Presentation (CTI)

OBJECTIVES: Given the importance of AMCP dossiers in formulary submissions and payer decision-making across the United States (US), optimizing accuracy and efficiency of content development is essential. This study explored the utility of generative artificial intelligence (GenAI) to automate generation of AMCP for a psychiatric disorder to produce reliable, traceable outputs through a human-in-the-loop approach.
METHODS: A Python-based interface was developed using the Claude 3.7 Sonnet GenAI. The tool, deployed on a secure AWS cloud infrastructure combined a Retrieval-Augmented Generation (RAG) framework with a multi-agent architecture to ensure that outputs were fully traceable to source documents. A total of 110 documents including journal articles, conference abstracts, treatment guidelines, and epidemiology data sources were uploaded into the RAG pipeline for processing of text, tables, and figures/plots. The agents were configured to generate different AMCP sections, in accordance with established Format for Formulary Submissions, version 5.0, and the outputs were validated by subject matter experts (SMEs) for relevance, completeness, accuracy, language, and overall quality, using a 5-point Likert scale.
RESULTS: An AMCP dossier was generated, encompassing sections on executive summary, product and disease description, clinical, value and modeling report, and appendices. Output included tables and visualizations, such as bar graphs, pie charts and line graphs. Based on Likert scale assessment, SMEs strongly agreed that the generated content was relevant and accurate, and somewhat agreed that the responses were largely complete. In all sections, there were instances of repetition, however SMEs concurred that the responses were mostly well-written (Strongly or Somewhat agreed). Overall, the automated approach generated a draft that was ∼90% complete, reduced development time by ∼80% compared to manual methods, and resulted in 70-75% cost reduction.
CONCLUSIONS: This study highlights potential of GenAI to streamline AMCP dossier development process, substantially reducing timeline from weeks/months to days, while maintaining accuracy, traceability and adherence to established guidelines.