Exploring AI Simulation in Diabetes Remission Prediction for Bariatric Surgery

A proof-of-concept study explored the use of AI-simulated patients to select diabetes remission prediction models for bariatric surgery and found it can improve model fit for individual preoperative scenarios. The team tested a reproducible framework that applied six validated models across 100 simulated cases, providing a practical testbed to inform model selection in practice.

The investigators generated 100 structured simulated profiles with realistic ranges for age, BMI, HbA1c, C‑peptide and medication use, then applied a standardized prompt and decision algorithm to six validated tools to quantify remission-prediction concordance and prompt stability. The study emphasized reproducibility through repeated LLM sessions and human oversight, positioning these methods as a preclinical testbed upstream of real-world validation in the clinical workflow.

Methodological strengths include the structured, reproducible simulation that enables direct model-to-model comparison under controlled inputs and the use of human expert review to limit AI errors

AI-simulated outputs can clarify relative predictions from competing diabetes remission prediction models, helping clinicians compare likely remission estimates for individual patients and structure clearer risk conversations.