Introduction

Out-of-hospital cardiac arrest (OHCA) survival rates remain critically low despite decades of research, with cardiogenic causes representing the highest mortality burden. To address limitations in existing prediction tools, we developed two novel artificial neural network (ANN)-based predictive scores for return of spontaneous circulation (ROSC) and survival-to-discharge specifically in cardiogenic OHCA. This study directly compared their performance against the established Utstein comparator group—the current benchmark for OHCA outcome prediction.

Methods

Using prospectively collected data from a EuReCa_One-aligned OHCA registry (October 2014–September 2023; n=3,369 confirmed cardiogenic cases with initiated CPR), multilayer perceptron ANN modeling identified key predictive factors through iterative feature importance analysis. Scores (0–10 points) were weighted by factor contribution and categorized into low (0-3), intermediate (4-6), and high (7-10) probability tiers. Validation occurred in a temporally distinct, propensity-matched cohort (October 2023–December 2024; n=628) controlling for age, bystander involvement, and arrest location. The Utstein comparator group (witnessed arrest + bystander CPR + initial shockable rhythm) served as the primary benchmark for performance comparison using multivariate logistic regression.

Results

In the derivation cohort, ROSC occurred in 23.4% (790/3,369) and survival-to-discharge occurred in 5.0% (170/3,369). The ANN-derived scores demonstrated high discriminative accuracy: ROSC score OR 2.741 (95% CI 2.017–3.724; p<0.001) and survival score OR 4.850 (95% CI 2.083–11.294; p<0.001). Compared to the Utstein predictor, the ANN ROSC score showed non-inferior performance (Utstein OR 2.841, 95% CI 2.819–4.438; p<0.001). Critically, the ANN survival-to-discharge score significantly outperformed the Utstein predictor, which demonstrated no statistical significance for survival prediction (Utstein OR 1.610, 95% CI 0.647–4.010; p=0.306), indicating limited clinical utility for this endpoint.

Conclusion

ANN-generated scores provide clinically actionable, granular prediction of cardiogenic OHCA outcomes. While comparable to Utstein for ROSC prediction, the survival-to-discharge score offers substantially superior discriminatory power (4.85 vs. 1.61 OR), enabling more precise early field triage and resource allocation. Future implementation studies should validate these tools in prehospital settings.