David K Bailly1, Jamie M Furlong-Dillard2, Melissa Winder3, Mark Lavering4, Ryan P Barbaro5, Kathleen L Meert6, Susan L Bratton1, Heidi Dalton7, Ron W Reeder1. 1. Division of Pediatric Critical Care, Department of Pediatrics, University of Utah, Salt Lake City, UT, USA. 2. Department of Pediatric Critical Care, Norton Children's Hospital/University of Louisville, Louisville, KY, USA. 3. Department of Pediatric Critical Care, Primary Children's Hospital, Salt Lake City, UT, USA. 4. University of Utah, Salt Lake City, UT, USA. 5. Department of Pediatrics and Communicable Diseases, University of Michigan, Ann Arbor, MI, USA. 6. Department of Pediatrics, Children's Hospital of Michigan, Detroit, MI, USA. 7. Department of Pediatrics, Inova Fairfax Hospital, Fall Church, VA, USA.
Abstract
INTRODUCTION: The Pediatric Extracorporeal Membrane Oxygenation Prediction (PEP) model was created to provide risk stratification for all pediatric patients requiring extracorporeal life support (ECLS). Our purpose was to externally validate the model using contemporaneous cases submitted to the Extracorporeal Life Support Organization (ELSO) registry. METHODS: This multicenter, retrospective analysis included pediatric patients (<19 years) during their initial ECLS run for all indications between January 2012 and September 2014. Median values from the BATE dataset for activated partial thromboplastin time and internationalized normalized ratio were used as surrogates as these were missing in the ELSO group. Model discrimination was evaluated using area under the receiver operating characteristic curve (AUC), and goodness-of-fit was evaluated using the Hosmer-Lemeshow test. RESULTS: A total of 4,342 patients in the ELSO registry were compared to 514 subjects from the bleeding and thrombosis on extracorporeal membrane oxygenation (BATE) dataset used to develop the PEP model. Overall mortality was similar (42% ELSO vs. 45% BATE). The c-statistic after external validation decreased from 0.75 to 0.64 and model calibration decreases most in the highest risk deciles. CONCLUSION: Discrimination of the PEP model remains modest after external validation using the largest pediatric ECLS cohort. While the model overestimates mortality for the highest risk patients, it remains the only prediction model applicable to both neonates and pediatric patients who require ECLS for any indication and thus maintains potential for application in research and quality benchmarking.
INTRODUCTION: The Pediatric Extracorporeal Membrane Oxygenation Prediction (PEP) model was created to provide risk stratification for all pediatric patients requiring extracorporeal life support (ECLS). Our purpose was to externally validate the model using contemporaneous cases submitted to the Extracorporeal Life Support Organization (ELSO) registry. METHODS: This multicenter, retrospective analysis included pediatric patients (<19 years) during their initial ECLS run for all indications between January 2012 and September 2014. Median values from the BATE dataset for activated partial thromboplastin time and internationalized normalized ratio were used as surrogates as these were missing in the ELSO group. Model discrimination was evaluated using area under the receiver operating characteristic curve (AUC), and goodness-of-fit was evaluated using the Hosmer-Lemeshow test. RESULTS: A total of 4,342 patients in the ELSO registry were compared to 514 subjects from the bleeding and thrombosis on extracorporeal membrane oxygenation (BATE) dataset used to develop the PEP model. Overall mortality was similar (42% ELSO vs. 45% BATE). The c-statistic after external validation decreased from 0.75 to 0.64 and model calibration decreases most in the highest risk deciles. CONCLUSION: Discrimination of the PEP model remains modest after external validation using the largest pediatric ECLS cohort. While the model overestimates mortality for the highest risk patients, it remains the only prediction model applicable to both neonates and pediatric patients who require ECLS for any indication and thus maintains potential for application in research and quality benchmarking.
Authors: Matthieu Schmidt; Aidan Burrell; Lloyd Roberts; Michael Bailey; Jayne Sheldrake; Peter T Rycus; Carol Hodgson; Carlos Scheinkestel; D Jamie Cooper; Ravi R Thiagarajan; Daniel Brodie; Vincent Pellegrino; David Pilcher Journal: Eur Heart J Date: 2015-06-01 Impact factor: 29.983
Authors: Nazima Pathan; Deborah A Ridout; Elizabeth Smith; Allan P Goldman; Katherine L Brown Journal: Intensive Care Med Date: 2008-08-01 Impact factor: 17.440
Authors: Carrie L Freeman; Tellen D Bennett; T Charles Casper; Gitte Y Larsen; Ania Hubbard; Jacob Wilkes; Susan L Bratton Journal: Crit Care Med Date: 2014-03 Impact factor: 7.598