Prediction of Difficult Tracheal Intubation by Artificial Intelligence: A Prospective Observational Study

Aim: Many predictive clinical tests are used together for preoperative detection of patients with difficult airway risk. In this study, we aimed to predict difficult intubation with different artificial intelligence algorithms using various clinical tests and anthropometric measurements, besides, to evaluate the accuracy performance of Cormack and Lehane (C-L) classification with artificial intelligence. Material and Methods: This study was conducted as a single-blind prospective observational study between 2016 and 2019. A total of 1486 patients with American Society of Anesthesiologists physical status I-III, scheduled to undergo elective surgery and requiring endotracheal intubation, were included. Demographic variables, clinical tests and anthropometric measurements of the patients were recorded. Difficult intubation was evaluated using the 4-grade C-L system according to the easy and difficult intubation criteria. Difficult intubation was tried to predict using 16 different artificial intelligence algorithms. Results: The highest success rate among artificial intelligence algorithms was obtained by the RandomForest method. With this method, difficult intubation was predicted with 92.85% sensitivity, 96.94% specificity, 93.69% positive predictive value and 96.52% negative predictive value. C-L classification accuracy performance also determined as 95.60%. Conclusion: Artificial intelligence has been considerably successful in predicting difficult intubation. Besides, C-L classifications of easy and difficult intubated patients were successfully predicted with artificial intelligence algorithms. Using a 6-grade modified C-L classification for laryngeal view may provide stronger difficult intubation prediction. A safer and more potent prediction in training artificial intelligence can be achieved by adding individual differences and clinical features that support the definition of difficult intubation.