Artificial intelligence for direct-to-physician reporting of ambulatory electrocardiography.

Summary

In 14,606 ambulatory ECG recordings, an ensemble AI (DeepRhythmAI) achieved markedly higher sensitivity for critical arrhythmias than certified technicians (98.6% vs 80.3%), reducing false negatives by ~14-fold per patient compared to human review. Although AI increased false positives modestly, its strong negative predictive value supports direct-to-physician reporting.

Key Findings

• AI sensitivity for critical arrhythmias was 98.6% vs 80.3% for technicians; false negatives were 3.2 vs 44.3 per 1,000 patients.
• Relative risk of missed diagnosis was 14.1-fold higher for technicians compared with AI.
• AI had higher false-positive event rate (median 12 vs 5 per 1,000 patient-days), trading specificity for sensitivity.

Clinical Implications

AI-only preliminary analysis could enable direct-to-physician reports, reducing delays and technician workload, while flagging high-risk arrhythmias promptly. Clinical pathways should incorporate human oversight for false-positive management.

Limitations

• Higher false-positive rate may increase downstream review burden.
• Model performance may vary across devices, populations, and acquisition protocols not represented.

Future Directions

Prospective implementation studies assessing clinical outcomes, cost-effectiveness, and optimal human-AI oversight models, plus domain shift robustness across vendors and care settings.