INFREQUENT AND LIMITED TO LITERATURE REVIEWS: AI USE IN NICE HTA SUBMISSIONS IN ONCOLOGY
Author(s)
Steven Y. Yap, PhD1, Vertika Awasthi, PharmD2, Juliette Torres Ames, MSc1, Martina Giulietti Smith, BSc1, Alyssa Simon, MPH3, Grace E. Fox, PhD3.
1OPEN Health HEOR & Market Access, OPEN Health, London, United Kingdom, 2OPEN Health HEOR & Market Access, OPEN Health, Bangalore, India, 3OPEN Health HEOR & Market Access, OPEN Health, New York, NY, USA.
1OPEN Health HEOR & Market Access, OPEN Health, London, United Kingdom, 2OPEN Health HEOR & Market Access, OPEN Health, Bangalore, India, 3OPEN Health HEOR & Market Access, OPEN Health, New York, NY, USA.
OBJECTIVES: To identify the extent, transparency and assessment of use of artificial intelligence (AI) in HTA submissions to NICE focusing on oncology.
METHODS: NICE HTA submissions in oncology published between January 2022 and December 2025 were identified using EVERSANA’s NAVLIN. Committee papers (CPs) associated with the identified submissions were retrieved using a script and screened for use of AI using an agentic ChatGPT-5 workflow based on predefined AI-related keywords. Human review verified outputs and checked for omissions. Flagged submissions underwent a second-stage review to confirm AI use and extract information on AI applications, disclosure practices, and EAG assessments.
RESULTS: Of 191 HTA submissions identified, 142 had CPs available for screening (241 CPs in total); 49 were flagged for potential AI use, but only 5/142 (3.5%) submissions were confirmed as using AI following human review. All 5 instances related to systematic literature review (SLR) activities; 4 occurred in single technology appraisals and 1 in a cost-comparison appraisal. Beyond the 5 confirmed cases, 1 CP explicitly disclosed that AI had not been used. Among the 5 confirmed cases, 2 described the use of AI or natural language processing methods. The remaining 3 reported the use of AI-enabled SLR platforms. In all 5 submissions, details on AI use were limited; disclosures generally identified tools or methods but provided little information on how AI functionalities were applied. Where AI use was disclosed, EAG comments were favourable, describing the methods as robust and appropriate.
CONCLUSIONS: AI use in NICE submissions in oncology was infrequently reported and limited to SLR activities, but where used EAG assessments were favourable. Where AI-enabled tools were reported, disclosures often lacked detail on how AI functionalities were used, even after the NICE AI guidance became available. Human oversight was necessary to correctly identify the submissions reflecting the challenges in distinguishing AI use from AI-related terminology.
METHODS: NICE HTA submissions in oncology published between January 2022 and December 2025 were identified using EVERSANA’s NAVLIN. Committee papers (CPs) associated with the identified submissions were retrieved using a script and screened for use of AI using an agentic ChatGPT-5 workflow based on predefined AI-related keywords. Human review verified outputs and checked for omissions. Flagged submissions underwent a second-stage review to confirm AI use and extract information on AI applications, disclosure practices, and EAG assessments.
RESULTS: Of 191 HTA submissions identified, 142 had CPs available for screening (241 CPs in total); 49 were flagged for potential AI use, but only 5/142 (3.5%) submissions were confirmed as using AI following human review. All 5 instances related to systematic literature review (SLR) activities; 4 occurred in single technology appraisals and 1 in a cost-comparison appraisal. Beyond the 5 confirmed cases, 1 CP explicitly disclosed that AI had not been used. Among the 5 confirmed cases, 2 described the use of AI or natural language processing methods. The remaining 3 reported the use of AI-enabled SLR platforms. In all 5 submissions, details on AI use were limited; disclosures generally identified tools or methods but provided little information on how AI functionalities were applied. Where AI use was disclosed, EAG comments were favourable, describing the methods as robust and appropriate.
CONCLUSIONS: AI use in NICE submissions in oncology was infrequently reported and limited to SLR activities, but where used EAG assessments were favourable. Where AI-enabled tools were reported, disclosures often lacked detail on how AI functionalities were used, even after the NICE AI guidance became available. Human oversight was necessary to correctly identify the submissions reflecting the challenges in distinguishing AI use from AI-related terminology.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA108
Topic
Health Technology Assessment
Topic Subcategory
Systems & Structure
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Oncology