Zeitschrift für Neonatalbiologie

Abstrakt

Can machine learning methods predict the outcome of extubation in premature infants as well as clinicians?

Martina Mueller, Jonas S. Almeida, Romesh Stanislaus and Carol L. Wagner

Rationale: Although management of respiratory failure in preterm infants using a mechanical ventilator has made great progress in the last decades, predicting extubation outcome at a given time point remains challenging. Numerous studies have been conducted to identify predictors of extubation outcome, but the rate of failed extubation attempts has not decreased. Objective: To develop a decision support tool to predict extubation outcome in preterm infants using a set of machine learning algorithms. Mechanical: Using a dataset of 486 ventilated preterm infants, predictive models were developed using machine learning algorithms such as artificial neural networks (ANN), support vector machine (SVM), naive Bayesian classifier (NBC), boosted decision trees (BDT), and multivariable logistic regression (MLR). The performance of all models was evaluated using area under the curve (AUC). Results For some models (KNN, MLR and NBC) the results were satisfactory (AUC: 0.63–0.76); however, two algorithms (SVM and BDT) showed poor performance with AUCs of ~0.5. Conclusion: Clinician predictions still outperform machine learning due to the complexity of the data and the contextual information that may not be captured in the clinical data used as input for developing the machine learning algorithms. Incorporating preprocessing steps in future studies may improve the performance of the predictive models.