Session (CTI)

PURPOSE: Health economic models must be reproducible, auditable, and clinically credible. As agentic AI tools proliferate, modelers face a practical question: which parts of the modelling workflow benefit from AI assistance, and which are too consequential to delegate? Speakers will share perspectives from industry, academia, HTA, and vendor experience, addressing where human oversight remains non-negotiable and exploring the emerging potential of agentic AI in model development, validation, and “living HTA.” DESCRIPTION: This workshop gives attendees a cross-stakeholder view of agentic AI in health economic modelling. [Beth] will open the session by framing the core challenge: agentic AI is a powerful, general-purpose technology that can be directed at nearly any task, but not every task it can do should be delegated to it. (10 min.) [Tim] will demonstrate where AI provides clear value in the health economic modelling workflow: data extraction, incorporating “live” NMA results into decision models, model building and rapid scenario testing, and report generation. (15 min.) [Liza] will share practical lessons from developing agentic health economic modelling systems, making the case for deliberately excluding AI from model calculations, with computation instead delegated to open-source packages such as heemod and PyDICE. (10 min.) [Dan] will bring an HTA perspective, describing work done to date on model testing, validation, and replication as well as conversion (e.g., Excel to R), and will discuss the prospects of agentic AI for model development and support of "living HTA" activities. (10 min.) [Beth] will then facilitate audience interaction with probing questions (15 min.). This workshop is intended for health economists and HTA practitioners who want a practical perspective on where the technology earns trust—and where it does not.