An expert-AI co-creation framework for robust health economic simulations

Health economic (HE) evaluations are essential for healthcare decision-making, yet the rapid pace of innovation demands accelerated HE evaluations. Integrating Generative AI (GenAI), particularly Large Language Models (LLMs), into HE simulation modeling offers a promising opportunity to enhance model development efficiency and speed.

We propose a framework for automating HE model development through expert-AI co-creation. This framework enables GenAI to generate simulation models in a structured and transparent manner. The proposed approach is expert-led and incorporates iterative prompt-response cycles. GPT5-mini and GPT-5.2-codex are employed using zero-shot prompting. As LLM results are non-deterministic, we introduced a Prompt Robustness Test (PRT) to assess the stability of LLM-generated outputs, indicating whether a prompt is sufficiently reliable to support automated model development. The methodology is applied to replicate a published discrete-event simulation comparing two post-surgery treatment strategies for colon cancer. The original model is complex and representative of real-world HE practice.

Using our framework, we replicated the model in Python, achieving cost and QALY estimates within 2% of the original. While the overall pathway structure was successfully reconstructed, several logic discrepancies remained compared with the original model.

Our framework provides methodological guidance on the expert-AI co-creation of complex HE models, emphasizing that expert oversight is essential throughout the process.