Preliminary Results from Using Gen-AI to Personalized Medication Leaflets.

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine

Background:

• Traditional patient information leaflets (PILs) are static and often difficult to understand.
• The European Medicines Agency (EMA) is transitioning to electronic product information (ePI) using HL7 FHIR® standards.
• Personalized medicine requires tailored patient information for improved comprehension and adherence.

Purpose of the Study:

• To explore the potential of Large Language Models (LLMs) for generating personalized PIL summaries.
• To assess the feasibility of using International Patient Summary (IPS) data for tailoring medication information.
• To enhance patient understanding and adherence through simplified, individualized PIL content.

Main Methods:

• Experimentation with three open-source LLMs to generate PIL summaries.
• Utilizing prompts with and without personalization, guided by IPS data.
• Evaluating the coherence, accuracy, and personalization of LLM-generated summaries.

Main Results:

• LLMs demonstrated the ability to generate coherent summaries of PILs.
• Current LLM outputs require improvement in accuracy and personalization.
• Personalized PILs show potential for enhancing patient-physician communication and shared decision-making.

Conclusions:

• LLMs offer a promising avenue for creating personalized patient information.
• Further development in prompt engineering and model fine-tuning is crucial for clinical application.
• Personalized PILs can support shared decision-making and improve patient engagement with their treatment.