Toward an explainable AI-Based clinical decision support system for predicting adverse outcomes in Rhabdomyolysis

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Nephrology

Background:

• Rhabdomyolysis presents significant morbidity and mortality risks, often due to acute kidney injury.
• Current risk stratification methods for rhabdomyolysis are insufficient for complex patient data.
• Early identification of high-risk patients is crucial for timely intervention.

Purpose of the Study:

• To develop and validate an explainable AI (XAI) model for predicting a composite outcome of renal replacement therapy or 90-day mortality in rhabdomyolysis patients.
• To lay the foundation for an XAI-based clinical decision support system (CDSS).
• To enable real-time risk scoring at the point of care.

Main Methods:

• Utilized routinely available admission data from 1031 adult patients.
• Applied multivariate imputation, Boruta feature selection, and ADASYN for class imbalance.
• Developed and evaluated a CatBoost machine learning model, employing SHAP for explainability.

Main Results:

• The CatBoost model demonstrated high predictive performance (AUC=0.942, accuracy=0.913).
• Key predictors identified by SHAP include creatinine, troponin T, and albumin, aligning with clinical knowledge.
• Decision-curve analysis indicated a superior net benefit compared to existing strategies.

Conclusions:

• An interpretable AI model can effectively predict severe rhabdomyolysis outcomes, enhancing clinical decision-making.
• The proposed framework allows integration into electronic health records for real-time risk assessment.
• Age-specific model refinement is necessary, highlighting the importance of transparent AI in healthcare.