From Pilot to Platform: Expanding Generative AI Coauthoring Across Global Value Dossier (GVD) Sections With Proven Quality and Validation
Author(s)
Larisa Gofman, PhD1, Sahil Sharma, M. Pharm2.
1ZS Associates, Princeton, NJ, USA, 2ZS Associates, Gurugram, India.
1ZS Associates, Princeton, NJ, USA, 2ZS Associates, Gurugram, India.
OBJECTIVES: Evaluate the scalability of our global value dossier (GVD) Coauthoring Accelerator in developing additional dossier chapters on disease burden; including clinical, humanistic, and economic; without compromising accuracy or quality.
METHODS: Building on our existing methodology and proof-of-concept of our Gen AI GVD Accelerator, we curated an outline template to demonstrate how our tool can support other chapters of a GVD including burden of disease which included clinical burden, economic burden, and humanistic burden. Our Retrieval-Augmented Generation (RAG) framework was used to apply the configured prompts to the reference documents. Prompt engineering thoughtfully aligned with key evidence types relevant to each burden domain. We evaluated prompts and quality of outputs for those three sections of the GVD to validate the following key metrics (1) accuracy of extracted content, (2) traceability to references of the original evidence sources, (3) efficiency and time to generate each chapter, and (4) completeness.
RESULTS: The Accelerator demonstrated strong content development for all three chapters with efficiency and a similar accuracy rate to our previous validation (~95%) in extracting the key required evidence. Our tool demonstrated strong accuracy, efficiency, and completeness for the additional sections. Traceability to input references was observed, with similarity scores generally in the moderate-to-high range. Creation time was reduced by ~70% compared to manual methods, accelerating the time develop an output.
CONCLUSIONS: We produced modular GVD content across clinical, economic, and humanistic chapters of a GVD with high accuracy, completeness, and traceability. This further demonstrates and supports our original proof-of-concept for the potential to leverage a GVD Coauthoring Accelerator to effectively support dossier development. These findings further support strategic quality and value of using AI to expedite and support alignment across stakeholders in driving content generation with a focus on HTA and payer requirements.
METHODS: Building on our existing methodology and proof-of-concept of our Gen AI GVD Accelerator, we curated an outline template to demonstrate how our tool can support other chapters of a GVD including burden of disease which included clinical burden, economic burden, and humanistic burden. Our Retrieval-Augmented Generation (RAG) framework was used to apply the configured prompts to the reference documents. Prompt engineering thoughtfully aligned with key evidence types relevant to each burden domain. We evaluated prompts and quality of outputs for those three sections of the GVD to validate the following key metrics (1) accuracy of extracted content, (2) traceability to references of the original evidence sources, (3) efficiency and time to generate each chapter, and (4) completeness.
RESULTS: The Accelerator demonstrated strong content development for all three chapters with efficiency and a similar accuracy rate to our previous validation (~95%) in extracting the key required evidence. Our tool demonstrated strong accuracy, efficiency, and completeness for the additional sections. Traceability to input references was observed, with similarity scores generally in the moderate-to-high range. Creation time was reduced by ~70% compared to manual methods, accelerating the time develop an output.
CONCLUSIONS: We produced modular GVD content across clinical, economic, and humanistic chapters of a GVD with high accuracy, completeness, and traceability. This further demonstrates and supports our original proof-of-concept for the potential to leverage a GVD Coauthoring Accelerator to effectively support dossier development. These findings further support strategic quality and value of using AI to expedite and support alignment across stakeholders in driving content generation with a focus on HTA and payer requirements.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
HTA156
Topic
Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Decision & Deliberative Processes, Systems & Structure, Value Frameworks & Dossier Format
Disease
Respiratory-Related Disorders (Allergy, Asthma, Smoking, Other Respiratory)