Dietary amino acid composition and elevated fasting glucose in children and adolescents from four Chinese provinces: a cross-sectional study combining regression and machine learning

Objective Using an integrated analytical pipeline that combines regression and machine learning, with nonlinear dose–response modeling, this cross-sectional study sought to examine dietary amino acid composition in relation to elevated fasting glucose (EFG) in children and adolescents. Methods Data from 12,567 children and adolescents (6–18 years) from four provinces in a national nutrition and health surveillance program were analyzed. Feature selection was performed using the Boruta algorithm, and propensity score matching (PSM) was utilized to address confounding. Conditional logistic regression was used to assess the relationships between EFG risk and 20 amino acids expressed as percentages of total protein intake; odds ratios (ORs) were reported per 1-percentage-point increment in the proportion of total protein intake. Predictive models were built with six machine learning algorithms (LightGBM, XGBoost, SVM, KNN, CatBoost, and NB) and the SHAP method was used to measure feature contributions and determine the most important amino acids. Further stratified analyses and a generalized additive model (GAM) were conducted to assess association strength and investigate possible nonlinear dose–response relationships. Results 6,960 participants were included in the final analysis. The best-performing model was LightGBM (AUC = 0.831), which included this model included 20 amino acids plus 12 confounding factors (including fat intake, protein intake, hypertension, and general/central obesity). Conditional logistic regression combined with SHAP-based feature importance ranking identified methionine (OR = 2.03, 95% CI: 1.61–2.57) and threonine (OR = 1.59, 95% CI: 1.32–1.92) as positively associated with EFG, whereas arginine (OR = 0.82, 95% CI: 0.74–0.92) and tyrosine (OR = 0.71, 95% CI: 0.59–0.86) were inversely associated (all p < 0.01). Stratified analyses showed generally consistent directions of association. The GAM showed that methionine, arginine, and tyrosine had significant nonlinear relationships with EFG risk ( P for nonlinearity < 0.05). Conclusion The dietary amino acid intake is strongly associated with EFG levels in children and adolescents. Methionine and threonine are positively associated, and arginine and tyrosine are inversely associated. Collectively, these four amino acids may be important dietary correlates associated with abnormal glucose metabolism in this study population.