A Machine-Learning Approach for Decision Support and Risk Stratification Of Pediatric Perioperative Patients Based on the APRICOT Dataset

Abstract

Background: Anesthesia has evolved to provide a high level of safety for children, yet a small chance of a serious perioperative complication remains, even in those considered at low risk. In practice, prediction of at-risk patients currently relies on the American Society of Anesthesiologists (ASA) Physical Status score, despite reported inaccuracies with this method. Methods: In this study, we used machine learning to develop predictive models that can classify children as low risk for anesthesia at the point of surgical booking and on the day of the procedure after anesthetic assessment. This information can be used to support decision-making in the safe allocation of patients to non-tertiary care locations and providers, enabling the right care to be delivered in the right place on a personalized basis. Findings: Our selected models achieved accuracies of >0.9, areas under the receiver operating curve of 0.6-0.7, and negative predictive values greater than 96%. Gradient boosting models were the best performing for both the booking phase and the day-of-surgery phase. Interpretation: By using our models with the feature values available for an individual patient when the model has utility in the clinical workflow, the perioperative team can obtain a tailored point-of-care risk assessment to support decision making. Model use illustrates that accurate, clinically meaningful prediction of patients at low risk of critical perioperative adverse events is possible on an individual, rather than population-based, level. Funding: This study was funded internally by the Department of Anesthesia and Pain Medicine at Johns Hopkins All Children's Hospital. Declaration of Interest: None to declare. Ethical Approval: Ethical approval was obtained by all participating centers.