Can Machine Learning Methods Predict Extubation Outcome in Premature Infants as well as Clinicians?

Area of Science:

• Neonatal Medicine
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Mechanical ventilation is common in premature infants, but predicting extubation success remains a significant clinical challenge.
• Despite advances, extubation failure rates in premature infants have not decreased, highlighting the need for improved predictive tools.

Purpose of the Study:

• To develop a machine learning-based decision-support tool for predicting extubation outcomes in premature infants.
• Evaluate the efficacy of various machine learning algorithms in forecasting extubation success.

Main Methods:

• Utilized a dataset of 486 premature infants on mechanical ventilation.
• Developed and compared predictive models using artificial neural networks (ANN), support vector machine (SVM), naïve Bayesian classifier (NBC), boosted decision trees (BDT), and multivariable logistic regression (MLR).
• Model performance was assessed using the area under the curve (AUC).

Main Results:

• Artificial neural networks, multivariable logistic regression, and naïve Bayesian classifiers demonstrated satisfactory performance with AUCs ranging from 0.63 to 0.76.
• Support vector machine and boosted decision trees exhibited poor predictive performance, with AUCs around 0.5.

Conclusions:

• Current machine learning models do not consistently outperform clinician predictions for extubation outcome in premature infants.
• The complexity of clinical data and uncaptured contextual information limit the performance of current machine learning algorithms.
• Future research incorporating data preprocessing steps may enhance the predictive accuracy of these models.