A smartphone-based PPG algorithm detected AF with accuracy comparable to standard electrocardiography in a large, multinational cohort. Data presented at the American Heart Association (AHA) 2025 Sessions evaluated algorithm performance across smartphone models and clinical subgroups.

The study enrolled 236 participants from five academic centers in the United States and Europe. The system analyzed PPG signals recorded on smartphones to detect individual heartbeats, estimate heart rate, and classify rhythm using convolutional neural networks. Rhythm classification was compared with the standard 12-lead electrocardiogram (ECG).

The algorithm demonstrated high diagnostic performance without technician verification. Overall accuracy was 98.5% (95% confidence interval [CI] 98.0%-99.0%). Sensitivity was 96.3% (95% CI 94.4%-97.7%), and specificity was 99.3% (95% CI 98.8%-99.7%). Positive predictive value was 98.0% (95% CI 96.5%-98.9%), and negative predictive value was 99.8% (95% CI 99.6%-99.9%).

Performance remained consistent across ten smartphone models and across clinical subgroups, including heart failure, vascular disease, hypertension, diabetes, and prior stroke. Sensitivity was reduced in individuals with darker skin tone and higher BMI. Technician verification mitigated this reduction.These findings confirm 

that smartphone-based PPG algorithms can accurately detect AF across diverse patient populations and clinical contexts.