Background: Nonalcoholic fatty liver disease (NAFLD) is highly prevalent in children and is associated with obesity. Objectives: To test whether addition of bioelectrical impedance analysis (BIA) parameters to BMI and anthropometric indices improves the prediction performance of NAFLD than BMI z score (BAZ) alone. Methods: This cross-sectional study recruited 933 children 6-12 years of age for anthropometric measure, BIA, and liver ultrasound. Prediction models of the BAZ, anthropometric, and BIA sets were built in children with obesity using machine learning algorithms. Results: Prevalences of NAFLD were 44.4% (59/133) and 20% (12/60) in boys and girls with obesity, respectively. In both sexes, BAZ set performed worst; adding anthropometric indices into the model improved the model performance, whereas BIA parameters were the best approach for predicting NAFLD. The best result in boys achieved had an accuracy of 75.9% and area under receiver operating characteristic curve of 0.854. In girls, the best result achieved had an F-measure score of 0.615, Matthews correlation coefficient of 0.512, and area under precision-recalled curve of 0.697. Conclusion: BIA is a simple and highly precise tool that yields better NAFLD prediction model than anthropometric indices, and much better performance than BAZ. This study suggests BIA as a potential predictor for pediatric NAFLD.
Keywords: bioelectrical impedance analysis; body composition; nonalcoholic fatty liver disease; pediatric obesity.