Exploiting Machine Learning Algorithms and Methods for the Prediction of Agitated Delirium After Cardiac Surgery:...

Area of Science:

• Medical Informatics
• Computational Medicine
• Surgical Outcomes Research

Background:

• Delirium is a temporary mental disorder affecting surgical patients, particularly after cardiac surgery.
• It is linked to adverse events, increased costs, and poor patient outcomes, including cognitive impairment, stroke, and death.
• Early identification of at-risk patients is crucial for implementing preventive interventions.

Purpose of the Study:

• To explore and compare the performance of various machine learning (ML) predictive models for preemptively identifying delirium in cardiac surgery patients.
• To assess the efficacy of different ML algorithms in predicting postoperative delirium.

Main Methods:

• Utilized a clinical dataset of over 5000 cardiac surgery patients.
• Developed predictive models using logistic regression, artificial neural networks (ANN), support vector machines (SVM), Bayesian belief networks (BBN), naïve Bayesian, random forest, and decision trees.
• Addressed class imbalance in the training dataset using random undersampling and validated performance on a separate test set.

Main Results:

• Support Vector Machines (SVM) demonstrated the best F1 score (40.2%), kappa (29.3%), and positive predictive value (29.7%) for the delirium class.
• Artificial Neural Networks (ANN) achieved the highest receiver-operator area under the curve (78.2%).
• Bayesian Belief Networks (BBN) showed the best precision-recall area under the curve (30.4%) for detecting positive cases.

Conclusions:

• Addressing class imbalance is vital for enhancing ML model performance in predicting postoperative delirium.
• Machine learning methods can uncover hidden patterns in complex, multifactorial conditions like delirium, potentially improving prediction accuracy.
• Proactive identification of at-risk patients through ML can lead to cost reduction and optimized patient outcomes by preventing complications.