Exploring Metabolic Changes in Children with Congenital Hypothyroidism: A Serum Metabolomic Study Combined by Machine Learning

Congenital hypothyroidism (CH) is a genetic endocrine disorder that can cause developmental delays if it is untreated. In this study, NMR-based metabolomics was employed to analyze serum samples from CH children and healthy controls across different age groups. Multivariate statistical analysis screened for 17, 16, 33, and 21 differential metabolites in the respective age groups and identified seven common metabolites, including lysine, 1-methylhistidine, glycerophosphocholine, phosphocholine, β-glucose, lipids, and creatine. The results indicated that CH children experienced metabolic disturbances in multiple pathways, particularly glycerophospholipid metabolism and glycine, serine, and threonine metabolism. Following recursive feature elimination (RFE) for feature selection, the top five core metabolites were selected to construct an optimized artificial neural network (ANN) model for CH diagnosis, achieving a prediction accuracy of 89.4%. These findings suggest that the identified metabolites can be used as potential diagnostic biomarkers for CH in children. This may help improve the early diagnosis accuracy of CH, serve as a rapid screening tool for newborns, and provide an auxiliary diagnostic method for suspected CH cases to facilitate early clinical intervention.