Key result
An automated machine learning prediction model based on the Utstein model accurately predicted CPR outcomes in OHCA patients, achieving a ROC-AUC of 0.9891.
Why the study?
The study aimed to construct an automated machine learning-based prediction model using the Utstein model for CPR outcomes in community OHCA patients to identify key predictive factors and support clinical risk stratification.
Does an AutoML-based prediction model using Utstein model variables accurately predict 30-day survival in community out-of-hospital cardiac arrest patients?
Observational (n=370)
No
Does an AutoML-based prediction model using Utstein model variables accurately predict 30-day survival in community out-of-hospital cardiac arrest patients?
An AutoML-based prediction model utilizing Utstein variables accurately predicts 30-day survival in OHCA patients, identifying on-site defibrillation and time to CPR as critical prognostic factors.
To construct an automated machine learning (AutoML)-based prediction model for cardiopulmonary resuscitation (CPR) outcomes in community out-of-hospital cardiac arrest (OHCA) patients using the Utstein model, systematically identify key predictive factors, and provide a basis for clinical risk stratification and individualized intervention. A single-center retrospective study design was employed, including all adult non-traumatic OHCA patients managed by the Emergency Department of The First People’s Hospital of Lin’an District, Hangzhou, from March 2019 to March 2024. All OHCA patients were analyzed to identify key factors associated with achieving return of spontaneous circulation (ROSC). In the second stage, a sub-cohort of patients who successfully achieved ROSC was selected to further investigate predictors influencing their 30-day survival status, ultimately developing the final prognostic prediction model. The Improved Fast Tree-seed Algorithm (IFATA) was utilized within the AutoML framework to optimize model hyperparameters. Feature importance was validated through dual-dimensional LASSO regression and SHAP interpretability models. An interactive web-based clinical decision support tool was developed for risk assessment and clinical communication. Among 370 enrolled patients, 256 (69.19%) died after resuscitation, while 114 (30.81%) achieved ROSC. The AutoML model demonstrated comprehensive optimal performance, with a ROC-AUC of 0.9891 and PR-AUC of 0.9490. Key features selected by AutoML were: on-site defibrillation, cumulative epinephrine dosage, time from cardiac arrest (CA) to CPR initiation, time from CA to endotracheal intubation, PCO2, and time from CA to intravenous access establishment. When evaluated on the test set, AutoML exhibited the strongest robustness, achieving a ROC-AUC of 0.9815 and PR-AUC of 0.9515. Its predictive calibration significantly outperformed other models, with the lowest Brier score (0.127). SHAP analysis ranked feature importance as follows: on-site defibrillation > cumulative epinephrine dosage > time from CA to CPR initiation > time from CA to endotracheal intubation > PCO2 > time from CA to intravenous access establishment. By integrating multidimensional clinical data with interpretable machine learning techniques, this study confirms the core role of the Utstein model in predicting CPR outcomes for OHCA patients. It achieves real-time translation of risk assessment into intervention decisions, providing innovative tools and methodologies for precise OHCA resuscitation.
No takes yet. Share an insight, caveat, or question.
Lei et al. (2026) conducted an observational in out-of-hospital cardiac arrest (OHCA) (n=370). AutoML-based prediction model vs. other models was evaluated on prediction of CPR outcomes (ROSC and 30-day survival) (ROC-AUC 0.9891). An automated machine learning prediction model based on the Utstein model accurately predicted CPR outcomes in OHCA patients, achieving a ROC-AUC of 0.9891.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: