RESUMO
PURPOSE: The present study validated a software-based electrocardiogram (ECG) analysis tool for detection of atrial fibrillation (AF) and risk for AF using polysomnography (PSG)-derived ECG recordings. METHODS: The Stroke Risk Analysis® (SRA®) software was applied to 3-channel ECG tracings from diagnostic PSG performed in enrolled subjects including a subgroup of subjects with previously documented AF. No subjects used positive airway pressure therapy. All ECG recordings were visually analyzed by a blinded cardiologist. RESULTS: Of subjects enrolled in the study, 93 had previously documented AF and 178 of 186 had an ECG that could be analyzed by either method. In subjects with known history of AF, automated analysis using SRA® classified 47 out of 87 ECG as either manifest AF or showing increased risk for paroxysmal AF (PAF) by SRA® (sensitivity 0.54, specificity 0.86). On visual analysis, 36/87 ECG showed manifest AF and 51/87 showed sinus rhythm. Among the latter subgroup, an increased risk for PAF was ascribed by SRA® in 11 cases (sensitivity 0.22, specificity 0.78) and by expert visual analysis in 5 cases (sensitivity 0.1, specificity 0.90). Among 36/178 ECG with manifest AF on visual analysis, 33 were correctly identified by the SRA® software (sensitivity and specificity 0.92). CONCLUSION: Sleep studies provide a valuable source of ECG recordings that can be easily subjected to software-based analysis in order to identify manifest AF and automatically assess the risk of PAF. For optimal evaluability of data, multiple channel ECG tracings are desirable. For assessment of PAF risk, the SRA® analysis probably excels visual analysis, but sensitivity of both methods is low, reflecting that repeated ECG recording remains essential.