communication.
• Despite potential benefits, the widespread adoption of LLMs in clinical settings faces substantial obstacles.

Purpose of the Study:

• To explore and synthesize the key barriers hindering the integration of LLMs into clinical practice.
• To propose a framework for understanding and addressing the technical, ethical, and practical challenges of LLM deployment in healthcare.

Main Methods:

• A comprehensive review and synthesis of recent evidence.
• Analysis of implementation challenges across five critical domains: workflow/safety, bias/equity, governance, technical vulnerabilities, and patient trust.
• Development of a conceptual framework based on empirical studies.

Main Results:

• Identified five major challenge domains: workflow misalignment and diagnostic safety, bias and equity concerns, regulatory and legal governance gaps, technical vulnerabilities (e.g., hallucinations, data poisoning), and maintaining patient trust and human connection.
• Highlighted the rapid advancement of LLM capabilities and mitigation strategies, suggesting potential for near-term clinical impact.

Conclusions:

• Successful integration of LLMs in healthcare requires a multi-faceted approach addressing identified barriers.
• Future research, robust governance, and a focus on responsible deployment are essential for realizing the benefits of LLMs in clinical environments.