Interpretable Machine Learning Framework for Predicting Major Adverse Cardiovascular Events in Rheumatoid Arthritis...

Area of Science:

• Rheumatology
• Cardiovascular Medicine
• Artificial Intelligence in Healthcare

Background:

• Patients with rheumatoid arthritis (RA) have an elevated risk of major adverse cardiovascular events (MACE).
• The clinical utility of machine learning (ML) for cardiovascular risk prediction in RA is limited by the "black-box" nature of complex algorithms.

Purpose of the Study:

• To develop interpretable survival models for predicting MACE in RA patients.
• To provide transparent and actionable insights for personalized cardiovascular risk management in RA.

Main Methods:

• Utilized data from 2461 RA patients (2011-2022) from the Taipei Medical University Clinical Research Database.
• Compared ML survival models (Random Survival Forest, DeepSurv, Cox-Time) against the Cox proportional hazards model.
• Integrated permutation importance and SHAP (Shapley additive explanations) for model interpretability.

Main Results:

• Random Survival Forest (RSF) exhibited superior performance (C-index: 0.8771, integrated Brier score: 0.0775).
• Key predictors identified include creatinine, conventional synthetic disease-modifying antirheumatic drugs, C-reactive protein, alanine aminotransferase, and age at RA diagnosis.
• SHAP analysis revealed specific protective and risk-increasing effects of medications and laboratory markers.

Conclusions:

• RSF with SHAP analysis provides interpretable and personalized cardiovascular risk predictions for RA patients.
• This approach enhances clinical decision-making and advances precision medicine in rheumatology.
• Future research should focus on temporal and external validation to ensure model generalizability.