Explainable Clinical Decision Support for Metabolic Index Prediction in Gout Patients Using GA-Optimized Ensemble...

Area of Science:

• Metabolic disease research
• Clinical informatics
• Machine learning in healthcare

Background:

• Metabolic indices (HOMA-IR, METS-IR, TyG) are crucial for assessing cardiometabolic risk and insulin resistance.
• Gout patients present complex metabolic profiles requiring accurate risk assessment tools.

Purpose of the Study:

• To develop and validate ensemble learning models for simultaneous prediction of HOMA-IR, METS-IR, and TyG.
• To utilize demographic, clinical, and laboratory data for predicting these metabolic indices in gout patients.
• To ensure clinical interpretability of the predictive models.

Main Methods:

• Analysis of retrospective data from 411 gout patients.
• Application of logarithmic transformation and genetic algorithms for feature selection.
• Optimization and validation of five ensemble learning models, including CatBoost, using Optuna and SHAP analysis.

Main Results:

• The CatBoost model, integrated with a genetic algorithm and logarithmic transformation, demonstrated superior prediction accuracy and minimal error rates.
• The model accurately simulated the TyG index formula and identified key predictors: insulin for HOMA-IR, BMI for METS-IR, and triglyceride variance for TyG.
• SHAP analysis confirmed the model's alignment with clinical pathophysiology.

Conclusions:

• The proposed GA + Log-Transformed CatBoost workflow offers a highly accurate, non-invasive, and interpretable clinical decision support system.
• Ensemble models can effectively capture complex pathophysiological patterns relevant to metabolic health.
• This approach enhances clinical decision-making for patients with gout and related metabolic conditions.