Clinical indicators associated with pericardial effusion in rheumatoid arthritis: a machine learning-based analysis

Area of Science:

• Rheumatology
• Cardiology
• Medical Informatics

Background:

• Pericardial effusion (PE) is a frequent, underdiagnosed complication of rheumatoid arthritis (RA) with significant mortality risk.
• Early PE detection in RA is challenging due to nonspecific symptoms and limited echocardiography access.
• Advanced machine learning (ML) approaches can identify clinical and laboratory features associated with PE in RA patients.

Purpose of the Study:

• To identify simple clinical and laboratory features associated with pericardial effusion (PE) in rheumatoid arthritis (RA) patients.
• To develop and validate machine learning models for predicting PE in RA.
• To enhance early detection and diagnostic suspicion of PE in RA.

Main Methods:

• A cross-sectional study evaluated 958 RA patients undergoing transthoracic echocardiography.
• Comprehensive demographic, clinical, and laboratory variables were analyzed.
• Eight ML algorithms were developed and validated using repeated train-test splits, with performance assessed by AUC-PR and AUC-ROC, and feature importance determined by SHAP analysis.

Main Results:

• Pericardial effusion (PE) was detected in 13.2% of RA patients.
• A random forest model demonstrated superior performance (mean AUC-PR: 0.674, mean AUC-ROC: 0.943).
• Key predictors for PE included palpitation, chest pain, older age, and longer RA disease duration.

Conclusions:

• Readily available clinical features, analyzed by ML, can increase diagnostic suspicion for PE in RA patients.
• RA patients with palpitations or chest pain, especially older individuals or those with long-standing disease, warrant closer echocardiographic evaluation.
• This study pioneers ML application to identify PE indicators in RA, aiding clinical decision-making for timely diagnosis.