Presentation (CTI)

OBJECTIVES: Oncology HTA increasingly requires complex, interconnected evidence packages spanning systematic literature reviews (SLRs), clinical study reports (CSRs), real-world evidence (RWE), and economic modelling. Growing evidence volumes and frequent updates created substantial operational bottlenecks. AI tools have been proposed to improve efficiency; however, concerns remain regarding validity, transparency, regulatory acceptance, and applicability to Living-HTA processes. We reviewed current AI applications across the oncology HTA evidence pathway, emerging guidance, and assessed improved efficiency from existing AI implementations for Living-HTA.
METHODS: A scoping review was conducted of peer-reviewed publications January 2023-December 2025, and HTA and regulatory guidance/methodological frameworks on AI-assisted evidence generation. We searched biomedical literature databases using terms on AI, machine learning, large-language models (LLMs), oncology, HTA, SLR, RWE, and economic modelling. Institutional websites and reference lists were reviewed. Selection focused on applications or guidance relevant to oncology HTA.
RESULTS: AI applications have demonstrated the ability to reduce manual effort in SLR citation screening ~30-70% and data extraction ~50%, while maintaining sensitivity >95% under appropriate human oversight. Reported applications extend to CSR summarization (40-60% time reduction), as well as RWE cohort identification/phenotyping, causal inference analyses, and extraction of parameters for economic models. Despite these gains, most implementations remain fragmented, static, and project-based. Few support continuously-updated evidence synthesis or materially shorten end-to-end Living-HTA timelines, as outputs typically require re-execution when scope, comparators, or jurisdictional requirements change. Across reviewed guidance, HTA bodies consistently emphasize transparency, validation, bias assessment, and human-in-the-loop governance as prerequisites for acceptability.
CONCLUSIONS: While AI-assisted workflows show promise across components of oncology HTA, current implementations have rarely translated into sustained efficiency gains for Living-HTA. Achieving this objective will require approaches that move beyond task-level automation toward integrated, governed systems capable of continuous evidence maintenance. Clear implementation frameworks and governance models are needed before AI can reliably support living-HTA.