Session (CTI)

Separate registration required.

Generative AI (GenAI) is rapidly transforming how HEOR and market access work is conducted, from literature reviews and evidence synthesis to dossier development and HTA submissions. As the field moves beyond experimentation, a new challenge emerges for subject matter experts: how to design and build AI tools that are rigorous enough to withstand scrutiny in regulatory, HTA, and payer engagement environments.

This intermediate-level course is designed for health economists, outcomes research professionals, market access specialists, and other HEOR subject matter experts who want to move beyond general-purpose chatbots and one-off pilots, and instead learn how to architect robust, purpose-built AI solutions. The emphasis throughout is on design: participants will learn how architectural choices determine whether an AI tool produces outputs that are reliable, traceable, and defensible.

The course is organized around three core themes, each illustrated with two applied examples that run throughout the course. These are complemented by ad hoc examples that showcase the diversity of possible architectures and applications across HEOR and market access.

Context Engineering
Large language models are only as good as the information they are given. Participants will learn how to design context: how external knowledge (eg, clinical data, published evidence, HTA guidance) is retrieved and incorporated into GenAI workflows. Retrieval-Augmented Generation (RAG) is treated as a cornerstone architecture, alongside complementary techniques such as tool use, that together determine factual accuracy, traceability, and domain fit.

Agentic AI
Participants will take an in-depth look at how autonomous and semi-autonomous AI agents can coordinate multi-step HEOR processes, such as structured data extraction and drafting workflows, while maintaining control, monitoring, and accountability. Faculty will discuss how to set boundaries for agents, orchestrate tasks, and design for human oversight.

Rigor for HTA
Building AI tools whose outputs will be scrutinized by HTA bodies, regulators, and payers demands a different standard than rapid prototyping. Faculty will address what separates "vibe-coded" solutions from HTA-ready tools and demonstrate how to evaluate and validate GenAI systems in terms of reliability, reproducibility, and regulatory alignment, drawing on frameworks such as ELEVATE-GenAI and guidance from NICE and the FDA. Ethical considerations around the application of AI are discussed in the same context. Participants will also learn how to set up a professional working environment, including effective use of an IDE, AI-assisted development, and reusable components such as skills.

The course provides full, runnable code for a worked example, discussed from a design and architecture perspective and highlighting a small number of key functions. Participants can run and adapt this example themselves after the course.

By the end of this course, participants will understand how to design AI architectures that live up to the evidentiary standards of HEOR and market access. They will leave with concrete design patterns, runnable reference implementations, and validation frameworks to build GenAI tools whose outputs can withstand scrutiny in HTA, regulatory, and payer settings. A basic understanding of Python or other similar scripting languages is recommended to get the most benefit from the provided worked examples.

PREREQUISITES: Attendance at “Applied Generative AI for HEOR: Introduction” or familiarity with concepts such as prompt engineering, APIs, and LLM workflows are necessary.