Statistical Applications in Pediatric Endocrinology: A Simulation on Metformin’s Effect on HbA1c in High-Risk Adolescents

Metformin has been increasingly used off-label in adolescents with type 1 diabetes (T1D) or prediabetic conditions to improve glycemic control. Despite clinical trials suggesting modest improvements in HbA1c, evidence remains mixed and methodologically limited. This paper introduces a simulation framework to evaluate advanced causal inference estimators—including Targeted Maximum Likelihood Estimation (TMLE), Double Machine Learning (DML), and Bayesian Causal Forests (BCF)—for estimating the causal effect of metformin on HbA1c reduction in high-risk adolescents. The framework incorporates realistic confounding structures based on empirical data and provides theoretical derivations for estimator properties.