Classifying Infection Risk Following Pediatric Cardiac Surgery

Abstract

Pediatric cardiac surgeries are commonly performed procedures. Postoperative infections complicate as many as 30% of procedures, increasing cost, length of stay, morbidity, and possibly mortality. If a patient’s risk of infection could be determined prospectively, targeted interventions could mitigate the risk. In this work, we developed a cohort of pediatric cardiac surgery patients using the Research Derivative, and manually validated the infection outcomes. We used this cohort to train and validate logistic regression and machine learning models to classify the risk of postoperative infections in this group, and derived an easily scorable risk classification rule for bedside use. The logistic regression model for a composite outcome achieved an AUC of 0.853, and the bedside rule achieved an AUC of 0.753.