Artificial intelligence and clinical stability after the Norwood operation

Background: Postoperative management of the neonate following the Norwood operation is among the most complex and challenging in pediatric critical care and high mortality remains. Artificial intelligence (AI) is poised to assist in monitoring of this complex population to improve clinical care, evaluation and outcomes. Methods: In a dedicated Pediatric Cardiac Intensive Care Unit in a quaternary Children’s Hospital, a convolutional neural network (CNN) model was developed and trained on electrocardiogram (ECG) waveforms from 45 neonates after the Norwood procedure. Waveforms from the first two postoperative days (critical) and the day prior to transfer from the intensive care unit (ICU) (stable) were used for training. The model was evaluated on a separate cohort of 10 neonates following the Norwood procedure. Models were compared to traditional machine learning algorithms on non-waveform data, and then combined in a final model. Retrospective clinical observation scoring was completed for comparison. Results: The CNN model yielded an area under the curve of the receiver operating characteristic (AUC-ROC) of 0.97 (±0.02). The final model combining the CNN, random forest (RF) on vital signs, and logistic regression achieved an AUC-ROC of 0.98 (±0.02) and an AUC of precision recall (AUC-PR) of 0.97 (±0.04) for distinguishing critical from stable. Clinical observations to assess patient stability agreed with the final model 78% of the time. This suggests that opportunities exist to improve the assessment of overall clinical state through the implementation of an AI based data monitoring tool. Conclusions: This novel, combined AI models can accurately detect changes in clinical status as patients progress from critically ill to stable following the Norwood procedure. This work provides the basis of a novel bedside monitoring tool and suggests new ways AI may influence clinical care beyond predicting deterioration events.