Abstract
OBJECTIVE: This study aimed to investigate the clinical characteristics and plasma metabolites of nonalcoholic fatty liver disease (NAFLD) in obese Chinese children and to develop machine learning-based NAFLD diagnostic models. METHODS: We recruited 222 obese children aged 4 to 17 years and divided them into an obese control group and an obese NAFLD group based on liver ultrasonography. Mass spectrometry metabolomic analysis was used to measure 106 metabolites in plasma. Binary logistic regression was used to identify NAFLD-related clinical variables. NAFLD-specific metabolites were illustrated via volcano plots, cluster heatmaps, and metabolic network diagrams. Additionally, we applied 8 machine learning methods to construct 3 diagnostic models based on clinical variables, metabolites, and clinical variables combined with metabolites. RESULTS: By evaluating clinical variables and plasma metabolites, we identified 16 clinical variables and 14 plasma metabolites closely associated with NAFLD. We discovered that the level of 18:0 to 22:6 phosphatidylethanolamines was positively correlated with the levels of total cholesterol, triglyceride-glucose index, and triglyceride to high-density lipoprotein cholesterol ratio, whereas the level of glycocholic acid was positively correlated with the levels of alanine aminotransferase, gamma-glutamyl transferase, insulin, and the homeostasis model assessment of insulin resistance. Additionally, we successfully developed 3 NAFLD diagnostic models that showed excellent diagnostic performance (areas under the receiver operating characteristic curves of 0.917, 0.954, and 0.957, respectively). CONCLUSIONS: We identified 16 clinical variables and 14 plasma metabolites associated with NAFLD in obese Chinese children. Diagnostic models using these features showed excellent performance, indicating their potential for diagnosis.