The machine learning prediction model of non-alcoholic fatty liver; the role of hydrogen and methane breath tests

Nonalcoholic fatty liver disease (NAFLD) is now the leading cause of global chronic liver disease, affecting approximately 32.4% of the population in various regions and imposing healthcare and economic burdens. The gold standard for the diagnosis of NAFLD, such as liver biopsy, has numerous limitations in large-scale screening. Recent studies have explored the use of machine learning to diagnose NAFLD. In this study, we investigated the effect of the lactulose breath test (LBT) on a machine-learning model for predicting NAFLD. The input variables for machine learning included three combination sets to assess the effect of the LBT results: anthropometric characteristics and blood test results; anthropometric characteristics and LBT results; and anthropometric characteristics, blood test results, and LBT results. The machine learning models developed in this study included linear regression, support vector machine, K-nearest neighbour, Random forest, and extreme gradient boosting (XGBoost) with 536 participants. The model performance was evaluated using six metrics: Accuracy, Area Under the Receiver Operating Characteristic curve (AUROC), specificity, sensitivity, precision, and F1 score. Among the six models, XGBoost had the highest AUROC at 0.88. The AUROC results from the three combination variable sets indicate that the LBT results significantly improve the model performance. LBT results improve the NAFLD prediction model and provide an opportunity for additional NAFLD screening in patients receiving LBT.