Establishment of a machine learning predictive model for non-alcoholic fatty liver disease: a longitudinal cohort study

Background and Aims Non-alcoholic fatty liver disease (NAFLD) is a common chronic liver disease, which lacks effective drug treatments. This study aimed to construct an eXtreme Gradient Boosting (XGBoost) prediction model to identify or evaluate potential NAFLD patients. Methods and Results We conducted a longitudinal study of 22,140 individuals from the Beijing Health Management Cohort. Variable filtering was performed using the least absolute shrinkage and selection operator. Randomly Over Sampling Examples was used to address imbalanced data. Next, the XGBoost model and the other three machine learning (ML) models were built using balanced data. Finally, the variable importance of the XGBoost model was ranked. Among four ML algorithms, We got that the XGBoost model outperformed the other models with the following results: accuracy of 0.835, sensitivity of 0.835, specificity of 0.834, Youden index of 0.669, precision of 0.831, recall of 0.835, F-1 score of 0.833, and an area under the curve of 0.914. The top five variables with the greatest impact on the onset of NAFLD were aspartate aminotransferase, cardiometabolic index, body mass index, alanine aminotransferase, and triglyceride-glucose index. Conclusion The predictive model based on the XGBoost algorithm enables early prediction of the onset of NAFLD. Additionally, assessing variable importance provides valuable insights into the prevention and treatment of NAFLD.