Authors: Mohamed F. Abosamak et al
Anaesthesia Critical Care & Pain Medicine, 2025. DOI: 10.1016/j.accpm.2025.101589
This systematic review and meta-analysis evaluated the performance of artificial intelligence (AI) models in predicting difficult airways—one of anesthesia’s most critical and high-stakes challenges. Difficult airway complications can rapidly lead to hypoxia, brain injury, or death, and early prediction is essential for patient safety. Although AI tools have been developed to support clinical decision-making in multiple perioperative contexts, their accuracy and consistency in airway prediction have not been well quantified.
Researchers systematically searched PubMed and ScienceDirect for studies published up to March 2025 that developed or validated AI models for difficult airway prediction. Thirteen studies met inclusion criteria, including 11 involving surgical patients under general anesthesia and 2 involving emergency department populations. Using meta-analysis in R (version 4.4.2), the authors pooled model performance based on the area under the receiver operating characteristic curve (AUC or AUROC).
Among deep learning methods, the VGG (Visual Geometry Group) convolutional neural network demonstrated the best discriminative performance, with an AUC of 0.84 (95% CI, 0.83–0.84; I² = 0%). Traditional machine learning algorithms such as support vector machines (SVM, AUC 0.80; 95% CI, 0.65–0.96) and naïve Bayes classifiers (NB, AUC 0.81; 95% CI, 0.51–1.10) also performed reasonably well but showed high heterogeneity across studies (I² > 99%). Despite promising results, most existing models achieved only moderate predictive accuracy (AUC < 0.80) and lacked consistent external validation.
The authors conclude that while AI models show encouraging potential for identifying difficult airways preoperatively, current algorithms are limited by variable datasets, inconsistent definitions, and poor generalizability. Future research should focus on developing standardized, multicenter datasets and externally validated models to ensure safe, reliable deployment in clinical anesthesia practice.
What You Should Know
-
AI models can help predict difficult airways but vary widely in performance.
-
Deep learning (VGG) achieved the highest accuracy (AUC 0.84) among existing models.
-
Traditional machine learning methods such as SVM and NB showed moderate performance but high heterogeneity.
-
There is a need for larger, multicenter validation and standardized training datasets before clinical adoption.
Thank you to Anaesthesia Critical Care & Pain Medicine for publishing this timely meta-analysis emphasizing the emerging but still evolving role of AI in difficult airway prediction.