Predicting multifaceted risks using machine learning in atrial fibrillation: insights from GLORIA-AF study.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Atrial fibrillation (AF) patients face elevated risks of ischemic stroke and death.
• Anticoagulants reduce these risks but increase bleeding complications.
• Existing clinical risk scores have limited predictive accuracy, particularly for mortality.

Purpose of the Study:

• To evaluate a multi-label gradient boosting decision tree (ML-GBDT) model for predicting adverse outcomes in AF patients.
• To compare the ML-GBDT model's performance against established clinical risk scores.

Main Methods:

• Utilized data from the Global Registry on Long-Term Oral Anti-Thrombotic Treatment in Patients with Atrial Fibrillation (2011-2020).
• Trained and validated the ML-GBDT model to predict all-cause death, ischemic stroke, and major bleeding within one year.
• Compared model discrimination using Area Under the Curve (AUC) and Net Reclassification Index (NRI) against Charlson Comorbidity Index, CHA₂DS₂-VASc, and HAS-BLED scores.

Main Results:

• The ML-GBDT model demonstrated superior prediction for death (AUC 0.785 vs. 0.747), ischemic stroke (AUC 0.691 vs. 0.613), and major bleeding (AUC 0.698 vs. 0.607) compared to clinical scores.
• Significant improvements in NRI were observed for all outcomes (10.0% for death, 12.5% for stroke, 23.6% for bleeding).
• Included 25,656 patients with a mean age of 70.3 years; 1-year event rates were 3.5% for death, 0.8% for stroke, and 1.6% for bleeding.

Conclusions:

• The ML-GBDT model significantly outperforms traditional clinical risk scores in predicting adverse outcomes for AF patients.
• This advanced model can serve as a comprehensive tool for optimizing patient risk assessment.
• Implementing the ML-GBDT model can aid in mitigating adverse events and improving management strategies for atrial fibrillation.