Risk-Guided Screening for Atrial Fibrillation Using Electronic Health Records.

Summary

FIND-AF 2.0 used age, sex, and 10 comorbidities to predict new AF within 6 months across EHR datasets from the United Kingdom, Japan, Israel, Canada, and China. The model showed good to excellent discrimination, with AUROC values ranging from 0.747 to 0.835 across national cohorts, and was prospectively tested as a tool to guide scalable AF screening.

Key Findings

• FIND-AF 2.0 was developed and externally validated across five countries using EHR data.
• AUROC values for new AF prediction ranged from 0.747 in Canada to 0.835 in Israel, with AUROC greater than 0.7 in men and women across cohorts.
• A prospective study tested the model for identifying a high-risk population for scalable AF screening.

Clinical Implications

Health systems could use routinely collected EHR data to prioritize individuals for opportunistic or systematic AF screening, potentially improving detection before stroke occurs. Clinical deployment should include local recalibration, equity assessment, workflow integration, and prospective evaluation of effects on AF detection, anticoagulation, stroke, and health-care utilization.

Limitations

• The abstract does not provide the complete prospective screening results or downstream clinical outcomes.
• Prediction performance varied across countries, indicating possible calibration and transportability challenges.
• The observational EHR design may be affected by differences in coding, access to care, and ascertainment of AF.

Future Directions

Future studies should report the prospective screening yield, confirmatory testing burden, treatment changes, stroke prevention, cost-effectiveness, and equity impact. Pragmatic trials should compare risk-guided screening with usual care and evaluate recalibration across additional health systems and underrepresented populations.