A novel multimodal computational system using near-infrared spectroscopy predicts the need for ECMO initiation in neonates with congenital diaphragmatic hernia

Background/purpose: The purpose of this study was to develop a computational algorithm that would predict the need for ECMO in neonates with congenital diaphragmatic hernia (CDH). Methods: CDH patients from August 2010 to 2016 were enrolled in a study to continuously measure cerebral tissue oxygen saturation (cStO2) of left and right cerebral hemispheres. NIRS devices utilized were FORE-SIGHT, CASMED and INVOS 5100, Somanetics. Using MATLAB (c), a data randomization function was used to deidentify and blindly group patient's data files as follows: 12 for the computational model development phase (6 ECMO and 6 non-ECMO) and the remaining patients for the validation phase. Results: Of the 56 CDH patients enrolled, 22 (39%) required ECMO. During development of the algorithm, a difference between right and left hemispheric cerebral oxygenation via NIRS (Delta HCO) was noted in CDH patients that required ECMO. Using ROC analysis, a Delta HCO cutoff > 10% was predictive of needing ECMO (AUC: 0.92; sensitivity: 85%; and specificity: 100%). The algorithm predicted need for ECMO within the first 12 h of life and at least 6 h prior to the clinical decision for ECMO with 88% sensitivity and 100% specificity. Conclusion: This computational algorithm of cerebral NIRS predicts the need for ECMO in neonates with CDH. (c) 2017 Elsevier Inc. All rights reserved.