Acute cardiac care & resuscitation
How much information is needed to predict outcomes after cardiac arrest?
Authors' conclusion
Multiple data sources were needed to achieve optimal sensitivity.
Study context
- Population
- Consecutive adults unresponsive after cardiac arrest, single hospital, January 2010–February 2022773 (70 %) poor outcome
- Participants
- 1,106 participants
- Exposure
- Machine learning (ML) models using combinations of registry, electronic health record (EHR) and EEG data
- Outcome
- Sensitivity predicting poor outcome (discharge Cerebral Performance Category (CPC) 4 or 5) at perfect specificity (zero false positives)
- Follow-up
- Presentation and 12, 24, 48 and 72 h post-arrest
Methodological detailssingle-centre, machine learning prognostic models
- Single-centre
- Machine learning prognostic models
Main finding
As reported in the abstract
Sensitivity for poor outcome at perfect specificity · 24 h
Peak median sensitivity at perfect specificity (24 h)
70 %
65–73 %
Observational data: associations, not effects.
Further findings
Discrimination of best model (24 h)
- Median AUC
- 0.949 [0.947-0.951]
Original publication
How much information is needed to predict outcomes after cardiac arrest? Resuscitation. Published 15 August 2025. doi:10.
Summarised from the published abstract; every number is checked against it