USE OF ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING (AI/ML) IN HEALTH TECHNOLOGY ASSESSMENT (HTA) EVIDENCE GENERATION: A LANDSCAPE REVIEW OF HTA GUIDANCE AND EMERGING STANDARDS
Author(s)
Eirini Poimenidou, MSc1, Anagha R. Nath, MSc2, Anna Maria Mavrigiannaki, MSc1, Natalie Blackburn, PhD3, Georgios Kantidakis, PhD1, Julia Oddsdottir, MSc4, Theodora Oikonomidi, PhD1, Sheily Kamra, MSc5, Yogesh Punekar, MSc6.
1IQVIA, Athens, Greece, 2IQVIA, Bengaluru, India, 3IQVIA, Porto Salvo, Portugal, 4IQVIA, London, United Kingdom, 5IQVIA, Gurugram, India, 6IQVIA, Pinner, United Kingdom.
1IQVIA, Athens, Greece, 2IQVIA, Bengaluru, India, 3IQVIA, Porto Salvo, Portugal, 4IQVIA, London, United Kingdom, 5IQVIA, Gurugram, India, 6IQVIA, Pinner, United Kingdom.
OBJECTIVES: The use of AI/ML in HTA evidence generation, including systematic literature reviews (SLRs), is accelerating, yet expectations from HTA bodies and methodological organisations remain fragmented. This review maps HTA guidance on AI/ML use, identifies key methodological developments, and highlights gaps relevant to sponsors and evidence synthesis teams.
METHODS: A landscape review of publicly available guidance, position statements, and frameworks was conducted across major HTA bodies (NICE, CDA-AMC, IQWiG, HAS, NCPE, SMC, TLV, G-BA, PBAC) and relevant methodological frameworks, including the Responsible AI use in evidence SynthEsis (RAISE) framework, developed by Cochrane, Campbell, JBI, and the Collaboration for Environmental Evidence.
RESULTS: The identified guidance addresses evidence generation broadly, with direct implications for SLRs. NICE (ECD11, August 2024) provides the most comprehensive operational AI/ML guidance, setting end-to-end expectations across the HTA evidence package: justification of use, validation, transparent reporting, human oversight, adherence to reporting standards (e.g., PRISMA, PALISADE), and early engagement with NICE (e.g., via NICE Advice). CDA-AMC (April 2025) aligns closely with NICE's position. IQWiG permits validated ML tools case-by-case, referencing the RAISE framework. HAS conducted exploratory testing, with promising yet preliminary results, but issued no submission requirements. No AI/ML-specific HTA submission guidance was identified for NCPE, SMC, TLV, G-BA, or PBAC. RAISE (2025-2026) classifies task-specific AI applications across a 5-tier acceptability scale, from acceptable (e.g., deduplication tools) to not acceptable (e.g., fully automated reviews, cross-study synthesis via large language models [LLMs]). Cochrane formally endorses RAISE (November 2025), requiring that AI does not compromise methodological rigour, is used with human oversight, and is transparently reported.
CONCLUSIONS: HTA AI/ML guidance is maturing, with NICE and RAISE providing the most actionable direction. However, the absence of task-specific minimum validation thresholds or standardised acceptance criteria across HTA bodies increases the documentation and justification burden for sponsors preparing multi-country submissions.
METHODS: A landscape review of publicly available guidance, position statements, and frameworks was conducted across major HTA bodies (NICE, CDA-AMC, IQWiG, HAS, NCPE, SMC, TLV, G-BA, PBAC) and relevant methodological frameworks, including the Responsible AI use in evidence SynthEsis (RAISE) framework, developed by Cochrane, Campbell, JBI, and the Collaboration for Environmental Evidence.
RESULTS: The identified guidance addresses evidence generation broadly, with direct implications for SLRs. NICE (ECD11, August 2024) provides the most comprehensive operational AI/ML guidance, setting end-to-end expectations across the HTA evidence package: justification of use, validation, transparent reporting, human oversight, adherence to reporting standards (e.g., PRISMA, PALISADE), and early engagement with NICE (e.g., via NICE Advice). CDA-AMC (April 2025) aligns closely with NICE's position. IQWiG permits validated ML tools case-by-case, referencing the RAISE framework. HAS conducted exploratory testing, with promising yet preliminary results, but issued no submission requirements. No AI/ML-specific HTA submission guidance was identified for NCPE, SMC, TLV, G-BA, or PBAC. RAISE (2025-2026) classifies task-specific AI applications across a 5-tier acceptability scale, from acceptable (e.g., deduplication tools) to not acceptable (e.g., fully automated reviews, cross-study synthesis via large language models [LLMs]). Cochrane formally endorses RAISE (November 2025), requiring that AI does not compromise methodological rigour, is used with human oversight, and is transparently reported.
CONCLUSIONS: HTA AI/ML guidance is maturing, with NICE and RAISE providing the most actionable direction. However, the absence of task-specific minimum validation thresholds or standardised acceptance criteria across HTA bodies increases the documentation and justification burden for sponsors preparing multi-country submissions.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA244
Topic
Health Technology Assessment
Topic Subcategory
Value Frameworks & Dossier Format
Disease
No Additional Disease & Conditions/Specialized Treatment Areas