Adaptive refinement and prompt-guided conditioning for clinically realistic LLM-generated synthetic pediatric data Frontiers