Identification of atrial fibrillation using heart rate variability: a meta-analysis.

Background: Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and is associated with significant cardiovascular complications. Recently, artificial intelligence (AI) algorithms have leveraged heart rate variability (HRV) patterns to enhance the accuracy of AF identification. Methods: We conducted a systematic review of the literature by searching four major biomedical databases-PubMed, Web of Science, Embase, and Cochrane Library-spanning from their inception to December 13, 2024, following the PRISMA guidelines. We extracted data on true positives, false positives, true negatives, and false negatives from the included studies, which were then synthesized to evaluate sensitivity and specificity comprehensively. Results: Our final analysis included 12 diagnostic studies. Hierarchical summary receiver operating characteristic modeling revealed excellent discriminative ability, with a pooled sensitivity of 0.94 and specificity of 0.97. In detecting AF, the AI model demonstrated exceptional performance (sensitivity = 0.96, specificity = 0.99, AUC = 1.00). Subgroup analyses revealed that both deep learning algorithms (sensitivity = 0.95, specificity = 0.98, AUC = 0.99) and multi-database studies (sensitivity = 0.96, specificity = 0.97, AUC = 0.99) demonstrated enhanced accuracy in AF identification compared to other approaches. Conclusion: Machine learning can effectively identify AF with HRV in ECG, especially in diagnosis and detection, with deep learning algorithms and multiple-databases outperforming other diagnostic methods. Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/, PROSPERO (CRD42025634406).