Objective:

To develop interpretable machine learning (ML) models for predicting low health-related quality of life (HRQoL) in patients with metabolic dysfunction–associated steatotic liver disease (MASLD) and identify key predictors.

Methods:

In 821 MASLD patients, HRQoL was assessed (Chronic Liver Disease Questionnaire) and categorized. Six predictors were selected via feature selection. Data were split into training/test sets. Six ML algorithms were trained and evaluated using ROC curves, calibration, and decision curve analysis. SHAP provided interpretability.

Results:

Key predictors included alcohol use, exercise, smoking, age, education, and arthritis. Ensemble models (e.g., Gradient Boosting, AUC=0.942) performed best. SHAP identified lifestyle factors (alcohol, exercise) as primary drivers, with age modifying their impact.

Conclusion:

Interpretable ML models accurately predict low HRQoLin MASLD. Lifestyle behaviors are dominant predictors, supporting potential use for risk screening and targeted interventions.

In 821 MASLD patients, HRQoL was assessed (Chronic Liver Disease Questionnaire) and categorized. Six predictors were selected via feature selection. Data were split into training/test sets. Six ML algorithms were trained and evaluated using ROC curves, calibration, and decision curve analysis. SHAP provided interpretability.

Key predictors included alcohol use, exercise, smoking, age, education, and arthritis. Ensemble models (e.g., Gradient Boosting, AUC=0.942) performed best. SHAP identified lifestyle factors (alcohol, exercise) as primary drivers, with age modifying their impact.

Interpretable ML models accurately predict low HRQoLin MASLD. Lifestyle behaviors are dominant predictors, supporting potential use for risk screening and targeted interventions.