Identifying Subgroups with Differential Responses to Amiodarone among Cardiac Arrest Patients with a Shockable Rhythm at Hospital Arrival using the Machine Learning Approach

Abstract

Background: There are few reports of studies on the differential effects of amiodarone among out-of-hospital cardiac arrest (OHCA) patients with a shockable rhythm at hospital arrival. The present study aimed to investigate the clinical heterogeneity of OHCA patients with a shockable rhythm upon hospital arrival and to identify subgroups with differential responses to amiodarone, using a machine learning approach. Methods: We used the Japanese nationwide OHCA registry of the Japanese Association for Acute Medicine for this study; data from OHCA patients with a shockable rhythm at hospital arrival were included in the analyses. The primary outcome was a favorable neurological outcome at 30 days. We developed a scoring system by the weighting method with logistic likelihood loss to identify patient subgroups showing differential effects of amiodarone from the point of view of the neurological outcome and survival at 30 days. Results: Among the 68,111 cases of OHCA in the registry, the data of 2333 OHCA patients with an initial shockable rhythm at hospital arrival were analyzed. The developed score identified higher age, longer interval between the call to the emergency medical service and hospital arrival, absence of a “witness”, no defibrillation prior to hospital arrival, hypothermia at hospital arrival, and pre-hospital epinephrine administration as variables that were significantly associated with a beneficial effect of amiodarone. Based on the results of the developed scoring system, 47% (1107/2333) of the patients were considered to greatly benefit from amiodarone administration, whereas 53% (1226/2333) of patients were considered to not benefit from amiodarone administration. The effect of amiodarone on the neurological outcome at 30 days varied significantly among the subgroups identified by the developed score (ORinteraction: 1.07 [95% confidence interval (CI): 0.99–1.13], p = 0.005). Conclusions: We successfully developed a model that could discriminate between OHCA patients with an initial shockable rhythm at hospital arrival who would benefit or not benefit from the administration of amiodarone in terms of the neurological outcome at 30 days. There was clinical heterogeneity among OHCA patients with a shockable rhythm in terms of their response to amiodarone.

Keywords

• out-of-hospital cardiac arrest
• amiodarone
• shockable rhythm
• machine learning
• differential responsiveness