Machine learning for predicting metabolic-associated fatty liver disease including NHHR: a cross-sectional NHANES study

Abstract

Objective Metabolic - associated fatty liver disease (MAFLD) is a common hepatic disorder with increasing prevalence, and early detection remains inadequately achieved. This study aims to explore the relationship between the non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio (NHHR) and MAFLD, and to establish a predictive model for MAFLD using NHHR as a key variable. Methods All participants were selected from the NHANES cohort, spanning from 2017 to March 2020. Multiple linear regression models were employed to examine the relationship between the non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio (NHHR) and the controlled attenuation parameter (CAP). To explore the non-linear association between NHHR and CAP, smooth curve fitting and restricted cubic splines (RCS) of the adjusted variables were utilized. Subgroup analyses were conducted to identify variations in the relationships between the independent and dependent variables across different populations. Finally, a metabolic - associated fatty liver disease (MAFLD) prediction model was developed using seven machine learning methods, including eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Multilayer Perceptron (MLP), Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and logistic regression. The SHAP (SHapley Additive exPlanations) value was employed to interpret the importance of various features. Result Weighted multiple linear regression models revealed a significant positive correlation between the NHHR and the CAP (Beta =  7.42, 95% CI: 5.35-9.50, P <  0.001). Smooth curve fitting and RCS demonstrated a non-linear relationship between NHHR and CAP. Subgroup analyses indicated that this relationship was more pronounced in females. Among the seven machine learning predictive models incorporating NHHR, the XGBoost algorithm exhibited the highest predictive performance, with an area under the curve (AUC) of 0.828. Furthermore, NHHR was identified as the second most important feature in the SHAP analysis, following body mass index (BMI), highlighting its potential in predicting MAFLD. Conclusion A significant positive correlation was identified between the NHHR and the CAP. The inclusion of NHHR in the XGBoost predictive model for MAFLD demonstrated robust predictive capability, providing a valuable tool for the early detection of MAFLD with considerable clinical application potential.