Abstract / Summary
Given the looming crisis of rapid increase in childhood obesity globally, chronic metabolic diseases have consequently emerged but without sufficient awareness among children. Reliable and accessible screening tools thus are required to screen for significant liver fibrosis in the general pediatric population. This prospective cohort study comprised three independent cohorts: a training cohort, and two validation cohorts from China and the USA, respectively. Eight machine learning algorithms were assessed, and 1767 participants were included. The PediatricLiver score, a K-nearest neighbors-based model incorporating age, sex, body mass index (BMI), low-density lipoprotein cholesterol (LDL-C), gamma-glutamyl transferase (GGT), albumin, and globulin levels, demonstrated good performance with an AUC of 0.82 (95% CI: 0.74–0.89) in a training cohort ( n = 437). The AUC was 0.87 (95% CI: 0.82–0.91) in the validation cohort ( n = 730) and 0.80 (95% CI: 0.72–0.87) in the NHANES cohort ( n = 600). Additionally, the score outperformed established non-invasive fibrosis scores and also demonstrated good performance in BMI-defined subgroups. A free, open-access web-based calculator for the PediatricLiver score is available at: https://pediatricliverscore.shinyapps.io/child_shiny/ . The PediatricLiver score might be a newly developed machine learning-based tool for screening significant liver fibrosis in the general pediatric population.