Presentation (CTI)

OBJECTIVES: In Health Economics and Outcomes Research (HEOR), effective data and knowledge translation across multidisciplinary teams is vital to ensure seamless collaboration and informed decision-making. Such teams often require information in different formats, such as code, natural language summaries, or visual representations like flow diagrams or charts. Converting data from one format to another can be time-intensive and prone to errors. This study explores the potential of Generative Artificial Intelligence (Gen-AI) as a tool to automate and enhance these translation processes in HEOR workflows.
METHODS: A proof-of-concept exercise was performed using a large language model (LLM) and subsequent tool development in a Python environment. The Gen-AI tool was designed to handle multiple tasks, including translating programming code (e.g., from VBA to Python), generating VBA macros from natural language descriptions, transforming mathematical expressions into structured methodological documentation for health economic models, creating flow diagrams and other graphical representations (using frameworks like the Mermaid library) from textual inputs. Specialized LLMs fine-tuned on various code bases were employed to facilitate code conversion. Retrieval augmented generation (RAG) provided coding examples and best-practice guidelines to improve the tool’s outputs. Additionally, multimodal LLMs enabled generation of both textual and visual outputs.
RESULTS: The resulting HEOR translator tool seamlessly converted data across formats, significantly reducing repetitive technical tasks. It effectively translated programming languages, created macros, explained code in plain language, and generated illustrative visuals like flow diagrams and data workflows from textual descriptions.
CONCLUSIONS: The findings highlight the potential of a Gen-AI-powered translator in optimizing HEOR workflows, reducing manual effort, and ensuring effective communication across stakeholders with varying technical expertise. Such a tool can act as a bridge between technical teams and decision-makers, facilitating the effective interpretation and dissemination of HEOR results.