Artificial intelligence is rapidly moving into one of medicine’s most critical areas — cardiac arrest care, with researchers reporting that AI tools may soon help doctors predict cardiac arrests, guide CPR decisions, improve emergency response systems, and support recovery planning after resuscitation.
A new review published in the World Journal of Emergency Medicine examined how artificial intelligence is being applied across the full chain of cardiac arrest care, from early risk detection to post-resuscitation recovery.
Researchers from Sun Yat-sen University and Sun Yat-sen Memorial Hospital analysed 114 studies and evaluated 92 different AI models after screening more than 2,100 scientific records. The team reviewed research covering both in-hospital and out-of-hospital cardiac arrest scenarios.
Cardiac arrest remains one of the deadliest medical emergencies worldwide, where even a few minutes of delay can significantly reduce survival chances and increase the risk of permanent brain injury. Despite advances in CPR, emergency response, and intensive care, survival rates remain low, particularly in out-of-hospital emergencies.
According to the review, AI systems showed particularly strong performance in predicting cardiac arrest risk before it occurs. In hospital settings, one multilayer perceptron model achieved an exceptionally high predictive accuracy with an area under the receiver operating characteristic curve (AUROC) of 0.998. For out-of-hospital cardiac arrest prediction, machine learning models such as extreme gradient boosting and random forest algorithms achieved AUROC values of 0.950.
Researchers also found promising results in AI-assisted CPR decision support. Convolutional neural network models demonstrated strong performance in analysing resuscitation-related decisions, while other AI systems showed potential in predicting neurological recovery and survival outcomes after cardiac arrest.
The review highlighted several emerging applications of AI in emergency medicine, including wearable-device-based cardiac arrest detection, AI-supported emergency call recognition, automated external defibrillator localisation, and the use of large language models similar to generative AI systems for medical education and clinical support.
Researchers noted that AI is no longer being explored for just one stage of cardiac arrest treatment. Instead, it is increasingly being integrated across the entire care pathway — from identifying warning signs before collapse to assisting doctors during resuscitation and supporting rehabilitation after recovery.
However, the authors cautioned that significant challenges remain before AI becomes a routine part of emergency care. Many existing models still lack large-scale real-world validation, and concerns remain regarding data quality, privacy protection, algorithmic bias, and unequal access to advanced healthcare infrastructure.
The researchers emphasised that future studies should focus on prospective clinical trials, explainable AI systems, and ensuring these technologies improve real patient outcomes rather than only achieving high performance scores in laboratory testing.
