Arrhythmias & electrophysiology
Development of an AI-Driven QT Correction Algorithm for Patients in Atrial Fibrillation
Authors' conclusion
A CNN model applied to AF ECGs accurately predicted the sinus QTc interval, outperforming current alternatives and exhibiting a high negative predictive value.
Study context
- Population
- Patients with a 12-lead ECG in atrial fibrillation (AF) within 10 days of a sinus ECG with similar (±10 ms) QRS durations, 2001–202144% female
- Participants
- 6,432 patients
- Comparison
- Convolutional neural network (CNN) applied to AF electrocardiograms (ECGs) vs AFQTc method for estimating sinus QTc
- Outcome
- Accuracy of predicting the software-generated sinus QTc value from an AF ECG
Performance
As reported in the abstract
Prediction of sinus QTc
| Test or model | Mean absolute error | Root mean squared error | Predictions within 1 SD (±30.6 ms) | Predictions within 2 SD (±61.2 ms) |
|---|---|---|---|---|
| CNN | 22.2 ms | 30.6 ms | 84% | 97% |
CNN
Mean absolute error22.2 ms
Root mean squared error30.6 ms
Predictions within 1 SD (±30.6 ms)84%
Predictions within 2 SD (±61.2 ms)97%
- CNN vs AFQTc: P < 0.0001 — CNN outperformed, narrower error ranges
Further findings
Ruling out QTc prolongation, by sex
- NPV, male
- 0.82
- NPV, female
- 0.92
- Specificity, male
- 0.92
- Specificity, female
- 0.97
- Model performed best for ruling out QTc prolongation
Model vs AFQTc method
- Model outperformed AFQTc, narrower error ranges (mean absolute error comparison)
- P < 0.0001
Predictions within 1 SD (±30.6 ms) and 2 SD (±61.2 ms) of sinus QTc
- Within 1 SD (±30.6 ms)
- 84%
- Within 2 SD (±61.2 ms)
- 97%
- Approximately
Original publication
Development of an AI-Driven QT Correction Algorithm for Patients in Atrial Fibrillation. JACC Clin Electrophysiol. Published 30 November 2022. doi:10.
Summarised from the published abstract; every number is checked against it