Magnetic Resonance-Based Diagnostics for Bleeding Assessment in Neonatal Cardiac Surgery

Abstract

Infants undergoing a cardiac operation are at high risk for postsurgical bleeding. To date, there are no highly predictive models for postsurgical bleeding in this population. This study's objective was to assess the predictive ability of T2 magnetic resonance (T2MR). T2MR uses magnetic resonance to detect clot formation characteristics on a small blood sample and provides hemostatic indicators that can assess bleeding risk. This prospective, single-institution study enrolled 100 patients younger than 12 months old undergoing a cardiac operation from April 27, 2015, to September 21, 2016. The primary end point was postsurgical bleeding within 24 hours after the procedure. T2MR data were modeled with a binary recursive partitioning algorithm with randomized cross-validation. The tight clot metric produced the highest univariate discrimination of bleeding (receiver operator characteristic curve, 0.64; classification accuracy, 72%), and along with the platelet function metric, demonstrated highest relative importance based on Gini index splitting (Salford Systems, San Diego, CA). Multivariate modeling with cross-validation showed mean receiver operator characteristic curve area of 0.74 and classification accuracy of 82%. T2MR tight clot and platelet function metrics were associated with bleeding events.