Pre-operative Characteristics for Prediction of Supraglottic Airway Failure Using Machine Learning (ERICA)

Overview

Supraglottic airway devices (SGA) are a safe and well-established technique for airway management. Nowadays, up to 60% of general anaesthetics performed in European countries use SGA. In 0.2-4.7% SGA fail and require conversion to tracheal tubes. The ERICA study will use artificial intelligence methods to develop a model that can predict the risk of an unplanned SGA conversion based on pre-operative characteristics available during the premedication visit.

An intraoperative change of procedure not only leads to time delays but also time delays, but also involves measures that are stressful for the patient, such as deepening the anaesthesia and manipulating the airway again. Therefore, the objective of ERICA is to develop a machine learning algorithm based on preoperative information 1) that can accurately predict the risk of an unplanned SGA conversion and 2) identifies characteristics leading to conversion from SGA to tracheal tube. I. Developing the model • The final dataset will be split in a training, testing, and validation cohort. Five models will be created to predict intraoperative conversion from SGA to tracheal tube including generalized linear models (GLM), deep learning, distributed random forest (DRF), xgboost and gradient boosting machine (GBM). Then, a stacked ensemble model will be constructed through combination of the five models. Finally, the best artificial intelligence model will be chosen. II. Identify characteristics leading to the airway conversion and categorisation. * Intraoperative changes of the patient's position can alter the risk of conversion, therefore operations with positional changes should be considered * Identify patient- and procedure-dependent characteristics that lead to conversion from SGA to tracheal tube and their importance.

Conditions