Development of a Machine Learning Model for Predicting Weaning Outcomes Based Solely on Continuous Ventilator...

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Discontinuing mechanical ventilation is a complex clinical challenge.
• Accurate prediction of patient readiness for extubation is crucial for optimizing care and reducing complications.
• Current methods may not fully leverage continuous monitoring data available during spontaneous breathing trials (SBTs).

Purpose of the Study:

• To develop and validate a machine learning model for predicting mechanical ventilation weaning outcomes.
• To utilize continuous monitoring parameters from ventilators during SBTs as input for the predictive model.
• To assess the model's performance and provide interpretable insights for clinical decision-making.

Main Methods:

• A convolutional neural network (CNN)-based model was developed to analyze diverse-length data, including three waveforms and 25 numerical parameters from ventilators during SBTs.
• Data from 138 patients in a medical intensive care unit were used, with a random 8:2 split for training and testing.
• Gradient-weighted class activation mapping (Grad-CAM) was employed for model interpretability.

Main Results:

• The model achieved an area under the receiver operating characteristic curve (AUC-ROC) of 0.912 for predicting weaning success.
• The area under the precision-recall curve (AUC-PR) was 0.767, indicating robust performance.
• Grad-CAM highlighted influential features, providing visual explanations for the model's predictions.

Conclusions:

• The developed machine learning model accurately predicts mechanical ventilation weaning outcomes using readily available SBT data.
• The model offers an interpretable tool to assist clinicians in real-time extubation decisions.
• This approach has the potential to improve patient outcomes by facilitating timely ventilator discontinuation.