Artificial intelligence in airway management: a narrative review

Comprehensive Summary

This article summarizes the current evidence for machine learning and AI in intubation. The authors indicate that AI has the ability to more accurately predict difficult airways and even perform intubations with expert-level precision. This benefit stems from AI’s ability to outperform conventional bedside tests by analyzing voice recognition, facial profiles, and neck structures alongside other patient data. Consequently, the authors envision a future where physicians use smartphones pre-operatively to gauge airway difficulty, enhancing safety in the operating room. Another key focus of the article is AI’s role in emergency situations where stress might otherwise slow a provider's response. Robotic intubation systems have demonstrated accurate and precise performance while minimizing airway trauma, which could significantly improve outcomes in high-risk scenarios. However, important caveats remain, including high false-positive rates (false alarms) when detecting esophageal intubation and the potential for provider “deskilling.” Ultimately, machine learning should be viewed as a supplemental tool for airway management rather than a replacement for clinical providers.

Outcomes and Implications

This study outlines clear future directions based on the summarized evidence. AI continues to improve as a powerful educational tool, particularly for students learning airway management. It serves as a valuable resource for gauging accuracy and enhancing bedside tests, acting as a helpful adjunct for clinical training. However, significant hurdles remain, specifically the 50% specificity rate in detecting esophageal intubations (which causes false alarms) and the persistent limitations in positive predictive value for airway complexity. These factors justify the current hesitation in integrating AI into everyday clinical practice. Instead, its immediate utility may lie in preoperative preparation for potentially difficult airways. Ultimately, the authors indicate that several technical challenges still prevent AI from significantly transforming routine clinical outcomes at this stage.