Machine learning-based model for predicting metabolic dysfunction-associated steatotic liver disease using...

Area of Science:

• Hepatology
• Machine Learning
• Preventive Medicine

Background:

• Metabolic dysfunction-associated steatotic liver disease (MASLD) is a growing concern in young adults, leading to severe long-term liver complications.
• Early detection of MASLD is difficult in asymptomatic individuals, necessitating advanced risk assessment tools.
• Accurate and non-invasive methods are crucial for identifying individuals at risk for MASLD.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting MASLD risk in adults aged 20-40 years.
• To assess the efficacy of non-invasive parameters in predicting MASLD.
• To establish a tool for early MASLD screening in clinical settings.

Main Methods:

• A machine learning model was developed and validated using data from 13,047 participants (training) and 1,335 (external validation).
• MASLD was defined by hepatic steatosis (ultrasonography) and at least one cardiometabolic risk factor.
• Models incorporated age, sex, BMI, blood pressure, body fat percentage (PBF), and skeletal muscle index (SMI), with logistic regression, random forest, and XGBoost algorithms applied.

Main Results:

• The most comprehensive model (Model 3) achieved high internal validation performance with AUROCs up to 0.91.
• External validation demonstrated strong predictive power with AUROCs ranging from 0.88 to 0.89.
• Body mass index (BMI) and PBF were key predictors, while higher skeletal muscle index (SMI) unexpectedly correlated with increased MASLD risk.

Conclusions:

• The developed ML model accurately predicts MASLD risk in young adults using non-invasive parameters.
• This model shows potential for facilitating early MASLD screening in routine clinical practice.
• Further research into the role of SMI in MASLD development is warranted.