From Sleep Patterns to Heart Rhythms: Predicting Atrial Fibrillation from Overnight Polysomnograms.

Area of Science:

• Cardiology
• Biomedical Engineering
• Sleep Medicine

Background:

• Atrial fibrillation (AF) is often asymptomatic, increasing risks of stroke and heart failure.
• Obstructive sleep apnea (OSA) is highly prevalent in AF patients (60-90%).
• Polysomnography (PSG) offers a unique opportunity for early AF prediction via ECG analysis.

Purpose of the Study:

• To identify individuals at high risk of future atrial fibrillation (AF).
• To develop a predictive model using single-lead ECG data from standard PSG recordings.
• To leverage existing PSG data for proactive AF screening.

Main Methods:

• Analyzed 18,782 single-lead ECG recordings from 13,609 subjects.
• Extracted 1,800 hand-crafted features and updated a pre-trained deep neural network.
• Trained a shallow neural network on ECG and AF probability features for future AF prediction.

Main Results:

• The model achieved 67% sensitivity and 81% specificity in predicting AF.
• Survival analysis indicated a significant hazard ratio of 8.36 for AF outcomes (p < 1.93 × 10^-52).
• Modest precision (0.3) suggests potential for false positives.

Conclusions:

• ECG analysis of overnight PSG data shows potential for AF prediction.
• This approach could enable low-cost screening and proactive treatment for high-risk individuals.
• Further refinement with additional physiological parameters may reduce false positives and enhance clinical utility.