Redefining digital health interfaces with large language models.

Area of Science:

• Digital Health
• Artificial Intelligence in Medicine
• Clinical Informatics

Background:

• Digital health tools offer potential healthcare improvements but face adoption barriers due to usability and trust issues.
• Large Language Models (LLMs) demonstrate advanced text processing capabilities, suggesting broad healthcare applications.
• Direct clinical use of LLMs is challenging due to potential inconsistencies and nonsensical outputs (hallucinations).

Purpose of the Study:

• To demonstrate LLM-based systems utilizing external tools as a novel interface for digital health.
• To enhance the utility and practical impact of digital healthcare tools and AI models.
• To address current limitations of LLMs in clinical settings, such as hallucinations.

Main Methods:

• LLM-based systems designed as agents to interact with external tools.
• Development of a novel interface connecting clinicians with digital technologies via LLMs.
• Quantitative assessment of LLM-based interfaces using cardiovascular disease and stroke risk prediction examples.

Main Results:

• LLM-based systems effectively utilized external tools, creating a functional interface.
• Performance was quantitatively assessed in cardiovascular disease and stroke risk prediction.
• The novel interface demonstrated benefits compared to traditional digital tool interfaces.

Conclusions:

• LLM-based agent systems offer a viable solution for integrating advanced AI into clinical workflows.
• These systems improve the usability and trustworthiness of digital health tools.
• The approach mitigates LLM-specific challenges like hallucinations, paving the way for wider AI adoption in healthcare.