Interpreting Variable Importance in Metabolically Co-Regulated Systems: Considerations for Metabolic Dysfunction-associated Steatotic Liver Disease Risk Prediction.

• Metabolic co-regulation limits independent interpretation of predictors. • Shared latent structure can distort supervised variable-importance rankings. • Nonlinear and threshold behavior may alter apparent predictor contributions. • Random Forest–based importance is highly sensitive to collinearity. • Unsupervised filters can stabilize feature attribution and improve clarity.