Risk-Guided Screening for Atrial Fibrillation Using Electronic Health Records

Area of Science:

• Cardiology and Artificial Intelligence
• Predictive modeling for cardiovascular disease
• Public health screening strategies

Background:

• Current atrial fibrillation (AF) screening methods may be enhanced by risk stratification.
• Electronic Health Records (EHRs) offer a valuable data source for developing predictive models.
• The need for scalable and efficient AF screening tools is critical for early intervention.

Purpose of the Study:

• To develop and externally validate a machine learning prediction model (FIND-AF 2.0) using EHR data to guide AF screening.
• To prospectively test the model's ability to identify individuals requiring targeted AF screening.
• To assess the stroke risk associated with AF in high-risk populations identified by the model.

Main Methods:

• A random forest model (FIND-AF 2.0) was developed using age, sex, and 10 comorbidities from EHRs across multiple countries.
• External validation was performed in large international cohorts (UK, Japan, Israel, Canada, China).
• A prospective study involved participants aged ≥30 years with elevated stroke risk, stratified by FIND-AF 2.0 risk, undergoing multi-lead ECG monitoring.

Main Results:

• FIND-AF 2.0 demonstrated good to excellent predictive performance across all validation cohorts (AUROC > 0.7).
• The prospective study found significantly higher AF diagnosis rates in high-risk (4.5%) versus low-risk (0.6%) FIND-AF 2.0 individuals.
• High-risk AF patients without anticoagulation in the FinACAF registry had a 6.0 events per 100 patient-years ischemic stroke rate.

Conclusions:

• The EHR-based FIND-AF 2.0 machine learning model effectively identifies a high-risk subpopulation for AF diagnosis.
• This model enables scalable, EHR-driven, risk-guided AF screening, facilitating targeted interventions.
• FIND-AF 2.0 has the potential to improve early detection of AF and subsequent stroke prevention strategies.