Session (CTI)

Purpose The growing adoption of Generative AI (GenAI) presents new opportunities and challenges for evidence generation in HEOR. As these tools evolve, stakeholders including HTA agencies, regulatory bodies, and internal teams increasingly require confidence that AI-generated outputs meet standards of transparency, traceability, and reliability. This workshop will focus on RAG as a practical framework to meet these expectations in real-world HEOR workflows. Presenters will introduce key components of the RAG pipeline including chunking, embeddings, and retrieval using semantic search and explain how these design approaches may impact the accuracy, reproducibility, and transparency of GenAI outputs. Using real-world examples, the session will demonstrate how RAG-based systems can be applied to common HEOR tasks such as literature reviews, evidence synthesis, and dossier development. The workshop will also explore how complex data formats including tables, graphs, and mixed media require tailored preprocessing to be integrated effectively into GenAI workflows. Although technical in nature, this session is designed for a wide HEOR audience. Participants will gain essential insight into how specific technical decisions shape the trustworthiness of GenAI outputs, equipping them to make informed judgments when evaluating or implementing GenAI tools. Description (10 min) Bill Malcolm will first introduce Generative AI with RAG in HEOR and the importance of transparency, traceability and reliability. (15 min) Rajdeep Kaur will discuss ongoing developments in RAG relevant to HEOR applications and will explain how to build a strong RAG pipeline for complex HEOR data. (15 min) Sven L Klijn will walk through key steps of RAG pipeline to explain how the approach supports transparent and traceable evidence generation. (10 min) Pall Jonsson will share on how RAG based approaches align with GenAI guidance from NICE, CDA, and ELEVATE-AI for HEOR and HTA use. (10 min) Attendees will have the opportunity to engage with the panellists and ask questions.