Artificial intelligence in resuscitation: a scoping review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Cardiopulmonary Resuscitation Research

Background:

• Artificial intelligence (AI) is increasingly explored for improving outcomes in cardiac arrest (CA).
• The current landscape and specific applications of AI in resuscitation are not well-defined.
• Understanding AI's role in CA is crucial for advancing emergency medical care.

Purpose of the Study:

• To conduct a scoping review of AI applications in cardiac arrest and resuscitation.
• To identify existing research, methodologies, and outcomes associated with AI in resuscitation.
• To pinpoint research gaps for future investigation in AI-driven resuscitation strategies.

Main Methods:

• Systematic literature search using PubMed, EMBASE, and Cochrane databases.
• Adherence to PRISMA-ScR framework and ILCOR guidelines for comprehensive review.
• Screening and classification of AI applications by methodology, study design, outcomes, and setting, with AI-assisted data extraction validated manually.

Main Results:

• 197 studies were included, predominantly retrospective (90%), with limited prospective trials (16) and randomized controlled trials (2).
• AI was mainly used for CA prediction, rhythm classification, and post-resuscitation prognostication, with machine learning being the most common method (50%).
• High reported performance (AUROC > 0.85) was noted, but external validation and real-world implementation were scarce.

Conclusions:

• AI demonstrates potential in prediction and decision support for resuscitation, but evidence for improved patient outcomes is limited.
• Current clinical use of AI in resuscitation is not widespread.
• Future research should prioritize prospective validation, data equity, AI explainability, and clinical workflow integration.