Cardiac arrest & resuscitation
Deep learning-based return of spontaneous circulation (ROSC) prediction and electrocardiogram (ECG) phenotyping in out-of-hospital cardiac arrest
Authors' conclusion
The CNN-based model accurately predicted ROSC and shockable rhythm from ECG signals and identified five novel ECG phenotypes in OHCA.
Study context
- Population
- Adults with nontraumatic out-of-hospital cardiac arrest (OHCA) with prehospital ECG signals, Korean OHCA Registry and Seoul EMS ECG data, July 2021–December 2023
- Participants
- 3,452 adults
- Outcome
- Prediction of ROSC within 2 min (AUC)
- Follow-up
- Within 2 min
Methodological detailsdeep learning (one-dimensional convolutional neural network (CNN)), K-means clustering, registry-based
- Deep learning (one-dimensional CNN)
- K-means clustering
- Registry-based
Main finding
As reported in the abstract
Prediction of ROSC within 2 min (test set)
AUC for predicting ROSC within 2 min in the test set
0.921
95% CI 0.897–0.945
Observational data: associations, not effects.
Further findings
Classification of shockable rhythm (test set)
- AUC for shockable rhythm in the test set
- 0.983 (95% CI 0.979-0.988)
ECG phenotypes (K-means clustering) and ROSC probability
- Phenotype 1: shockable rhythms, high ROSC probability
- 30.4%
- Phenotype 2: shockable rhythms, low ROSC probability
- 4.8%
- Phenotype 3: pulseless electrical activity, relatively high ROSC probability
- 5.2%
- Phenotype 4: pulseless electrical activity, lower ROSC probability
- 0.5%
- Five ECG phenotype categories with varying morphologies and ROSC probabilities identified
- Phenotype 5: primarily asystole or near-asystole rhythms
- Transitions between ECG phenotypes associated with CPR quality
Original publication
Deep learning-based ROSC prediction and ECG phenotyping in out-of-hospital cardiac arrest. Resuscitation. Published 6 March 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →