Abstract
Introduction: Comparison of risk-adjusted outcomes is imperative in the evaluation of healthcare quality. Risk-adjustment for children undergoing cardiac surgery poses unique challenges and present risk models are not adequate. We developed a risk-adjustment tool for ICU mortality in the pediatric cardiac surgical population: Pediatric Index of Cardiac Surgical Intensive care Mortality (PICSIM)Score. Methods: We used 16,574 cardiac surgical patients from 55 pediatric ICUs in the VPS database (VPS, LLC, Alexandria, VA), a national pediatric critical care database. Thirty physiologic and diagnostic variables used for PIM2 and PRISM3 risk of mortality scores were combined with additional patient specific variables. Multivariate logistic regression with stepwise selection was used to develop a model. The model was developed with 75% of the data (training set), and tested with the remaining 25% (testing set). Results: 13 variables remained in the final model. Area under the Curve (AUC) was calculated using PICSIM as well as PIM2 and PRISM3 risk of mortality scores to compare performance. Two important variables were included: 1) The STAT Mortality Score, developed by the Society for Thoracic Surgeons (STS) and the European Association for Cardiothoracic Surgery (EACTS); 2) A unique variable indicating if the surgical procedure occurred prior to or after ICU admission. In the testing set PICSIM (AUC = 0.87) performed better than PIM2 (AUC=0.81) and PRISM3 (AUC=0.83). Conclusions: Using ICU admissions data, we developed the PICSIM mortality score, consisting of 13 risk variables capturing physiology, cardiovascular condition, and admission time data. PICSIM, showed better discrimination than PIM2 and PRISM3 for critical care mortality in a multi-site cohort of pediatric cardiac surgical patients. PICSIM was able to predict mortality whether or not the patient was admitted prior to or following a cardiac surgical procedure.