Presentation (CTI)

OBJECTIVES: HEOR teams increasingly rely on separate GenAI-assisted modules for literature review, comparative analyses, economic modelling, and dossier compilation. Operating in isolation, however, they remain fragmented and fall out of sync, leaving dossiers built on outdated evidence. We aimed to connect them into one continuously updated, human-governed pipeline from research question to HTA readiness.
METHODS: We built a connecting layer using the Model Context Protocol (MCP), which lets HEOR modules automatically exchange structured evidence through a common data model. Each module runs reasoning loops to reach the task goal. Human governance is built in, with expert validation and auditable, rationale-backed decisions at each step, aligned with Cochrane RAISE guidance for responsible AI. The pipeline is designed to meet the evidence requirements of the EU HTA Regulation, JCA guidance, and national submission formats.
RESULTS: The framework links the modules as ordered, MCP-connected stages, each governed by an expert checkpoint: (i) landscape assessment frames the research question; (ii) a living SLR layer keeps evidence current via scheduled and event-based updates, with confidence-scored decisions; (iii) extracted evidence is consolidated into evidence tables, maps, the PICOS set, and generation plan driving downstream work; (iv) comparative analysis runs feasibility assessment, NMA, and ITC; (v) economic analyses are developed, then updated and adapted as evidence and country requirements change; and (vi) reports and dossiers generation yields HTA-ready outputs with automated QA, appraisal simulation, and translation. The pipeline thus maintains one evidence base that is current, portable, and reusable across deliverables.
CONCLUSIONS: Connecting standalone GenAI-assisted HEOR modules through MCP into one human-governed pipeline keeps evidence current, portable, and reusable, while preserving expert oversight and auditable, RAISE-aligned decisions across the evidence chain.