Presentation (CTI)

OBJECTIVES: Excel cost-effectiveness models (CEMs) are a cornerstone of health economic evaluation, yet their complexity often hinders rapid interpretation, validation, and communication, especially for non-technical audiences. This study introduces an AI-powered analyzer tool designed to retrieve, interpret, summarize, explain and walkthrough Excel-based CEMs by extracting key model components and logic using natural language.
METHODS: The tool addresses the core challenge of reasoning over highly disaggregated Excel data where logic is fragmented across multiple sheets and no continuous text or semantic markup exists. Individual model components, such as transition probabilities, discounting, utilities, and treatment pathways are often embedded in thousands of sparsely populated cells with formulas obscuring the true computational logic. This fragmentation makes it non-trivial to chunk, vectorize, or index the retrieve efficiently. We engineered an agentic retrieval-augmented reasoning system that restructures the spreadsheet data into interpretable logical groupings while excluding high-density null regions that bloat memory and degrade performance. Context is dynamically constructed by aligning semantic signals from formulas, cell values, and sheet structures, enabling natural language reasoning without requiring the full workbook to be loaded into memory.
RESULTS: The analyzer demonstrated strong generalizability across diverse CEM architectures including Markov models, partitioned survival models, and cohort-based frameworks. It accurately identified structural assumptions, time horizons, discounting methods, and treatment pathways. Users were able to pose natural language questions (e.g., “How to change model settings” or “What are the scenario analysis assumptions?”), with the tool generating interpretable, evidence-linked responses. This significantly reduced model onboarding and audit time for analysts and reviewers.
CONCLUSIONS: The AI Excel analyzer provides support for understanding complex cost-effectiveness models without manual cell-by-cell inspection. It improves transparency, reduces reliance on model authors, and accelerates stakeholder engagement. Future enhancements will focus on integration with R-based workflows, version tracking, and validation audit trails to further support health technology assessment processes.