Presentation (CTI)

OBJECTIVES: To evaluate the feasibility of using generative AI to automatically generate and execute cost-effectiveness models based on user-defined clinical and economic parameters, reducing development time and improving reproducibility.
METHODS: A proof-of-concept framework was built using an LLM trained on health economics conventions. The AI system was evaluated across four representative cost-effectiveness models (two Excel-based and two R-based), including both state-transition (Markov) and partitioned survival structures, spanning different therapeutic areas and time horizons. For each model, a structured “Parameters” input sheet (e.g., transition probabilities, utilities, costs) was used to automatically generate both R- and Excel-based executable models. The AI also generates sensitivity analyses. The resulting models were compared against manually programmed reference models for accuracy and structure.
RESULTS: AI-generated health economic models reproduced the structural logic, state transitions, and ICER outputs of benchmark Excel- and R-based reference models. Base-case ICERs were within ±10% of published values, consistent with variation commonly observed in independent human-led replications due to rounding and incomplete reporting of intermediate calculations. Model build time was reduced by approximately 30-40%, compared with traditional manual programming, while maintaining full transparency and auditability of code and inputs
CONCLUSIONS: Generative AI can effectively and efficiently support health economic model development by automating model construction and execution from standard parameter inputs. AI-assisted models produce results consistent with published analyses and within expected replication variability observed in human-led implementations. With appropriate expert oversight, AI-assisted modeling can streamline HTA workflows, freeing analysts to focus on interpretation, scenario design, and value communication.