Arrhythmias & electrophysiology
Deep learning analysis of electrocardiogram for risk prediction of drug-induced arrhythmias and diagnosis of long QT syndrome
Authors' conclusion
CNN models applied to ECGs outperform QTc measurements to identify exposure to drugs altering the QT interval, congenital LQTS, and are greatest shortly after a diTdP episode.
Study context
- Population
- Healthy subjects before and after sotalol, congenital long-QT syndromes (cLQTS) patients and patients with drug-induced TdP (diTdP)3 cohorts
- Comparison
- Convolutional neural network (CNN) models vs QTc to identify sotalol exposure, cLQTS and diTdP risk
- Outcome
- Identification of sotalol exposure (ROC-AUC)
Performance
As reported in the abstract
Identification of sotalol exposure
| Test or model | ROC-AUC |
|---|---|
| CNN models | 0.98 |
| QTc | 0.72 |
ROC-AUC
CNN models0.98
QTc0.72
- CNN models vs QTc: P ≤ 0.001 — 0.98 vs. 0.72
Further findings
Multiple vs single 10-s electrocardiogram (ECG)
- Higher ROC-AUC using multiple vs single 10-s ECG
- P ≤ 0.001
8-lead vs single-lead models
- Performances comparable for 8-lead vs single-lead models
Detection of cLQTS vs healthy controls
- Accurate detection of presence and type of cLQTS; cLQT2 AUC-ROC
- 0.9
CNN model output after diTdP event
- Greatest shortly after a diTdP event, declining over time
- P ≤ 0.001
- After controlling for QTc and intake of culprit drugs
ECG segment analysis
- J-Tpeak interval best discriminator of sotalol intake
Original publication
Deep learning analysis of electrocardiogram for risk prediction of drug-induced arrhythmias and diagnosis of long QT syndrome. Eur Heart J. Published 1 September 2021. doi:10.
Summarised from the published abstract; every number is checked against it