Preliminary Results from Using Gen-AI to Personalized Medication Leaflets

The product information of a medicinal product includes the summary of product characteristics (SmPC), package label, and patient information leaflet (PIL), previously available only in paper or pdf format. The European Medicines Agency (EMA) in 2020 launched the electronic product information (ePI) project to make PILs available in HL7 FHIR®. The Vulcan/ Gravitate-Health ePI project created and tested a global HL7 FHIR Implementation Guide (IG). In parallel, Gravitate-Health conceptualized G-lens®, using information in the Electronic Health Record (EHR) and the International Patient Summary (IPS) to tailor leaflets to individual needs. This paper explores the potential of Large Language Models (LLMs) to generate concise personalized summaries of PILs, guided by data from IPS. The aim is to enhance patient understanding of and adherence to medication by providing digestible personalized summaries of PILs in simplified language to foster shared decision making and enhance patient communication with their health team. We experimented with three different open-source LLMs, employing prompts with and without personalization. Results indicate that LLMs can generate coherent summaries. However, accuracy and personalization need to improve, perhaps by engaging tailored PILs in patient-physician interaction and shared decision-making. Promising future research directions in personalization of PILs include prompt engineering, fine-tuning, and evaluation models for testing at scale.