A MULTI-AGENT GENERATIVE AI FRAMEWORK FOR AUTOMATED LANDSCAPE ASSESSMENTS IN RARE DISEASES: A CASE STUDY IN MULTIPLE SYSTEM ATROPHY
Author(s)
Jag Chhatwal, PhD1, Mine Tekman, PhD2, Ismail F. Yildirim, MSc2, Sumeyye Samur, PhD2, Turgay Ayer, PhD2, Marko Mychaskiw, PhD3, Rinat Ariely, BSc, MBA, MSc, PhD4;
1Harvard Medical School / Massachusetts General Hospital, Boston, MA, USA, 2Value Analytics Labs, Boston, MA, USA, 3Teva Pharmaceuticals, West Chester, PA, USA, 4Teva Pharmaceuticals, Parsippany, NJ, USA
1Harvard Medical School / Massachusetts General Hospital, Boston, MA, USA, 2Value Analytics Labs, Boston, MA, USA, 3Teva Pharmaceuticals, West Chester, PA, USA, 4Teva Pharmaceuticals, Parsippany, NJ, USA
OBJECTIVES: Market access strategy requires rapid, comprehensive understanding of decision-maker evidence expectations. This need is particularly challenging in rare diseases due to low prevalence, phenotypic heterogeneity, and limited comparative evidence. We evaluated whether an agentic generative AI framework could address these challenges by accurately and efficiently synthesizing evidence on epidemiology, natural history, economic burden, and unmet needs, using Multiple System Atrophy (MSA) as a case study.
METHODS: We developed a "Deep Agent" architecture using the LangGraph framework to orchestrate >1,000 specialized sub-agents packaged in ValueGen.AI platform. The system operates via three functional layers: 1) a Main Orchestrator for query decomposition; 2) a Deep Agent Layer for parallelized retrieval tasks; and 3) a Tool Layer utilizing the Model Context Protocol to interface with heterogeneous sources (databases, registries, APIs). The system was tasked with generating market access landscape assessment for MSA for 10 countries. AI generated results were validated via human review for quality assurance.
RESULTS: The system generated a comprehensive 100-page market access landscape assessment for MSA in 48 hours, synthesizing 300 verifiable references—a process comparable to months of manual effort. The output accurately characterized MSA epidemiology (prevalence: 1.9-12.1 per 100,000), natural history (predictable UMSARS decline; median survival 6-10 years), and the symptomatic-only treatment landscape. The AI successfully identified critical evidence gaps, including the lack of established Minimal Clinically Important Differences (MCIDs) for key endpoints (UMSARS, MSA-QoL), the dominance of informal caregiving in societal costs, and the scarcity of cost-effectiveness analyses. It further provided a roadmap for RWE generation, recommending specific biomarkers (NfL, α-synuclein).
CONCLUSIONS: Multi-agent generative AI frameworks can substantially accelerate market access landscape assessments in rare diseases, compressing timelines from months to days while maintaining scientific rigor. Human oversight and quality control is still needed to verify the generated information and its quality.
METHODS: We developed a "Deep Agent" architecture using the LangGraph framework to orchestrate >1,000 specialized sub-agents packaged in ValueGen.AI platform. The system operates via three functional layers: 1) a Main Orchestrator for query decomposition; 2) a Deep Agent Layer for parallelized retrieval tasks; and 3) a Tool Layer utilizing the Model Context Protocol to interface with heterogeneous sources (databases, registries, APIs). The system was tasked with generating market access landscape assessment for MSA for 10 countries. AI generated results were validated via human review for quality assurance.
RESULTS: The system generated a comprehensive 100-page market access landscape assessment for MSA in 48 hours, synthesizing 300 verifiable references—a process comparable to months of manual effort. The output accurately characterized MSA epidemiology (prevalence: 1.9-12.1 per 100,000), natural history (predictable UMSARS decline; median survival 6-10 years), and the symptomatic-only treatment landscape. The AI successfully identified critical evidence gaps, including the lack of established Minimal Clinically Important Differences (MCIDs) for key endpoints (UMSARS, MSA-QoL), the dominance of informal caregiving in societal costs, and the scarcity of cost-effectiveness analyses. It further provided a roadmap for RWE generation, recommending specific biomarkers (NfL, α-synuclein).
CONCLUSIONS: Multi-agent generative AI frameworks can substantially accelerate market access landscape assessments in rare diseases, compressing timelines from months to days while maintaining scientific rigor. Human oversight and quality control is still needed to verify the generated information and its quality.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
HPR13
Topic
Health Policy & Regulatory
Topic Subcategory
Reimbursement & Access Policy
Disease
SDC: Rare & Orphan Diseases