Accuracy of Smartwatches in the Detection of Atrial Fibrillation

Figures

• 

  Figure 1. PRISMA Flow Diagram of Study Selection Process A total of 1,342 records were identified through database searches. After removing duplicates and ineligible records, 90 full-text articles were assessed for eligibility. Of these, 26 studies met the inclusion criteria and were included in the final review. Reasons for exclusion are detailed at each stage. PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

• 

  Figure 2. Risk of Bias Assessment Using QUADAS-2 Tool This figure presents the risk of bias assessment for the 26 included studies using the QUADAS-2 framework. Most studies demonstrated low risk of bias across domains, though a few showed methodological limitations in patient selection and index test conduct. The overall risk of bias is summarized in the final column. QUADAS = Quality Assessment of Diagnostic Accuracy Studies.

• 

  Figure 3. Forest Plot of Sensitivity for Device Subgroup Analysis The forest plot presents the pooled sensitivity estimates of smartwatches for detecting atrial fibrillation, stratified by device brand. The pooled sensitivity across all studies was 94.81% (95% CI: 91.94% to 96.70%).

• 

  Figure 4. Forest Plot of Specificity for Device Subgroup Analysis The forest plot presents the pooled specificity estimates of smartwatches for detecting atrial fibrillation, stratified by device brand. The pooled specificity across the included studies was 96.40% (95% CI: 94.12% to 97.82%).

• 

  Figure 7. Meta-Regression: log (DOR) vs log (Sample Size) by Device A positive association was observed in several subgroups. Apple Watch and Amazfit studies showed higher diagnostic ORs in larger samples, suggesting sample size as a driver of observed heterogeneity. DOR = diagnostic OR.

• 

  Figure 5. SROC Curves for Smartwatch Subgroups Each curve represents the diagnostic performance of a specific brand: Apple Watch (AUC: 0.968; SD: 0.01), Samsung (AUC: 0.983; SD: 0.013), Withing’s (AUC: 0.967; SD: 0.05), and Amazfit (AUC: 0.98; SD: 0.001). The plot highlights excellent overall diagnostic accuracy across all devices. SROC = summary receiver-operating characteristic.

• 

  Figure 6. Pooled SROC Curves by Smartwatch Detection Technology This figure illustrates the pooled diagnostic performance of smartwatches using different detection technologies—ECG and PPG—for the detection of AF. Both indicate high diagnostic accuracy for both modalities. AF = atrial fibrillation; ECG = electrocardiogram; PPG = photoplethysmography; other abbreviation as in Figure 5.

• 

  Central Illustration. Diagnostic Accuracy of Smartwatches for Atrial Fibrillation Detection Based on 26 Studies Including 17,349 Patients Showing excellent pooled sensitivity and specificity values by technology. Abbreviation as in Figure 6.