OBJECTIVES: This study aims 1) to test how health economists can take advantage of Copilot, a GPT-3-based artificial intelligence tool developed by GitHub and OpenAI, in their cost-effectiveness model development in MS Excel/VBA and 2) to draw attention on some non-programming issues related to the use of AI in programming.

METHODS: An existing CEM for the treatment of Urinary Tract Infection in hospitalized patients was stripped from its VBA code for some interface interactions, for the creation of the cost-effectiveness (CE) frontier and for the Probabilistic Sensitivity Analysis (PSA). Copilot was then prompted to rewrite the code from scratch, first with general prompts, then with step-by-step requests. The generated code was analyzed by an experienced VBA developer to assess quality, verboseness, and performance.

RESULTS: General prompts failed at generating any effective VBA code. Deconstructed algorithms for the CE frontier and the PSA generated effective working code. Performances (runtime) and quality were similar to code generated by an experienced developer.

CONCLUSIONS: This study showed it is feasible to use Copilot to generate VBA code in a CEM. However, the quality and verboseness of the generated code is depending on the user’s detailed requests. The CEM-specific user interface built in sheets contributes to difficulties in generating usable VBA code with Copilot. Copilot will probably accelerate model development for junior developers. But more experienced programmers will need to wait for improvements from GPT-4 and an increased VBA codex. Further research and development of more intelligent GAI algorithms is therefore necessary for the introduction on daily routine. More broadly, potential issues of code explainability, ownership and licensing need to be clarified.