Machine learning models integrating dietary data predict all-cause mortality in U.S. NAFLD patients: an NHANES-based...

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Nutritional Science

Background:

• Non-alcoholic fatty liver disease (NAFLD) is a major cause of chronic liver disease, linked to metabolic issues and lifestyle.
• Current prognostic models for NAFLD often overlook crucial dietary factors.
• This study addresses the need to incorporate nutritional data into NAFLD mortality prediction.

Purpose of the Study:

• To develop and validate machine learning models for predicting all-cause mortality in NAFLD patients.
• To integrate demographic, serological, and nutritional data, with a focus on dietary interventions.
• To identify key predictors of mortality, emphasizing the role of diet.

Main Methods:

• Analysis of 2,589 NAFLD participants from the NHANES 2007-2018 dataset.
• Utilized LASSO-Cox regression to identify significant survival-associated variables.
• Developed and evaluated five machine learning models (RSF, GBM, CoxBoost, SurvivalSVM, XGBoost) using AUC, C-index, Brier score, and SHAP values for interpretability.

Main Results:

• LASSO-Cox identified 13 significant variables, including age, income, glucose, physical activity, and dietary fiber.
• GBM and RSF models achieved strong predictive performance (AUC ~0.8) for 5- and 10-year survival.
• Poor prognosis linked to age, low income, hyperglycemia, and sedentary behavior; higher dietary fiber intake correlated with better survival.

Conclusions:

• Machine learning models incorporating dietary data effectively predict NAFLD patient mortality.
• The Random Survival Forest (RSF) and Gradient Boosting Machine (GBM) models demonstrated high accuracy.
• Increased dietary fiber intake emerged as a significant protective factor, highlighting its therapeutic potential in NAFLD.