Presentation (CTI)

OBJECTIVES: Developing an HTA strategy for a new therapeutic asset requires systematic review of prior agency submissions to map accepted cost-effectiveness modeling approaches, identify methodological drivers, and anticipate regulatory critique patterns. This review is conventionally conducted manually by health economists, requiring substantial time to process large volumes of complex documentation while often yielding incomplete cross-agency integration. We assessed the performance of a generative AI (GenAI)-assisted workflow for structured review of HTA cost-effectiveness evidence.
METHODS: The framework was applied to nine HTA agency documents (>1,200 pages) from three jurisdictions in an ophthalmic indication as proof of concept. Using a predefined extraction grid aligned with the Drummond checklist, the GenAI systematically extracted model structure, health state definitions, comparator sets, utility and cost inputs, sensitivity analyses, and cost-effectiveness outcomes. Cross-agency synthesis mapped convergence and divergence in modeling assumptions. Outputs were benchmarked against manual expert review.
RESULTS: The workflow processed all nine documents and generated a structured cross-agency synthesis. Review time was reduced by 75% (8 expert-days versus 2 days). Extraction accuracy reached 96% across 42 predefined data points (30% random audit); residual discrepancies (4%) reflected only interpretive divergences in attributing shared versus treatment-specific assumptions. Beyond structured extraction, the GenAI surfaced emergent cross-agency patterns not explicitly targeted, including undocumented divergence on cost-effectiveness acceptance criteria, demonstrating the capacity to generate strategic intelligence beyond predefined parameters. Output quality was rated 4.5/5.0 by a senior health economist.
CONCLUSIONS: GenAI-assisted review of HTA cost-effectiveness evidence demonstrated substantial efficiency gains while preserving analytical rigor comparable to conventional expert-led review. The framework enables rapid, reproducible synthesis of agency expectations and modeling precedent, supporting evidence-informed strategy for new assets. This proof-of-concept suggests broad applicability across therapeutic areas.