Development and validation of machine learning models for nonalcoholic fatty liver disease

Area of Science:

• Hepatology and data science applications in clinical diagnostics.

Background:

• Nonalcoholic fatty liver disease (NAFLD) is the most common liver condition globally.
• Early diagnosis of NAFLD is crucial for reducing associated morbidity and mortality.

Purpose of the Study:

• To develop and validate a novel predictive model for NAFLD by integrating various risk factors.
• To assess the performance of multiple machine learning models in predicting NAFLD.

Main Methods:

• Utilized a training set of 578 participants with abdominal ultrasound data.
• Employed LASSO regression and Random Forest for predictor screening.
• Developed and tuned five machine learning models: LR, RF, XGBoost, GBM, and SVM.
• Externally validated the best model using a testing set of 131 participants with MRI data.

Main Results:

• Identified key predictors including visceral adiposity index, abdominal circumference, BMI, ALT, ALT/AST ratio, age, HDL-C, and triglycerides.
• Achieved high Area Under Curve (AUC) values for all models, ranging from 0.900 to 0.928.
• The optimized XGBoost model demonstrated superior predictive performance with an AUC of 0.938.

Conclusions:

• Developed and validated five machine learning models for NAFLD prediction.
• The XGBoost model exhibited the highest predictive accuracy and reliability.
• XGBoost serves as a valuable tool for the early identification of individuals at high risk for NAFLD in clinical settings.