Multifactorial nutritional and metabolic risk profiles for behavioral outcomes in Chinese children: a cross-sectional study at a Health Management Center using explainable machine learning and principal component analysis.

Background: Behavioral problems affect 10-20% of children worldwide. Conventional pediatric assessment of risk factors for behavioral problems relies largely on single-biomarker analysis, which fails to capture the synergistic or antagonistic interactions among nutritional and metabolic factors and is further hampered by multicollinearity among biomarkers. Whether integrated nutritional-metabolic patterns improve risk identification beyond individual biomarkers remains unclear. We therefore combined multivariable regression, explainable machine learning, and principal component analysis (PCA) to identify integrated nutritional-metabolic risk patterns for behavioral problems in children. Methods: This cross-sectional study included 5,468 children aged 6-18 years who underwent comprehensive health check-ups at the Health Management Center of a tertiary Children's Hospital in China between August 2022 and December 2025 and who had complete biomarker and behavioral data. Blood magnesium, iron, copper, zinc, and lead were measured by inductively coupled plasma mass spectrometry (ICP-MS); triglycerides (TG), calcium, and phosphorus by enzymatic or colorimetric assays; and 25-hydroxyvitamin D [25(OH)D] by chemiluminescent immunoassay (CLIA). Behavioral problems were assessed using the Conners' Parent Rating Scale (CPRS) based on Chinese norms. Covariates (age, sex, and body mass index) were obtained from standardized anthropometric measurements. Multivariable logistic regression, XGBoost with SHapley Additive exPlanations (SHAP), and PCA were applied. Results: Among the participants (median age 9.41 years; 61.5% males), 624 (11.41%) had behavioral problems. Higher blood lead [odds ratio (OR) 2.26, 95% confidence interval (CI): 1.51-3.37] and TG (OR 1.77, 95% CI: 1.10-2.82) were associated with increased odds of behavioral problems, whereas iron was protective (OR 0.06, 95% CI: 0.01-0.39). SHAP analysis ranked lead, iron, vitamin D, and TG as the most influential predictors. PCA derived four patterns; Pattern 4 (lead exposure-nutritional antagonism: elevated lead with low calcium and vitamin D) showed the strongest, dose-response association (highest Conclusions: In this health check-up sample, a pattern characterized by elevated lead combined with insufficient calcium and vitamin D was associated with behavioral problems in a dose-response manner. Although causality cannot be inferred, pattern-based nutritional-metabolic profiling during routine pediatric check-ups may help identify children who could benefit from closer behavioral monitoring, nutritional optimization, and lead exposure reduction.