Predicting the risk of bone erosion in rheumatoid arthritis using a SHAP-based interpretable machine learning model

Area of Science:

• Rheumatology
• Machine Learning
• Medical Imaging

Background:

• Bone erosion (BE) is a critical prognostic indicator in rheumatoid arthritis (RA).
• Accurate prediction of BE risk is crucial for timely intervention.
• Existing methods may lack interpretability or predictive power.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting BE risk in RA patients.
• To utilize the Shapley Additive exPlanations (SHAP) framework for model interpretability.
• To identify key predictors of BE in RA patients.

Main Methods:

• A multi-center retrospective study of 412 RA patients without baseline BE.
• Data preprocessing included imputation and resampling for class imbalance.
• Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature selection, followed by training ten ML algorithms.
• Model performance evaluated using AUC, Brier score, and Decision Curve Analysis (DCA).
• SHAP framework used for interpreting the optimal model's global and individual feature contributions.

Main Results:

• The eXtreme Gradient Boosting (XGBoost) model achieved high performance (AUC 0.896 internal, 0.893 external).
• Key predictors identified by SHAP analysis included synovial Power Doppler Imaging (PDI), synovial hyperplasia, disease duration, and anti-cyclic citrullinated peptide antibody.
• DCA indicated the XGBoost model provided clinical utility across a range of threshold probabilities.

Conclusions:

• The SHAP-interpretable XGBoost model effectively predicts BE risk in RA patients.
• Synovial PDI and synovial hyperplasia are identified as pivotal factors in BE development.
• This interpretable ML approach offers a promising tool for personalized RA management.