Postoperative Delirium Prediction through Machine Learning in Patients Undergoing Aortocoronary Bypass Surgery

Objectives: Postoperative delirium is a prevalent and disabling mental disorder that occurs regularly in patients undergoing cardiac surgery. Delirium is associated with increased morbidity and mortality as well as a prolonged hospital stay. Identifying patients likely to develop delirium postextubation at the earliest possible point in time may help guide treatment decisions. This study investigates whether machine learning models are feasible to predict postoperative delirium.

Methods: In this prospective monocentric clinical trial, we included 256 patients who underwent coronary artery bypass grafting (CABG) with or without cardiopulmonary bypass (CPB) between 2017 and 2018. Demographic, clinical, and hemodynamic data until extubation were used to develop a machine learning model. The model used a support vector and naive-Bayes backbone with fivefold cross-validation and was trained on 70% of the data. The remaining data were spared for testing purposes. Patients were considered having delirium if either the 4AT or Confusion Assessment Method (CAM)-ICU test conducted within 7 days after extubation was positive and neuroleptic or sedative medication was required. Model evaluation was conducted using accuracy after accounting for class imbalance.

Result: A total of 256 patients were enrolled, of which 53 patients developed delirium as defined. Patients with delirium were significantly older (72.3 ± 7.9 vs. 66.4 ± 9.3 years, p = 0.002) and had a worse logistic EuroSCORE (p < 0.001). In addition, the duration of ventilation was significantly longer (19.8 ± 14.4 vs. 13.0 ± 6.3 hours, p = 0.003) and these patients showed significant amount of fluid overload (p = 0.016). Due to the class imbalance, we applied a weighted oversampling approach to account for the considerably lower fraction of patients with delirium (20.7%). The machine learning algorithm could identify patients with delirium at the time of extubation with an accuracy of more than 90%. Comparing our training and testing performance there was no indication for model overfitting. The computing time of the trained model was no longer than 1.3 seconds per patient.

Conclusion: Our results indicate that our machine learning algorithm was feasible and useful for predicting treatment-relevant delirium at the time of extubation in patients undergoing CABG with CPB. To further test our model, validation on a different cohort will be performed.

Publication History

Article published online:
19 February 2021