EP Edge Journal Watch: A Late Breaking Overview September 2026-EP Edge Journal Watch Ep 38 Transcript

EP Edge Journal Watch Ep 38 Transcript

Pdf Summary

This podcast transcript from EP Edge at HRX summarizes three electrophysiology abstracts.<br /><br />First, the hosts discuss a study testing commercially available large language models on single-lead ECG interpretation from implantable loop recorder strips. Models including ChatGPT, Claude, Gemini, Qwen, and MedGemma were compared against electrophysiologist interpretation on 1,000 strips. Performance was poor overall: ChatGPT did best with only 52.7% rhythm accuracy, while the medical-specific model MedGemma performed worst at 28%. The models often misread atrial flutter as sinus rhythm and undercalled atrial fibrillation, suggesting that current public LLMs are not reliable for ECG interpretation.<br /><br />Second, they review a retrospective study of 3,522 ambulatory ECG monitors looking for ventricular tachycardia during exercise and daily activities. The observed VT rate was 1.05%, higher than expected for a general population, though the speakers note likely selection bias because these were patients already referred for monitoring. Higher PVC burden was associated with more nonsustained VT, suggesting longer monitoring may reveal additional arrhythmias in some patients.<br /><br />Third, they examine the COSPIRE registry analysis of attempted conduction system pacing. Core lab adjudication reclassified about 19% of implants, and outcomes differed markedly: mortality was 7.5% in patients classified as deep septal pacing versus 0.8% in true conduction system pacing. The speakers interpret this as evidence that if proper left bundle branch area pacing criteria are not met, especially in heart failure patients, clinicians should consider alternative strategies such as a coronary sinus LV lead rather than accepting suboptimal deep septal pacing.<br /><br />Overall, the episode emphasizes cautious interpretation of AI-generated ECG reads, the importance of patient selection in arrhythmia prevalence studies, and the need for rigorous standards in conduction system pacing.

Keywords

electrophysiology

ECG interpretation

large language models

atrial fibrillation

atrial flutter

ventricular tachycardia

ambulatory ECG monitoring

conduction system pacing

left bundle branch area pacing

implantable loop recorder