Comparative analysis: artificial intelligence vs. anesthesiologists in difficult airway identification

Abstract

Accurate identification of the difficult airway (VAD) represents a critical challenge in anesthesiology, due to its direct impact on patient safety during anesthetic induction. Traditional clinical tools, such as the Mallampati classification and the Cormack-Lehane scale, have limitations in sensitivity and specificity, which has motivated the exploration of artificial intelligence (AI) models as a diagnostic alternative. General objective: To evaluate the effectiveness of artificial intelligence models against the clinical evaluation of anesthesiologists who are experts in the identification of difficult airway (VAD). In adult patients scheduled for surgical procedures under general anesthesia at the La Trinidad Teaching Medical Center from August-October 2025 Materials and methods: An observational, descriptive, cross-sectional and prospective study was carried out at the La Trinidad Teaching Medical Center, with a sample of 95 patients. 285 Standardized images were captured using smartphones and processed with deep learning algorithms (Qwen 3.0). Conventional clinical tests were applied and diagnoses were compared by statistical analysis with SPSS and EPIDAT. Results: The AI models achieved a sensitivity of 85.29%, specificity of 87.10%, positive predictive value of 93.55% and an area under the curve (AUC) of 0.857. Concordance with the clinical assessment was acceptable (kappa index = 0.681), Conclusions: AI proves to be a reliable and accurate tool for the identification of VAD, with potential to improve patient safety, optimize anesthetic planning, and strengthen institutional innovation in resource-limited contexts.