FRAMING AI-ENABLED EVIDENCE FOR HTA: A PRACTICAL TAXONOMY BY EVIDENCE DOMAINS AND LEVELS OF AI INVOLVEMENT
Author(s)
Tristan Gonzalez, MPharm, MSc1, Shavinder Girn, MPharm2, Fabio Kistner, BSc3, Anna Marimón, MSc4, Bengt Liljas, PhD5, Ilse van Oostrum, MA, MSc, PhD6, Melanie Koehle, MSc7, Graham Tatham, .2.
1Director, AstraZeneca AG, Baar, Switzerland, 2Simon Kucher & Partners, London, United Kingdom, 3Simon Kucher & Partners, Munich, Germany, 4AstraZeneca, Barcelona, Spain, 5AstraZeneca, Gaithersburg, MD, USA, 6AstraZeneca, Den Haag, Netherlands, 7AstraZeneca, Baar, Switzerland.
1Director, AstraZeneca AG, Baar, Switzerland, 2Simon Kucher & Partners, London, United Kingdom, 3Simon Kucher & Partners, Munich, Germany, 4AstraZeneca, Barcelona, Spain, 5AstraZeneca, Gaithersburg, MD, USA, 6AstraZeneca, Den Haag, Netherlands, 7AstraZeneca, Baar, Switzerland.
OBJECTIVES: AI-enabled evidence refers to evidence and materials used in HTA/payer decision-making where AI is applied to search, extract, analyze, synthesize, or generate evidentiary content. Currently, no standard taxonomy exists for AI use in HTA. As use cases expand rapidly and stakeholder acceptance remains uncertain, a practical framework is needed to standardize terminology and align expectations for validation, transparency, and acceptability. The objective of this project was to design an HTA-oriented taxonomy that classifies AI-enabled evidence by evidence domain and level of AI involvement to structure HTA discussions of use cases, safeguards, and acceptance.
METHODS: A taxonomy was developed through focused landscape mapping (payer and regulator publications, relevant industry pilots and policy initiatives), covering two dimensions: HTA-aligned evidence domains and level of AI involvement through a combination of desk research and stakeholder interviews to capture perspectives beyond published positions. It was iteratively refined for clarity, reproducibility, and applicability. This framework was validated via six internal workshops and expert interviews to capture insights beyond published positions (with input across France, Germany, Italy and the UK, and two with pan‑European thought leaders).
RESULTS: First dimension: Evidence domains define where AI-enabled evidence is applied across HTA evidence generation and assessment workflows, and comprise four categories: submission outputs, economic evidence, real-world evidence, and clinical & comparative evidence. Second dimension: Levels of AI involvement define the degree to which AI impacts evidence generation, analytical processes, and decision-grade conclusions, and are divided into three levels: Level 1 (Support); Level 2 (Inform) - and Level 3 (Determine)This framework provides clear and widely understood definitions, evidence domains reflecting HTA evidence generation workflows and AI involvement levels reflecting perceived evidentiary risk.
CONCLUSIONS: This taxonomy provides a clear framework that enables structured discussion of use cases, safeguards, and acceptability across evidence domains and levels of AI influence.
METHODS: A taxonomy was developed through focused landscape mapping (payer and regulator publications, relevant industry pilots and policy initiatives), covering two dimensions: HTA-aligned evidence domains and level of AI involvement through a combination of desk research and stakeholder interviews to capture perspectives beyond published positions. It was iteratively refined for clarity, reproducibility, and applicability. This framework was validated via six internal workshops and expert interviews to capture insights beyond published positions (with input across France, Germany, Italy and the UK, and two with pan‑European thought leaders).
RESULTS: First dimension: Evidence domains define where AI-enabled evidence is applied across HTA evidence generation and assessment workflows, and comprise four categories: submission outputs, economic evidence, real-world evidence, and clinical & comparative evidence. Second dimension: Levels of AI involvement define the degree to which AI impacts evidence generation, analytical processes, and decision-grade conclusions, and are divided into three levels: Level 1 (Support); Level 2 (Inform) - and Level 3 (Determine)This framework provides clear and widely understood definitions, evidence domains reflecting HTA evidence generation workflows and AI involvement levels reflecting perceived evidentiary risk.
CONCLUSIONS: This taxonomy provides a clear framework that enables structured discussion of use cases, safeguards, and acceptability across evidence domains and levels of AI influence.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA409
Topic
Health Policy & Regulatory, Health Technology Assessment, Organizational Practices
Topic Subcategory
Value Frameworks & Dossier Format
Disease
No Additional Disease & Conditions/Specialized Treatment Areas