Abstract 396: A Novel Machine Learning Approach for Identifying Subgroups With Differential Responses to Amiodarone in Patients With Cardiac Arrest With Shockable Rhythm

Abstract

Introduction: Although there is a possibility that the effectiveness of amiodarone may differ among a particular subgroup, there are few studies on the differential effects of amiodarone for the out-of-hospital cardiac arrest (OHCA) patients with a shockable rhythm. Objective: To identify subgroups with differential responses to amiodarone in OHCA with shockable rhythm by a novel machine learning approach. Methods: We used data from the Japanese nationwide OHCA registry of the Japanese Association for Acute Medicine. The OHCA patients who had initial shockable rhythm upon hospital arrival were included. The primary outcome was a good neurological outcome at 30 days, defined as a Cerebral Performance Categories score ≤2. We developed a linear score by outcome weighting learning (OWL) method with logistic loss to identify subgroups with differential effects of amiodarone. The difference in the effect of amiodarone between subgroups based on the developed score was evaluated by repeated cross-validation (RCV), and hypothesis testing was performed by the permutation method. Results: Among 68,111 OHCA patients in the registry, data of 2333 patients with initial shockable rhythm upon hospital arrival were analyzed. Our developed score showed that the longer duration from call to EMS to hospital arrival, hypothermia at hospital arrival, and no defibrillation until hospital arrival were important variables that had a trend of increasing the beneficial effect of amiodarone (Table). The effect of amiodarone on neurological outcomes varied significantly among the subgroups identified by the developed scores (OR: 1.05 (95%CI: 0.99-1.13), P [RCV] = 0.038). Conclusions: We successfully developed a model that discriminates between the patients for whom amiodarone had a harmful effect on their neurological outcome from those for whom amiodarone had a beneficial effect. In the patients for whom amiodarone had a harmful effect, the other treatment approaches may be needed.