Abstract 214: A Counterfactual Machine Learning Model Study To Estimate The Impact Of Coronary Angiography For Out Of Hospital Cardiac Arrest

Abstract

Background: The American Heart Association guidelines for resuscitation 2020 recommends emergency coronary angiography (CAG) in cases of out-of-hospital cardiac arrest (OHCA) after return of spontaneous circulation (ROSC) with ST-segment elevation(STE) and, it is unclear how much CAG actually contributes to improved outcomes in various OHCA patient populations. The aim of this study is to estimate the impact of CAG on OHCA using a counterfactual machine learning model. Methods: We analyzed adult non-traumatic OHCA patients (43,977 cases) from June 2014 to December 2019 from the JAAM (Japan Association for Acute Medicine)-OHCA registry, which includes post-hospitalization data from 140 Japanese hospitals. We defined outcomes as favorable neurological outcomes (Cerebral Performance Categories (CPC) 1-2) and survival at 30 and 90 days. We estimated the outcome assuming that all patients had received CAG using a machine learning and counterfactual hypothetical model, and also estimated the outcome assuming that all patients had not received CAG. An average of each individual difference between the two estimations was calculated as average treatment effect (ATE) (%), which represents the effect of CAG on improving outcomes for the population. In addition, subgroup analysis was performed based on ROSC, STE and initial documented rhythm. Results: The ATE(%, 95% confidence interval) of CAG in all OHCA patients for CPC 1-2 at 30 and 90 days was 3.11% (2.92-3.30) and 1.85% (1.74-1.95), and survival at 30 and 90 days was 6.25%(5.93-6.57) and 2.20%(1.88-2.52), respectively. The subgroup analysis of ATE for CPC 1-2 at 90 days was particularly lower in the no STE group, initial pulseless electrical activity (PEA) group and initial asystole group than in the STE group and initial ventricular fibrillation(VF) group (Figure). Conclusions: In the JAAM-OHCA registry, we used a counterfactual machine learning model to estimate the effect of CAG on OHCA.