Predicting the spontaneous cardioversion of atrial fibrillation using artificial intelligence-enabled...

Area of Science:

• Cardiology
• Artificial Intelligence
• Health Economics

Background:

• Spontaneous cardioversion (SCV) is common in emergency department (ED) patients with primary atrial fibrillation (AF).
• Predicting SCV can optimize patient discharge and reduce healthcare costs.
• Current methods for predicting SCV in the ED setting are limited.

Purpose of the Study:

• To evaluate the efficacy of artificial intelligence-enabled electrocardiograms (AI-ECGs) in predicting SCV in primary AF patients.
• To assess the potential cost savings of an AI-ECG-guided 'wait-and-see' protocol.

Main Methods:

• Recruited patients presenting to EDs with primary AF.
• Utilized convolutional neural network (CNN) architectures (ResNet50, EfficientNet, DenseNet) and an ensemble learning model for SCV prediction.
• Performed a cost-minimization analysis to evaluate the economic impact of the AI-ECG protocol.

Main Results:

• An ensemble learning AI-ECG model achieved 69.7% accuracy and a ROC AUC of 0.742.
• The AI-ECG-guided 'wait-and-see' protocol reduced per-patient costs from $4681 to $3398.
• This protocol resulted in a 33.3% reduction in overall hospitalizations.

Conclusions:

• AI-ECGs can accurately predict SCV in patients with primary AF presenting to the ED.
• Implementing an AI-ECG-guided 'wait-and-see' protocol offers significant cost savings and reduces hospital admissions.