Presentation (CTI)

OBJECTIVES: Health technology assessment (HTA) workflows from evidence synthesis to economic modeling and dossier generation are often fragmented across multiple workflows and software environments. The objective of this research was to develop and validate a comprehensive generative AI (GenAI) suite capable of producing end-to-end, HTA-ready outputs within a unified and transparent framework, while maintaining human oversight and methodological rigor.
METHODS: The suite integrates five interconnected modules: (1) Evidence Synthesis, enabling landscape assessment and systematic literature reviews for generating standardized extraction grids; (2) Model Conceptualization, producing model conceptualization protocols with model structure, inputs, and assumptions; (3) Model Development, generating fully functional HTA-ready Excel or R/Python models for decision tree, Markov, semi-Markov and hybrid frameworks; (4) Report Generation, drafting technical reports including sensitivity analyses; and (5) Dossier Generation, for market-specific submissions. The system was tested using diverse therapeutic areas, including oncology, rare diseases, and mental health indications. Validation metrics included traceability, reproducibility, structural accuracy, and narrative coherence.
RESULTS: Across six test cases spanning three therapeutic areas, the suite achieved 80-85% full-process accuracy (from evidence synthesis to dossier generation) and 92-95% partial-process accuracy when individual modules were used independently. Automation reduced approximately 50%-85% development time compared with manual workflows across multiple modules (50% in evidence synthesis, 90% in model conceptualization, 75% in model development, 85% in report writing, and 90% in dossier generation). Domain experts verified all AI-generated excel/R models for face validity, structural integrity, and compliance with established HTA methodological standards.
CONCLUSIONS: This GenAI suite demonstrates that comprehensive automation of HTA workflows with human oversight is achievable while preserving methodological rigor and transparency. The substantial time savings observed across all modules suggest meaningful potential for accelerating HTA submission timelines while maintaining quality standards. Future research will focus on evaluating performance across additional use cases and real-world HTA submissions to further assess robustness, generalizability, and practical applicability.