A MULTI-AGENT GENAI SYSTEM FOR TRACEABLE, MULTI-COUNTRY LANDSCAPE ASSESSMENT IN MASH
Author(s)
Jag Chhatwal, PhD1, Mine Tekman, PhD2, Ismail F. Yildirim, MSc2, Sumeyye Samur, PhD2, Turgay Ayer, PhD2;
1Massachusetts General Hospital/Harvard Medical School, Boston, MA, USA, 2Value Analytics Labs, Boston, MA, USA
1Massachusetts General Hospital/Harvard Medical School, Boston, MA, USA, 2Value Analytics Labs, Boston, MA, USA
OBJECTIVES: Generative AI has the potential to accelerate HEOR and market access tasks, but commonly used models (e.g., ChatGPT, Perplexity) often lack the accuracy, depth, and traceability required for rigorous evidence synthesis. We developed HEOR-specific multi-agent AI system, ValueGen.AI to address these limitations. This study evaluated the system’s ability to autonomously generate a comprehensive, multi-country HEOR landscape assessment for metabolic dysfunction-associated steatohepatitis (MASH).
METHODS: We developed a hierarchical "Deep Agent" architecture using the LangGraph framework to orchestrate over 1,000 specialized sub-agent invocations. The system functions through three functional layers: 1) A Main Orchestrator that decomposes queries and evaluates information sufficiency; 2) A Deep Agent Layer executing parallel retrieval tasks; and 3) A Tool Layer utilizing the Model Context Protocol (via FastMCP) to interface with heterogeneous data sources (databases, APIs, and web repositories). Inter-agent communication is managed via an asynchronous message queue, enabling non-linear, iterative data gathering. This structure ensures every generated insight is verified and directly traceable to a primary reference.
RESULTS: The ValueGen.AI generated a >100-page, multi-regional HEOR landscape assessment for adult MASH in under 48 hours, compared with months using conventional manual approaches. The report synthesized more than 300 verifiable references, covering epidemiology, clinical trials, real-world evidence, and economic models through October 2025, with projections to 2050. The assessment comprehensively addressed critical HEOR domains, including comparative effectiveness, HRQoL utilities, budget impact, and stakeholder policy, alongside granular subgroup analyses (e.g., T2D, obesity, genetic risk). The content and references were verified by human in the loop.
CONCLUSIONS: Multi-agent AI workflows can dramatically improve HEOR productivity by compressing months of work into days. By moving beyond simple prompting to a structured, agentic architecture, this approach demonstrates that Gen AI can perform complex HEOR tasks without compromising the scientific rigor and traceability essential for decision-making.
METHODS: We developed a hierarchical "Deep Agent" architecture using the LangGraph framework to orchestrate over 1,000 specialized sub-agent invocations. The system functions through three functional layers: 1) A Main Orchestrator that decomposes queries and evaluates information sufficiency; 2) A Deep Agent Layer executing parallel retrieval tasks; and 3) A Tool Layer utilizing the Model Context Protocol (via FastMCP) to interface with heterogeneous data sources (databases, APIs, and web repositories). Inter-agent communication is managed via an asynchronous message queue, enabling non-linear, iterative data gathering. This structure ensures every generated insight is verified and directly traceable to a primary reference.
RESULTS: The ValueGen.AI generated a >100-page, multi-regional HEOR landscape assessment for adult MASH in under 48 hours, compared with months using conventional manual approaches. The report synthesized more than 300 verifiable references, covering epidemiology, clinical trials, real-world evidence, and economic models through October 2025, with projections to 2050. The assessment comprehensively addressed critical HEOR domains, including comparative effectiveness, HRQoL utilities, budget impact, and stakeholder policy, alongside granular subgroup analyses (e.g., T2D, obesity, genetic risk). The content and references were verified by human in the loop.
CONCLUSIONS: Multi-agent AI workflows can dramatically improve HEOR productivity by compressing months of work into days. By moving beyond simple prompting to a structured, agentic architecture, this approach demonstrates that Gen AI can perform complex HEOR tasks without compromising the scientific rigor and traceability essential for decision-making.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR155
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Gastrointestinal Disorders