Utilizing Machine Learning to Identify Predictors of Corticosteroid Discontinuation 1-Year After Adult Heart...

Area of Science:

• Cardiology
• Immunology
• Data Science

Background:

• Corticosteroids (CS) are crucial post-heart transplantation (HT) but associated with adverse effects.
• Identifying predictors of CS discontinuation is vital for optimizing long-term outcomes.

Purpose of the Study:

• To leverage machine learning (ML) to identify factors associated with CS discontinuation 1-year post-adult HT.
• To develop a predictive model for CS cessation using recipient, donor, and transplant variables.

Main Methods:

• Retrospective analysis of the United Network for Organ Sharing (UNOS) database.
• Utilized eXtreme Gradient Boosting (XGBoost) and Shapley Additive Explanations (SHAP) for predictive modeling and feature importance.
• Included 72,730 adult first-time HT recipients discharged on CS.

Main Results:

• 17.9% of recipients discontinued CS within 1 year.
• Predictors of CS discontinuation included lower transplant center volume, shorter donor ischemic time, and LVAD use at transplant.
• The XGBoost model demonstrated strong performance (AUC=0.854).

Conclusions:

• Machine learning effectively identified key predictors of 1-year CS discontinuation post-heart transplant.
• Transplant center volume, donor ischemia time, and LVAD use are significant factors influencing CS cessation.