Abstract 274: Identifying Target Populations For Extracorporeal Cardiopulmonary Resuscitation Among Patients With Refractory Out-of-hospital Cardiac Arrest Using Machine Learning

• Published in: Circulation (New York, N.Y.) N.Y.), 2022-11, Vol.146 (Suppl_1), p.A274-A274
• Main Authors: Osawa, Itsuki, Hirayama, Ichiro, Hayakawa, Mineji, Shimada, Yuichi J, Doi, Kent
• Format: Article
• Language: English
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.146.suppl_1.274

IntroductionExtracorporeal cardiopulmonary resuscitation (ECPR) can be used on patients with out-of-hospital cardiac arrest (OHCA) when the cause is potentially reversible. However, little is known about the optimal target populations for ECPR among those with refractory OHCA. MethodsWe retrospectively analyzed nationwide OHCA registry data on 57,754 patients from 93 hospitals in Japan, 2014-2019. We identified 1,273 patients having a witness or initially shockable rhythms with refractory OHCA of presumed cardiac origin without return of spontaneous circulation after >3 doses of adrenaline administration and defibrillations when necessary at the emergency department (ED). We used the causal forest—a machine learning method to estimate heterogeneous treatment effects—to determine individual treatment effects of ECPR on a favorable 30-day neurologic outcome defined as the Cerebral Performance Category scale of 1 or 2, adjusting for confounders (e.g., age, sex, initial cardiac rhythms, and laboratory data). We compared patient characteristics in each quintile of individual ECPR effects to identify potential determinants of ECPR effects in patients with refractory OHCA. Finally, to determine optimal targets of ECPR, we evaluated how ECPR effects could vary depending on patient characteristics. ResultsWe found that the treatment effects of ECPR on a favorable neurologic outcome were heterogeneous (top 20% vs. others, 0.15 vs. 0.00; p