Presentation (CTI)

OBJECTIVES: Excel-based health economic models are widely used in health technology assessment to inform reimbursement and pricing decisions, but require advanced technical expertise, limiting their usability for broader stakeholder engagement and regulatory review. This study aimed to develop a conversational AI assistant capable of interacting with these models, enabling natural language queries, input changes, scenario analysis, and output retrieval.
METHODS: The system was developed to interact with Excel-based models using natural language. The architecture was built using Claude 3.7 Sonnet and a Retrieval-Augmented Generation framework designed to process complex Excel files while preserving formulas, charts, and embedded logic. A five-step processing pipeline was implemented, including model upload, preserving formulas, embedded logic, and converting the content into a format readable by the AI system. To evaluate the tool, a three-state cost-effectiveness Markov model was uploaded, and 50 test prompts were developed by subject matter experts.
RESULTS: The prompts were executed to assess the system’s ability to retrieve and interpret outputs from Excel mode. Parameter retrieval prompts (e.g., discount rates, transition probabilities, and results) were answered with 100% accuracy. Scenario-based prompts, such as modifying utility values or treatment costs, triggered correct recalculations in 92% of cases. The tool also provided interpretable explanations for underlying calculations, supporting transparency. Users were able to engage in follow-up queries for clarification or additional analysis. The average system response time was approximately 10 seconds, compared to 30-60 minutes for equivalent manual execution and validation.
CONCLUSIONS: The conversational AI assistant demonstrated strong potential to enhance efficiency and accessibility in health economics and outcomes research workflows by enabling interaction with Excel-based models. By facilitating natural language queries and dynamic scenario analysis, the tool reduces reliance on technical expertise while maintaining analytical integrity, supporting broader stakeholder engagement, and promoting transparency in decision making.