Machine learning and regular logistic regression predictive models for prediction of difficult laryngoscopies: a prospective cohort

Abstract

Background: Machine learning (ML) algorithms have been deemed to improve predictive performance as compared to regular logistic regression (LR) models. We compared the AUC for multiple predictive models including ML algorithms and LR for prediction of difficult laryngoscopies. Methods: A prospective cohort was conducted with patients undergoing general anesthesia for surgical procedures. We preoperatively collected data on sex, age, weight, height, ASA physical status, modified Mallampati test (MMT), mouth opening (MO), and sternomental distance (SMD). The main outcome was difficult laryngoscopies defined as Cormack and Lehane's classes 3 or 4. Uni and multivariable analyses were performed to evaluate association between variables and to build the predictive models. Only variables with p-values<0.2 were included into the models. Results: From a total of 453 patients, 29 (6.4%) presented difficult laryngoscopies. Sex (p=0.015), weight (p=0.009), height (p=0.001), MMT (p<0.001), MO (p<0.001), and SMD (p<0.001) were associated with difficult laryngoscopies. Age (p=0.212) and ASA (p=0.070) did not present significant difference between the 2 groups. Six predictive models were built including sex, weight, height, ASA, MT, MO, and SMD. The ROC curves for the predictive models are presented in Figure 1 and their AUC were as follows: Logistic Regression model - LR (82.58%; 95% CI: 67.25-97.91%); Linear Discriminant Analysis - LDA (82.58%; 95% CI: 70.7-94.46%); Classification and Regression Trees - CART (58.32%; 95% CI: 41.89-74.74%); k-Nearest Neighbors - kNN (58.32%; 95% CI: 37.53-79.11%); Random Forest - RF (69.05%; 95% CI: 44.22-93.87%); Generalized Boosted Model - GBM (84.35%; 95% CI: 73.85- 94.85%). Conclusions: ML algorithms did not improve airway prediction as compared to regular LR model.