AIGC-driven virtual patients for diabetes: a systematic review

This systematic review (SR) examines the application of Artificial Intelligence-Generated Content (AIGC) in developing virtual patients (VPs) for diabetes management. Through structured analysis of current literature, the study demonstrates how AIGC methodologies address key clinical challenges: generative adversarial networks (GANs) create synthetic continuous glucose monitoring trajectories to overcome data scarcity; reinforcement learning (RL) optimizes personalized insulin regimens through simulated treatment responses; and multimodal fusion techniques generate detailed models for diabetic complications. These approaches demonstrate technical feasibility in retrospective analyses and experimental settings for medical education and clinical decision support. However, their translation to clinical practice awaits rigorous prospective validation with diabetes-specific endpoints. However, validation remains primarily retrospective, with limited prospective trials reporting diabetes-specific endpoints like time-in-range and hypoglycemia metrics. Critical implementation barriers include data privacy concerns, model generalizability across diabetes subtypes, and clinical workflow integration. An empirical mapping of included studies onto a three-dimensional maturity framework reveals that while 92% of studies achieve technical validation, none demonstrate implementation maturity and only one has prospective evidence of clinical efficacy. Future progress hinges on interdisciplinary collaboration to develop robust validation frameworks aligned with regulatory standards and to proactively mitigate algorithmic biases, thereby bridging the gap between technical innovation and dependable clinical application.