Unravelling intubation challenges: a machine learning approach incorporating multiple predictive parameters.

Area of Science:

• Anesthesiology and Medical Informatics

Background:

• Difficult airway management is a critical patient safety concern during anesthesia requiring careful planning.
• Machine learning (ML) tools are increasingly adopted in medicine, often outperforming traditional methods.
• This study investigates ML techniques for predicting challenging airway management.

Purpose of the Study:

• To apply machine learning algorithms to identify predictive parameters for difficult airway management.
• To evaluate the performance of various ML models in forecasting airway challenges.

Main Methods:

• A cross-sectional study analyzed 622 patient records from two hospitals.
• Feature importance and undersampling techniques (SMOTE, edited nearest neighbor) were used for data balancing and feature selection.
• Seven ML algorithms, including K-Nearest Neighbors (KNN), were assessed using 10-fold cross-validation and standard performance metrics.

Main Results:

• Twenty-four significant features were identified from the initial 32.
• Undersampling methods outperformed SMOTE for data balancing.
• KNN demonstrated superior performance compared to other algorithms, achieving high accuracy (0.87) and AUC (0.87).

Conclusions:

• Machine learning algorithms offer valuable insights for predicting difficult airway management.
• Accurate forecasting of airway difficulties using ML can enhance clinical practice and improve patient outcomes.