From Data to Decisions: Decision-Makers' Perceptions of Artificial Intelligence-Powered Evidence Synthesis in HTA Submissions

OBJECTIVES: Evidence synthesis is a cornerstone of any health technology assessment (HTA), with systematic literature reviews (SLRs) serving as an unbiased, reproducible process for identifying, collecting, and synthesizing results from multiple sources. Since SLRs are highly labor-intensive, exploring avenues to integrate artificial intelligence (AI) with traditional methodologies is of interest. However, it is unclear whether AI-supported SLRs will be accepted when used for HTA submissions. The objective of this study was to engage with various HTA bodies to understand their perceptions regarding the use of AI-supported SLRs for HTA.

METHODS: We conducted a targeted literature review (TLR) in Embase and MEDLINE (January 2019 - May 2024) using terms for AI, natural language processing, and machine learning, combined with terms for HTA. Supplemental searches were conducted to identify policy documents and guidelines from HTA websites. The results of the TLR informed the development of a survey aimed to gather insights on the perceptions of HTA bodies on the use of AI for SLRs. This survey was shared with global HTA decision-makers.

RESULTS: The TLR found that, while HTA bodies express an interest in understanding the applications of AI in SLRs, many are currently unfamiliar with these methodologies. The majority do not mention the utilization or endorsement of AI. Early findings from the survey indicate that global HTA decision-makers are confident in the potential for AI in SLRs, and consider transparency, data privacy, and validated methodology to be key to its implementation. When asked specifically about responsibility, respondents believed that in 2-3 years, regulatory agencies should be responsible for the certification of AI tools.

CONCLUSIONS: Our findings suggest that AI-supported SLRs are likely to become an integral part of the HTA process. A collaborative approach across stakeholders will be required to address challenges and create best practices, ensuring transparency, methodological rigor, and reproducibility.