Supervised ensemble methods give higher AUCs than the tried-and-true risk scores, but this systematic review finds that most models cut corners on essentials like external validation and explainability. For now, bleeding risks of over-trusting these black boxes could rival the shortcomings of CHA2DS2-VASc. It’s hard to see a reason to switch in older adults until the methods mature.
Machine Learning for Stroke Risk in AF: Statistical Edge or Clinical Progress?
Do machine learning models fundamentally improve stroke risk stratification in atrial fibrillation compared to CHA2DS2-VASc, and are they truly deployment-ready?