Artificial Intelligence in Intensive Care: An Overview of Systematic Reviews with Clinical Maturity and Readiness...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Intensive care unit (ICU) care is time-critical and data-intensive, presenting high-risk scenarios for AI decision support tools.
• Existing AI evidence in ICUs is fragmented due to limited external validation, inconsistent reporting, and scarce real-world impact data.
• Key translational gaps were identified by mapping five ICU domains and assessing clinical and implementation maturity.

Purpose of the Study:

• To conduct a systematic overview of reviews on AI in ICUs.
• To assess the clinical and implementation maturity of AI tools across five predefined ICU domains.
• To identify translational gaps hindering the safe deployment of AI in ICUs.

Main Methods:

• A PRIOR-aligned overview of systematic reviews was performed, searching PubMed, Embase, and Web of Science.
• Data extraction and risk of bias assessment (ROBIS) were conducted by two independent reviewers.
• Findings were synthesized narratively, prioritizing AUROC ranges, without meta-analysis.

Main Results:

• 34 systematic reviews (2017-2025) covering prognostic and early warning AI applications in adult ICUs were included.
• Reporting predominantly focused on discrimination (AUROC ranges 0.54-0.99), with limited data on calibration and clinical utility.
• AI maturity signals were low-to-intermediate, with no evidence of routine or regulated clinical decision support (CDS) deployment.

Conclusions:

• A significant translational gap exists between AI's retrospective performance and its clinical maturity for safe ICU deployment.
• Priorities include external validation, prospective impact studies, standardized reporting (including calibration), and governance-focused implementation strategies.
• Addressing these gaps is crucial for advancing AI in critical care settings.