Machine learning modelling for predicting the utilization of invasive and non-invasive ventilation throughout the ICU...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Mechanical ventilation is a critical intervention for intensive care unit (ICU) patients.
• Predicting the need for ventilation is essential for timely and appropriate patient management.
• Existing methods for prediction may lack precision or generalizability.

Purpose of the Study:

• To develop and validate a Machine Learning (ML) model for predicting the requirement of invasive and non-invasive mechanical ventilation in ICU patients.
• To assess the model's performance using a large, multi-center dataset and an external validation set.
• To identify key clinical factors associated with the need for mechanical ventilation.

Main Methods:

• Analysis of 2.6 million patient records from the Philips eICU Research Institute (ERI) database (2010-2019).
• Random data splitting into training (63%), validation (27%), and test (10%) sets, with an additional external test set for generalizability assessment.
• Model performance evaluation using Area Under the Curve (AUC) by comparing predicted probabilities with actual ventilation use.

Main Results:

• The ML model achieved high prediction performance: AUC of 0.921 for overall ventilation, 0.937 for invasive, and 0.827 for non-invasive ventilation.
• Identified factors associated with lower ventilation likelihood include high Glasgow Coma Scores, younger age, lower Body Mass Index (BMI), and lower partial pressure of carbon dioxide (PaCO2).
• The model demonstrated generalizability on an external test set.

Conclusions:

• The developed ML model is a robust tool for predicting mechanical ventilation needs in ICU patients.
• This model can serve as a retrospective benchmarking tool for hospitals to evaluate their ventilation strategies.
• Future applications may include clinical decision support for optimizing mechanical ventilation management in ICUs.