The Future of Airway Management: A Comparative Analysis of Machine Learning and Ensemble Algorithms for Predicting Associated Factors of Difficult Intubation

Senobar  Naderian; Mojtaba  Mohammadzadeh Lame; Ata  Mahmoodpoor; Arezoo  Nejabatian; Kavous  Shahsavarinia; Maryam  Soleimanpour; Hassan  Soleimanpour

doi:10.18502/aacc.v11i5.19933

Senobar Naderian Student Research Committee, Tabriz University of Medical Sciences, Tabriz, Iran.
Mojtaba Mohammadzadeh Lame Department of Anesthesiology and Critical Care Medicine, Tabriz University of Medical Sciences, Tabriz, Iran.
Ata Mahmoodpoor Department of Anesthesiology and Critical Care Medicine, Tabriz University of Medical Sciences, Tabriz, Iran.
Arezoo Nejabatian Department of Emergency Medicine, Tabriz Medical Science, Islamic Azad University, Tabriz, Iran.
Kavous Shahsavarinia Research Center for Evidence-Based Medicine, Iranian EBM Centre: A JBI Centre of Excellence, Tabriz University of Medical Sciences, Tabriz, Iran.
Maryam Soleimanpour Clinical Research Development Unit of Tabriz Valiasr Hospital, Tabriz University of Medical Sciences, Tabriz, Iran.
Hassan Soleimanpour Medical Philosophy and History Research Center, Imam Reza General Hospital, Tabriz University of Medical Sciences, Tabriz, Iran.

DOI: https://doi.org/10.18502/aacc.v11i5.19933

Keywords: Difficult intubation; Machine learning; Ensemble learning; Mallampati score; Upper lip bite test; Feature-selection

Abstract

Background: This study aimed to evaluate the effectiveness of machine learning (ML) models in predicting difficult intubation among maxillofacial surgery patients by using clinical data from a previous study involving 132 patients. The study sought to enhance anesthesiologists' ability to identify patients at risk of difficult intubation, a critical concern in surgical settings.

Methods: The research applied various ML algorithms, including decision trees (DT), random forests (RF), Naive Bayes (NB), neural networks (NN), support vector machines (SVM), K-nearest neighbors (KNN), and ensemble voting methods, to the existing clinical dataset. This dataset contained a range of factors potentially associated with DI, such as the Mallampati score, Upper Lip Bite Test (ULBT) results, facial angle, and other relevant variables. A comprehensive approach was taken to explore the impact of different data preprocessing techniques, with a particular focus on feature selection and normalization methods.

Results: The study found that the combination of mutual information-based feature selection and robust scaler normalization consistently yielded high predictive accuracy. Notably, the decision tree algorithm achieved an accuracy of 0.84 and precision, sensitivity, and specificity scores of 0.95. The analysis also highlighted the strength of ensemble learning, which, by combining multiple classifiers, achieved an accuracy of 0.82. The results suggest that ML models, especially random forests and ensemble voting methods, can be highly accurate in predicting difficult intubation when trained on existing clinical data.

Conclusion: The research underscores the importance of data preprocessing in enhancing algorithmic performance, particularly the effectiveness of mutual information-based feature selection combined with robust scaler normalization. However, the study also indicates the need for further research to refine these models, ensuring their applicability and reliability in real-world clinical settings.