ABSTRACT
The purpose of this study was to investigate the performance of the Pediatric Risk of Mortality (PRISM III–24) in a Greek pediatric intensive care unit (PICU). We prospectively followed 300 PICU patients in an observational cohort study. PRISM III-24 performance was assessed in the whole population and in 4 preselected groups (infants, patients with length of PICU stay > 4 days, patients with co-morbidities, ventilated patients) with standard discrimination and calibration methods. Efficiency (on admission) was defined as admission of patients with mortality risk >1%, while effectiveness was determined as standardized mortality ratio (SMR). The area under the receiver operating curve (AUC) showed good discrimination of PRISM III-24 score (AUC 0.892, 95%CI 0.821-0.963) and PRISM III-24 predictive model (AUC 0.900, 95%CI 0.836-0.964). Hosmer-Lemeshow goodness-of-fit test showed good calibration of PRISM III-24 score (χ2 (8) =1.716, p = =0.989) and PRISM III-24 predictive model (χ2 (8) = 8.294, p = 0.405). Good performance was also found in the 4 preselected groups (AUC always>0.835, p values of the goodness-of-fit test always > 0.434). Efficiency was 76.7%, predicted PICU mortality was 11.1%, observed PICU mortality at discharge was 9.7%, and the SMR was 0.997 (95% CI 0.67-1.43). PRISM III-24 performed well in our population showing high discrimination and calibration capabilities. Mortality was higher than in relevant studies, probably due to case mix, patient characteristics and the distinct PICU policy of our country. However, efficiency and effectiveness we-re met by an international standard.