THE FUTURE OF AI-ENABLED HEALTH TECHNOLOGY ASSESSMENT
Author(s)
Michael J. Merchant, PhD1, Ahmed Yosef, MSc2, Stephen Duffield, PhD, MD3, Raphael Sonabend-Friend, PhD4, Pall Jonsson, BSc, PhD5.
1National Institute for Health and Care Excellence, Manchester, United Kingdom, 2National Institute for Health and Care Excellence, London, United Kingdom, 3NICE, Liverpool, United Kingdom, 4NICE, London, United Kingdom, 5National Institute for Health and Care Excellence (NICE), Manchester, United Kingdom.
1National Institute for Health and Care Excellence, Manchester, United Kingdom, 2National Institute for Health and Care Excellence, London, United Kingdom, 3NICE, Liverpool, United Kingdom, 4NICE, London, United Kingdom, 5National Institute for Health and Care Excellence (NICE), Manchester, United Kingdom.
OBJECTIVES: The National Institute for Health and Care Excellence (NICE) undertook an exploratory project to examine the potential role of artificial intelligence (AI) in health technology assessment (HTA) over a five year horizon.
METHODS: The project synthesised evidence from stakeholder workshops, a landscape review on AI governance in HTA, and empirical studies of AI performance in systematic literature review. It focused on three cross cutting themes: automation, personalisation, and reasoning. Automation refers to the application of AI to structured tasks, including literature screening, data extraction, and economic model development. Personalisation focuses on the use of AI to integrate data sources and causal approaches to individualise treatment effects beyond population averages. Reasoning‑oriented AI consists of systems capable of structured analysis that may support human judgement.
RESULTS: Stakeholders highlighted automation as a source of efficiency gains, supporting a staged adoption approach focused on low‑risk, high‑impact tasks with human oversight. Agentic AI systems were considered as an extension of this automation, with potential to coordinate multiple AI components across end‑to‑end evidence synthesis workflows. Dynamic, continuously updated approaches to evidence workflows, including pathways toward living HTA, were viewed as valuable but raising substantial governance and control challenges.Causal AI methods for exploration of treatment effects in subpopulations were considered clinically relevant but constrained by limitations in data quality and interpretability.Reasoning‑oriented AI was seen as promising for dossier preparation, including examining evidence using multiple AI personas to surface clinical and patient perspectives, but was viewed cautiously as a support for human judgement in evidence assessment and not yet ready for routine operational use.
CONCLUSIONS: The project identified opportunities and barriers to the adoption of AI across a range of HTA applications. The outputs have led to a second phase focused on agentic AI, and have also influenced the NICE AI Methods for Evidence framework
METHODS: The project synthesised evidence from stakeholder workshops, a landscape review on AI governance in HTA, and empirical studies of AI performance in systematic literature review. It focused on three cross cutting themes: automation, personalisation, and reasoning. Automation refers to the application of AI to structured tasks, including literature screening, data extraction, and economic model development. Personalisation focuses on the use of AI to integrate data sources and causal approaches to individualise treatment effects beyond population averages. Reasoning‑oriented AI consists of systems capable of structured analysis that may support human judgement.
RESULTS: Stakeholders highlighted automation as a source of efficiency gains, supporting a staged adoption approach focused on low‑risk, high‑impact tasks with human oversight. Agentic AI systems were considered as an extension of this automation, with potential to coordinate multiple AI components across end‑to‑end evidence synthesis workflows. Dynamic, continuously updated approaches to evidence workflows, including pathways toward living HTA, were viewed as valuable but raising substantial governance and control challenges.Causal AI methods for exploration of treatment effects in subpopulations were considered clinically relevant but constrained by limitations in data quality and interpretability.Reasoning‑oriented AI was seen as promising for dossier preparation, including examining evidence using multiple AI personas to surface clinical and patient perspectives, but was viewed cautiously as a support for human judgement in evidence assessment and not yet ready for routine operational use.
CONCLUSIONS: The project identified opportunities and barriers to the adoption of AI across a range of HTA applications. The outputs have led to a second phase focused on agentic AI, and have also influenced the NICE AI Methods for Evidence framework
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA66
Topic
Economic Evaluation, Health Technology Assessment, Methodological & Statistical Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas