Safety and accuracy of the computer interpretation of normal ECGs at triage.

Abstract

Background: ECGs performed at ED triage are mandatorily assessed by an emergency physician contributing to task interruptions, decreased quality of care and increased error risk. Recent literature suggests that a triage ECG interpreted as normal by the ECG machine software correlates with benign interpretation from attending cardiologists. Ambiguity persists regarding the safety of the normal computerized ECG interpretation and whether real-time physician review is needed.

Methods: In this prospective cohort study, we evaluated the GE Marquette 12SL ECG software's clinical accuracy for normal triage ECGs. We compared machine interpretation, ED physicians' real-time interpretation and management as well as cardiologists systematic but delayed review of the ECGs to establish diagnostic accuracy and safety of a normal software interpretation. Additional data were collected from the emergency department patient tracking system and charts to determine patient outcomes when machine normal interpretation differed from ED physicians and cardiologists' analysis.

Results: Between November 16th 2020 and November 19th 2021, we collected 1220 machine-normal ECGs performed at Charles-Le Moyne Hospital ED triage. 1051 ECGs were eligible for inclusion. ED physicians agreed with the machine for 1028 normal diagnoses (97.8% IC (0.95): 96.7-98.5) and cardiologists agreed with the machine for 1037 ECGs (98.7% IC (0.95) 97.8-99.2). Relevant changes in ED management were recorded in three patients (0.29% IC (0.95): 0.1-0.8) whose ECGs were read as normal by the software: two emergent referrals to the catheterization lab and one emergent transfer to the resuscitation bay with next day coronarography.

Conclusion: A normal ECG interpretation from the GE Marquette 12SL ECG software at ED triage has a very high accuracy and a very low probability of clinically relevant change in patient outcome and ED trajectory.