Here, we seek to identify metabolite predictors of dysglycemia in youth. In the discovery analysis among 391 youth in the Exploring Perinatal Outcomes among CHildren (EPOCH) cohort, we used reduced rank regression (RRR) to identify sex-specific metabolite predictors of impaired fasting glucose (IFG) and elevated fasting glucose (EFG: Q4 vs. Q1 fasting glucose) 6 years later and compared the predictive capacity of four models: Model 1: ethnicity, parental diabetes, in utero exposure to diabetes, and body mass index (BMI); Model 2: Model 1 covariates + baseline waist circumference, insulin, lipids, and Tanner stage; Model 3: Model 2 + baseline fasting glucose; Model 4: Model 3 + baseline metabolite concentrations. RRR identified 19 metabolite predictors of fasting glucose in boys and 14 metabolite predictors in girls. Most compounds were on lipid, amino acid, and carbohydrate metabolism pathways. In boys, no improvement in aurea under the receiver operating characteristics curve AUC occurred until the inclusion of metabolites in Model 4, which increased the AUC for prediction of IFG (7.1%) from 0.81 to 0.97 (p = 0.002). In girls, %IFG was too low for regression analysis (3.1%), but we found similar results for EFG. We replicated the results among 265 youth in the Project Viva cohort, focusing on EFG due to low %IFG, suggesting that the metabolite profiles identified herein have the potential to improve the prediction of glycemia in youth.
Keywords: dysglycemia; impaired fasting glucose; metabolomics; predictive model; youth type 2 diabetes.