Arrhythmias & electrophysiology
Development and validation of a novel machine learning–based algorithm to predict incident atrial fibrillation: A multicohort analysis
Authors' conclusion
Race-agnostic ML models demonstrated superior predictive performance and reduced bias compared with CHARGE-AF
Study context
- Population
- Participants free of atrial fibrillation (AF) at baseline; derivation in ARIC visit 5 (2011–2013) and Cardiovascular Health Study (1989–1990), validation in MESA and Framingham Offspring and Generation 3derivation 16,719
- Comparison
- Race-agnostic machine learning (ML) risk score vs CHARGE-AF score
- Outcome
- Incident AF
- Follow-up
- Within 5 years
Methodological detailsmulticohort, machine learning, external validation cohort
- Multicohort
- Machine learning
- External validation cohort
Performance
As reported in the abstract
Prediction of incident AF · 5 years
| Test or model | Concordance index | Brier score |
|---|---|---|
| ML model | 0.83 (95% CI 0.80-0.85) | 1.82 |
| CHARGE-AF | 0.77 (95% CI 0.74-0.79) | 1.92 |
Concordance index
ML model0.83 (95% CI 0.80-0.85)
CHARGE-AF0.77 (95% CI 0.74-0.79)
Brier score
ML model1.82
CHARGE-AF1.92
- ML model vs CHARGE-AF: P < 0.001 — discrimination
- ML model vs CHARGE-AF: P < 0.001 — calibration
Further findings
| Outcome | Values |
|---|---|
| Incident AF, by cohort · 5-year follow-up | Derivation cohort 3.0% · Validation cohort 1.9% |
| Population-attributable risk of AF from age, by race | Non-Hispanic black 14.3% · White 34.6% |
Algorithmic bias (disparate impact, equal opportunity difference, Theil index)
- ML model reduced algorithmic bias vs CHARGE-AF across all metrics
Key predictors
- Age, clinical factors (ECG parameters, cardiac biomarkers, BP) and education level
Original publication
Development and validation of a novel machine learning–based algorithm to predict incident atrial fibrillation: A multicohort analysis. Heart Rhythm. Published 5 December 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →