Presentation (CTI)

OBJECTIVES: AI/ML methods are increasingly being incorporated into HTA submissions both as health interventions (e.g. clinical decision support) and for evidence generation. However, there are concerns about using these methods due to challenges such as transparency and explainability. HTA bodies have started to issue guidelines and best practices regarding AI/ML use for evidence generation, but there is still a lack of clear guidance around its acceptability. The goal was to understand how widespread the use of AI/ML in HTA submissions is and how these submissions that include AI/ML methods are received by HTA bodies.
METHODS: Research was conducted using IQVIA’s Market Access & Insights platform and the IQVIA Linguamatics platform, a natural language processing (NLP) search engine, utilising a comprehensive set of AI/ML related keywords (e.g. LASSO, Random Forest, NLP), to systematically identify HTA reports that mentioned AI/ML methods from 2016 to 2024 from the UK, USA, Australia, and Canada.
RESULTS: In total, 6,097 HTA records were identified within the defined time period. Of these, 14 records used AI/ML methods in their submissions. Ten of the AI/ML applications in HTA submissions were from the National Institute for Health and Care Excellence, while the remaining 4 were from Canada's Drug Agency. Key applications of AI/ML use in HTA submissions included the use of AI-driven medical devices to support disease diagnosis and management, patient classification, feature selection for enhancing statistical model robustness and NLP models to support systematic literature reviews. Although 4 out of 14 records identified received a negative recommendation, none mentioned the use of AI/ML as the reason for the recommendation.
CONCLUSIONS: Concerns raised by the agencies on the use of AI/ML methods primarily focused on the need for ML model validation and a well-justified rationale for choosing AI/ML methods over traditional statistical approaches.