Barriers Delaying the Implementation of Artificial Intelligence-Based Literature Reviews (AILRs) by Health Technology Assessment (HTA) Bodies
Author(s)
Mangat G1, Sharma S2, Bergemann R3
1Parexel International, Mohali, PB, India, 2Parexel International, Chandigarh, India, 3Parexel International, Basel, Switzerland
Presentation Documents
OBJECTIVES: AILRs refer to literature reviews undertaken with AI-based tools for multiple review steps. Despite its significant potential, the uptake by HTA bodies has been slow, indicating the existence of barriers warranting further investigation.
METHODS: We conducted a rapid review to evaluate published studies, reviews, viewpoint papers, and presentations and identified multiple barriers. Based on our collective review experience, we shortlisted the most pertinent challenges (n=7) for HTA bodies in accepting AILRs.
RESULTS: The most significant barrier is the lack of availability of fit-for-purpose AI tools that suit the daily operational requirements of HTA bodies instead of HTA or tool developers. This results in isolated efforts by HTA bodies (e.g., developing internal RCT classifiers with minimal functionalities), further extending the gap between all stakeholders. Secondly, the over-reliance on AI data metrics alone, without consideration of its black-box phenomenon, is another significant barrier; none of the existing AI tools provide a qualitative rationale for excluding studies. Thirdly, the selection debate between indication-specific training of AI datasets vs. generative AI training is another concern, with some HTA bodies favoring the former approach. There is a lack of data on the impact of AI models using poor input/non-controlled data from the real world. The fourth and fifth administrative barriers relate to the lack of organizational (financial/political) support for HTA bodies and the potential legal implications if AI goes wrong (e.g., who owns the responsibility?). The sixth and seventh barriers include the potential loss of human intervention and decline in interpersonal communication (e.g., clinicians-patients-HTAs), impacting the determination of the clinically added value for the patient.
CONCLUSIONS: HTA bodies face an unprecedented burden to adopt AI. Strong collaboration between all relevant stakeholders—HTA developers, tool developers, and patients—should be encouraged, with HTA bodies mediating a central role instead of acting independently.
Conference/Value in Health Info
Value in Health, Volume 27, Issue 12, S2 (December 2024)
Code
HTA187
Topic
Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics, Decision & Deliberative Processes, Systems & Structure, Value Frameworks & Dossier Format
Disease
No Additional Disease & Conditions/Specialized Treatment Areas