Clinical and Metabolomic Panel for Noninvasive Screening of Biopsy-Confirmed MASH in Children and Adolescents

Noninvasive tests to identify pediatric metabolic dysfunction-associated steatohepatitis (MASH) remain a critical need. We aimed to develop and validate a screening panel that identifies biopsy-confirmed MASH in children and adolescents using clinical and metabolomics data.

Fasting serum from youth in NASH CRN studies and healthy participants underwent untargeted metabolomics by liquid chromatography-mass spectrometry (LC-MS). Clinical data included anthropometrics, lipids, liver enzymes, and markers of insulin resistance. MASH was determined by biopsy in NASH CRN participants and hepatic steatosis by MRI-PDFF in healthy participants. Clinical features were selected by Kolmogorov-Smirnov testing (p < 0.05), and metabolomics data were reduced to 52 features with non-zero LASSO coefficients. Feature importance was ranked using CatBoost. The final model included eight clinical features and the top 10 annotated metabolites. Data were split into training and test sets, with longitudinal validation in a sub-cohort with 96-week follow-up biopsies.

The cohort included 586 children (ages 5-18, 70% male, mean BMI z-score of 2.35 ± 0.84, 72% Hispanic). Of these, 390 had MASH (21% Zone 3, 41% Zone 1, 38% definite MASH), and 196 were non-MASH. The model incorporated ALT, AST, GGT, platelets, HOMA-IR, BMI z-score, TG:HDL ratio, and alkaline phosphatase. Key metabolic features included amino acid derivatives (e.g., glutamic acid, argininosuccinic acid), indole, diacetylspermine, and lipid metabolites. The model achieved an AUROC of 0.88 (92% sensitivity, 71% specificity) in the test set and 0.81 (69% sensitivity, 86% specificity) in the longitudinal cohort.

This 18-variable panel of eight clinical measures and 10 metabolites showed strong performance for noninvasive screening of biopsy-confirmed MASH in children and potential utility for longitudinal monitoring.

Currently, there are no clinically approved non-invasive biomarkers for diagnosing or monitoring pediatric MASH, despite its growing prevalence and long-term health risks. This study developed and validated a machine-learning-based panel that combines clinical and high-resolution metabolomics data, demonstrating strong performance for classifying MASH and monitoring disease status in a longitudinal cohort. These findings are particularly relevant for improving the care of children with MASH and could inform both clinical practice and research. With further validation, this approach could reduce reliance on liver biopsy, enable earlier detection, and support clinical decision-making for physicians and researchers.