Prediction of Ideal Depth in Neonatal Endotracheal Intubation

Accurate placement of Endotracheal Tubes (ETT) is critical for effective ventilation in neonates. Traditional formulae based solely on age or weight can be imprecise. This study investigates the suitability of various machine learning algorithms for predicting ideal ETT length based on patient weight and NTL (Nasotracheal Tip Location). We compared the performance of Linear Regression, Decision Trees, KNN (K-Nearest Neighbors), Random Forests, and lastly support vector regression (SVM) using the R statistical environment with the tidymodels and tidyverse packages. The coefficient of determination and the root-mean-squared error from cross-validation are used as performance measures. We find that KNN slightly outperforms the other machine-learning models and that the overall performance of all models was quite satisfactory. This approach has the potential to improve patient care by ensuring proper ETT placement and minimizing complications. Two approaches for explainable AI models, Shapley values and variable-importance analysis, are used to understand the contribution of the predictors when estimating the ideal ETT length.