[Development and validation of a machine learning-based prediction model for fatty liver in Wilson disease]

Area of Science:

• Hepatology
• Medical Informatics
• Machine Learning

Background:

• Wilson disease (WD) is a genetic disorder of copper metabolism.
• Fatty liver is a common complication in WD patients.
• Early detection and risk stratification of fatty liver in WD are crucial.

Purpose of the Study:

• To develop and evaluate a machine learning-based predictive model for fatty liver in WD patients.
• To identify key clinical variables for predicting fatty liver in WD.
• To compare the performance of seven different machine learning algorithms.

Main Methods:

• Retrospective analysis of clinical data from 1862 WD patients.
• Utilized Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection.
• Compared seven machine learning algorithms (LR, DT, RF, XGBoost, LightGBM, SVM, ANN) using AUC, PR curves, calibration, and DCA.
• Employed SHAP analysis to assess feature contributions.

Main Results:

• 69.60% of WD patients had fatty liver.
• LASSO identified PLT, RDW, ALT, TBA, CIV, and IBIL as key predictors.
• The LightGBM model achieved the highest performance with AUCs of 0.826 (training) and 0.815 (validation).
• SHAP analysis highlighted ALT, IBIL, TBA, and CIV as significant positive predictors.

Conclusions:

• The LightGBM model demonstrates superior predictive performance for fatty liver in WD patients.
• This model can aid in the early screening and risk stratification of WD patients.
• The identified key variables offer insights into the pathogenesis of fatty liver in WD.