External validation of the paediatric logistic organ dysfunction score

Abstract

External validation of the paediatric logistic organ dysfunction (PELOD) score in two paediatric intensive care units (PICU) in South America. Prospective observational cohort study including all PICU admissions from July 2003 to December 2004 in Porto Alegre, Brazil, and from January 2004 to December 2004 in Buenos Aires, Argentina. The data collected included demographic variables, diagnosis, need for mechanical ventilation, length of PICU stay and mortality, and the 12 variables in the PELOD score. For each PELOD score variable, the worst daily value and the worst value of the whole PICU stay were used for the daily PELOD (dPELOD) and PELOD scores, respectively. A total of 1,476 admissions (51.3% from Argentina and 48.7% from Brazil) were analysed. Observed and predicted mortality were, respectively, 4.7% and 6.6%, with a standardized mortality ratio of 0.72. The score showed excellent discrimination capacity, with an area under the receiver operator characteristic (ROC) curve of 0.93 (0.88-0.98). The dPELOD score on days 1-5 also showed good discrimination capacities, with areas under the ROC curve >0.85. However, PELOD and dPELOD scores showed poor calibration with the Hosmer-Lemeshow test (chi-square 72.3, p < 0.001). This poor calibration was explained by a deficiency in the PELOD score where it fails to identify two risk intervals; 3.1-16.2% and 40-80%. The PELOD score is reproducible, has excellent discrimination, but over-predicts mortality and has poor calibration. Although the lack of calibration may not invalidate the score, the PELOD score is a discontinuous variable and we advise careful consideration when using it as a surrogate endpoint in clinical trials.