Performance of Pediatric Risk of Mortality Score Among Critically Ill Children With Heart Disease

Authors:
Publisher website
Google Scholar
Abstract
Objective: To evaluate the performance of the Pediatric Risk of Mortality 3 (PRISM-3) score in critically ill children with heart disease. Methods: Patients <18 years of age admitted with cardiac diagnoses (cardiac medical and cardiac surgical) to one of the participating pediatric intensive care units in the Virtual Pediatric Systems, LLC, database were included. Performance of PRISM-3 was evaluated with discrimination and calibration measures among both cardiac surgical and cardiac medical patients. Results: The study population consisted of 87,993 patients, of which 49% were cardiac medical patients (n = 43,545) and 51% were cardiac surgical patients (n = 44,448). The ability of PRISM-3 to distinguish survivors from nonsurvivors was acceptable for the entire cohort (c-statistic 0.86). However, PRISM-3 did not perform as well when stratified by varied severity of illness categories. Pediatric Risk of Mortality 3 underpredicted mortality among patients with lower severity of illness categories (quintiles 1-4) whereas it overpredicted mortality among patients with greatest severity of illness category (fifth quintile). When stratified by Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery (STS-EACTS) categories, PRISM-3 overpredicted mortality among the STS-EACTS mortality categories 1, 2, and 3 and underpredicted mortality among the STS-EACTS mortality categories 4 and 5. Pediatric Risk of Mortality 3 overpredicted mortality among centers with high cardiac surgery volume whereas it underpredicted mortality among centers with low cardiac surgery volume. Conclusion: Data from this large multicenter study do not support the use of PRISM-3 in cardiac surgical or cardiac medical patients. In this study, the ability of PRISM-3 to distinguish survivors from nonsurvivors was fair at best, and the accuracy with which it predicted death was poor.

This publication has 20 references indexed in Scilit: