Diagnostic Accuracy of Wearable ECG Devices for Atrial Fibrillation and ST-Segment Changes: A Systematic Review

Background: Wearable electrocardiography (ECG) devices such as smartwatches offer a novel means for detecting cardiac arrhythmias, particularly atrial fibrillation (AF), and ST-segment abnormalities. Their role in complementing or replacing traditional ECG methods is being increasingly investigated. Objective: To evaluate the diagnostic performance (sensitivity, specificity) of wearable ECG devices in detecting AF and ST-segment changes, compared to 12-lead ECG as the gold standard. Methods: A systematic search was performed in PubMed, Scopus, and additionally, the SpringerLink platform was consulted up to June 2025, targeting open-access, English-language clinical studies from the last five years. Inclusion criteria: adult population, use of a wearable ECG device, 12-lead ECG comparator, and diagnostic accuracy reporting. Out of 145 records, 5 studies met the inclusion criteria. The systematic review protocol was not prospectively registered in PROSPERO due to the limited number of available studies and the exploratory nature of the topic, which focused on the most recent clinical evaluations of wearable ECG devices. However, the review strictly adhered to the PRISMA 2020 guidelines for systematic reviews to ensure methodological transparency and reproducibility. Results: Five studies encompassing a total of 1133 participants were incorporated into the analysis. Devices evaluated included Apple Watch (Series 4-6), KardiaMobile 6L, FibriCheck, Preventicus, and HUAMI dynamic ECG. Sensitivity ranged from 83% to 100%, and specificity from 79% to 100%. Algorithm improvements and repeated measurements significantly reduced inconclusive recordings. Multichannel ECG methods using smartwatches showed high agreement with 12-lead ECG in ST-elevation myocardial infarction detection. Conclusions: Wearable ECG devices demonstrate high diagnostic performance for AF and ST-segment abnormalities, especially in supervised environments. However, inconclusive recordings and algorithm limitations remain barriers to widespread clinical use. Real-world validation and algorithm refinement are needed for broader adoption.