The use of mortality prediction scores in clinical trials in the PICU is essential for comparing patient groups. Because of the decline in PICU mortality over the last decades, leading to a shift toward later deaths, recent trials use 90-day mortality as primary outcome for estimating mortality and survival more accurately. This study assessed and compared the performance of two frequently used PICU mortality prediction scores for prediction of PICU and 90-day mortality.
This secondary analysis of the randomized controlled Early versus Late Parenteral Nutrition in the Pediatric Intensive Care Unit trial compared the discrimination (area under the receiver operating characteristic curve) and calibration of the Pediatric Index of Mortality 3 and the Pediatric Risk of Mortality III scores for prediction of PICU and 90-day mortality.
Three participating PICUs within academic hospitals in Belgium, the Netherlands, and Canada.
One-thousand four-hundred twenty-eight critically ill patients 0–17 years old.
Although Pediatric Index of Mortality 3 only includes information available at the time of PICU admission, thus before any intervention in the PICU, it showed good discrimination (area under the receiver operating characteristic curve, 0.894; 95% CI, 0.892–0.896) and good calibration (no deviation from the diagonal, p = 0.58) for PICU mortality. Pediatric Risk of Mortality III, which involves the worst values for the evaluated variables during the first 24 hours of PICU stay, was statistically more discriminant (area under the receiver operating characteristic curve, 0.920; 95% CI, 0.918–0.921; p = 0.04) but poor in calibration (significant deviation from the diagonal; p = 0.04). Pediatric Index of Mortality 3 and Pediatric Risk of Mortality III discriminated equally well between 90-day mortality and survival (area under the receiver operating characteristic curve, 0.867; 95% CI, 0.866–0.869 and area under the receiver operating characteristic curve, 0.882; 95% CI, 0.880–0.884, respectively, p = 0.77), but Pediatric Risk of Mortality III was not well calibrated (p = 0.04), unlike Pediatric Index of Mortality 3 (p = 0.34).
Pediatric Index of Mortality 3 performed better in calibration for predicting PICU and 90-day mortality than Pediatric Risk of Mortality III and is not influenced by intervention or PICU quality of care. Therefore, Pediatric Index of Mortality 3 seems a better choice for use in clinical trials with 90-day mortality as primary outcome.
1Clinical Division and Laboratory of Intensive Care Medicine, Department of Cellular and Molecular Medicine, KU Leuven University Hospital, Leuven, Belgium.
2Virtual Pediatric Systems, LCC, Los Angeles, CA.
3Department of Pediatrics and Pediatric Surgery, Erasmus Medical Centre, Sophia Children’s Hospital, Rotterdam, The Netherlands.
*See also p. 189.
Dr. Jacobs and Ms. Flechet contributed equally.
Ms. Flechet and Dr. Verstraete received funding from the Research Foundation, Flanders (FWO) as a PhD fellow. Dr. Ingels’ institution received funding from a doctoral fellowship of FWO Belgium, and she received funding from the Clinical Research Foundation from the University Hospitals Leuven, Belgium. Dr. Casaer received support from a postdoctoral research grant by FWO. Dr. Verbruggen received funding from Fonds NutsOhra, Sophia Research Foundation, Stichting Agis Zorginnovatie and the Erasmus Trustfonds. Dr. Vanhorebeek received funding from the Methusalem program of the Flemish Government (Belgium, METH14/06). Dr. Van den Berghe received funding from the Methusalem program of the Flemish Government (Belgium, METH14/06), Flemish Agency for Innovation through Science and Technology (IWT-TBM110685), and the European Research Council (Advanced Grant AdvG-2012–321670). The remaining authors have disclosed that they do not have any potential conflicts of interest
Virtual Pediatric Systems has calculated Pediatric Risk of Mortality III probabilities of death, but has not been involved in data analysis, nor article preparation.
For information regarding this article, E-mail: firstname.lastname@example.org