Presentation (CTI)

OBJECTIVES: Technical reports accompanying cost-effectiveness models require both domain expertise and strict editorial compliance to support Health Technology Assessment (HTA) submissions. This study evaluated an AI-driven system for automated technical report generation from existing Excel-based cost-effectiveness models, benchmarking its outputs against manual development across two predefined validation domains: technical accuracy and editorial compliance.
METHODS: Two representative case studies, a partitioned survival model and a multi-state Markov model, were processed using an AI-based report generation system that extracts data directly from Excel models and associated documentation to draft complete technical reports. Performance was assessed across two domains. Domain validation evaluated consistency with health economics principles, accuracy of data extraction from Excel, correctness of result and figure interpretation, and the presence of unsupported or spurious technical statements. Editorial validation assessed HTA-style compliance, including automated generation of front matter (table of contents, lists of tables and figures), accuracy of captions and abbreviation legends, and adherence to predefined company-specific formatting templates. All outputs were independently reviewed by senior health economists and medical writers.
RESULTS: Across both case studies, the AI-generated reports accurately reflected model inputs and results from Excel, with no unsupported technical statements identified during expert review. Tables and figures were correctly structured from extracted data, and graphical outputs were appropriately interpreted. From an editorial perspective, the system generated required front matter (table of contents, lists of tables and figures), applied captions and abbreviation legends correctly, and adhered to predefined formatting templates suitable for HTA submissions. However, expert review identified residual issues related to verbosity and stylistic refinement in certain narrative sections. Compared with manual drafting, report development time was reduced from approximately 40 hours to under 8 hours.  
CONCLUSIONS: In two representative CE model case studies, AI-generated technical reports demonstrated strong domain accuracy and HTA-compliant editorial quality while substantially reducing authoring time.