# Combined analysis of the triglyceride–glucose index and melanin-concentrating hormone in metabolic dysfunction–associated fatty liver disease: a machine learning–based study

## TL;DR

This study explores how the triglyceride–glucose index and melanin-concentrating hormone can help identify people at risk for fatty liver disease using machine learning models.

## Contribution

The study introduces a machine learning-based model combining MCH and TyG index for MAFLD screening with strong predictive performance.

## Key findings

- MCH and TyG index are independent risk factors for MAFLD.
- A logistic regression model achieved high accuracy in predicting MAFLD risk.
- The TyG index partially mediates the association between MCH and MAFLD.

## Abstract

Metabolic dysfunction-associated fatty liver disease (MAFLD), a highly prevalent global liver disorder, requires simple and accessible screening approaches. As current diagnostic methods, such as the Controlled Attenuation Parameter (CAP), are limited in their applicability in obese patients and are primarily designed for fibrosis assessment. This study aim to investigate the associations of the serum melanin-concentrating hormone (MCH) and triglyceride–glucose (TyG) indices with MAFLD and to explore the risk factors and disease probability of MAFLD by developing machine learning models.

In this cross-sectional study of 212 MAFLD patients and 107 healthy controls, and feature selection were identified through the least absolute shrinkage and selection operator (LASSO) regression analysis and Variance Inflation Factor (VIF). Three predictive models—Logistic Regression, Random Forest, Support Vector Machinemodel (SVM)—were constructed using the training set and evaluated in an independent test set. Construction of nomogram using independent risk factors screened by machine learning. Multivariate logistic regression analysis was used to explore further assess independent risk factors. Mediation analysis was conducted to explore potential pathways.

Logistic regression model was found to outperform other classifier models in testing data [area under the curve (AUC) of 92.6, 95% CI: 0.865–0.987] and achieve the lowest Brier Score as well. Decision curve analysis suggested potential clinical utility. Logistic regression analysis indicated that MCH (OR, 2.193; 95% CI, 1.242–3.873; P = 0.007), TyG index (OR, 1.002; 95% CI, 1.001–1.003; P < 0.001), are independent risk factors for MAFLD. Subgroup analysis of the association between MCH and MAFLD stratified by sex, age, and body mass index (BMI) showed no significant effect modification after adjustment. Mediation analysis indicated that the TyG index accounted for a modest proportion of the association between MCH and MAFLD (mediation proportion: 10.89%).

Serum MCH and the TyG index were independently associated with MAFLD. A machine learning–based screening model was developed and internally validated, showing promising performance for identifying individuals at higher risk. However, external validation in larger multicenter prospective cohorts is warranted before broader clinical application.