USING GENERATIVE AI WITH RETRIEVAL-AUGMENTED GENERATION TO AUTOMATE SLIDE DECK DEVELOPMENT FROM AMCP FORMULARY DOSSIERS
Author(s)
Ankita Sood, PharmD, Gagandeep Kaur, M.Pharm, Rajdeep Kaur, PhD, Barinder Singh, RPh.
Pharmacoevidence, Mohali, India.
Pharmacoevidence, Mohali, India.
OBJECTIVES: AMCP dossiers are pivotal for formulary decision-making in the US, yet translating their content into clear, presentation-ready slide decks is resource-intensive and time-consuming. This study explored the utility of generative artificial intelligence (GenAI) to automate the development of a payer presentation slide deck derived from an AMCP dossier in a psychiatric disorder through a human-in-the-loop approach.
METHODS: A Python-based tool was developed using the Claude 4.6 Sonnet GenAI model. A Retrieval-Augmented Generation (RAG) framework was integrated to ensure that all slide content was grounded in and traceable to the underlying AMCP dossier and supporting source documents. Overall, 75 documents, including peer-reviewed journal articles, conference abstracts, and epidemiological data sources were ingested into the RAG pipeline. A multi-agent architecture was configured to generate structured slide content aligned with key AMCP dossier sections. Outputs were reviewed and validated by subject matter experts (SMEs) for relevance, completeness, accuracy, language, traceability, and overall quality.
RESULTS: An automated slide deck was generated reflecting key content aligned to all major AMCP dossier sections, with outputs including narrative slide content, summary tables, and data visualisations (bar graphs, pie charts, and line graphs). SMEs strongly agreed that generated slides were relevant and accurate, and somewhat agreed that responses were largely complete. Minor issues with content repetition across slides were noted; additionally, formatting gaps were identified that required manual intervention to meet final presentation standards. Overall, the automated pipeline produced a slide deck that was approximately 90% presentation-ready and reduced development time by approximately 80% compared to the manual approach.
CONCLUSIONS: This study demonstrates the potential of GenAI to support medical affairs and HEOR teams in accelerating evidence communication to payers, streamlining the translation of AMCP dossier content into payer-ready slide decks and compressing timelines from weeks to days, while maintaining accuracy, traceability, and alignment with formulary submission standards.
METHODS: A Python-based tool was developed using the Claude 4.6 Sonnet GenAI model. A Retrieval-Augmented Generation (RAG) framework was integrated to ensure that all slide content was grounded in and traceable to the underlying AMCP dossier and supporting source documents. Overall, 75 documents, including peer-reviewed journal articles, conference abstracts, and epidemiological data sources were ingested into the RAG pipeline. A multi-agent architecture was configured to generate structured slide content aligned with key AMCP dossier sections. Outputs were reviewed and validated by subject matter experts (SMEs) for relevance, completeness, accuracy, language, traceability, and overall quality.
RESULTS: An automated slide deck was generated reflecting key content aligned to all major AMCP dossier sections, with outputs including narrative slide content, summary tables, and data visualisations (bar graphs, pie charts, and line graphs). SMEs strongly agreed that generated slides were relevant and accurate, and somewhat agreed that responses were largely complete. Minor issues with content repetition across slides were noted; additionally, formatting gaps were identified that required manual intervention to meet final presentation standards. Overall, the automated pipeline produced a slide deck that was approximately 90% presentation-ready and reduced development time by approximately 80% compared to the manual approach.
CONCLUSIONS: This study demonstrates the potential of GenAI to support medical affairs and HEOR teams in accelerating evidence communication to payers, streamlining the translation of AMCP dossier content into payer-ready slide decks and compressing timelines from weeks to days, while maintaining accuracy, traceability, and alignment with formulary submission standards.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR245
Topic
Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Mental Health (including addiction)