Detecting atrial fibrillation in the polysomnography-derived electrocardiogram: a software validation study.

Area of Science:

• Cardiology
• Sleep Medicine
• Medical Informatics

Background:

• Atrial fibrillation (AF) is a significant risk factor for stroke.
• Polysomnography (PSG) provides electrocardiogram (ECG) recordings that can be analyzed for AF detection.
• Software-based ECG analysis offers a potential tool for automated AF detection and risk stratification.

Purpose of the Study:

• To validate a software-based ECG analysis tool for detecting atrial fibrillation (AF) and assessing AF risk.
• To evaluate the performance of Stroke Risk Analysis® (SRA®) software using PSG-derived ECGs.

Main Methods:

• Applied SRA® software to 3-channel ECG tracings from diagnostic PSG.
• Included subjects with and without documented AF; no subjects used positive airway pressure therapy.
• Compared software analysis to visual analysis by a blinded cardiologist.

Main Results:

• SRA® correctly identified 33 out of 36 ECGs with manifest AF (sensitivity and specificity 0.92).
• For subjects in sinus rhythm, SRA® identified increased paroxysmal AF (PAF) risk in 11/51 cases (sensitivity 0.22, specificity 0.78), compared to 5/51 by visual analysis (sensitivity 0.1, specificity 0.90).
• In subjects with known AF history, SRA® classified 47/87 ECGs as manifest AF or at increased risk (sensitivity 0.54, specificity 0.86).

Conclusions:

• Sleep studies are a valuable source for software-based ECG analysis to detect AF and assess PAF risk.
• SRA® analysis may outperform visual analysis for PAF risk assessment, though sensitivity for both is low.
• Multiple-channel ECG tracings are desirable for optimal data evaluation; repeated ECG recordings remain essential for accurate assessment.