AI misinterpretation occurred in 2.0% of 4969 ECGs, most commonly as atrial fibrillation (58.4%), leading to inappropriate medical therapies and increased resource utilization.
Observational (n=4,969)
Does AI misinterpretation of ECGs lead to inappropriate clinical management and resource utilization?
AI misinterpretation of ECGs, particularly false-positive atrial fibrillation, occurs in 2.0% of cases and can lead to inappropriate medical therapies and unnecessary cardiology consultations, highlighting the need for careful physician over-read.
INTRODUCTION: The advancement of artificial intelligence (AI) has aided clinicians in the interpretation of electrocardiograms (ECGs) serving as an essential tool to provide rapid triage and care. However, in some cases, AI can misinterpret an ECG and may mislead the interpreting physician. Therefore, we aimed to describe the rate of ECG misinterpretation and its potential clinical impact in patient's management. METHODS: We performed a retrospective descriptive analysis of misinterpreted ECGs and its clinical impact from May 28, 2020 to May 9, 2021. An electrophysiologist screened ECGs with confirmed diagnosis of atrial fibrillation (AF), sinus tachycardia (ST), sinus bradycardia (SB), intraventricular conduction delay (IVCD), and premature atrial contraction (PAC) that were performed in the emergency department. We then classified the misinterpreted ECGs as wrongly diagnosed AF, ST, SB, IVCD, or PAC into the correct diagnosis and reviewed the misinterpreted ECGs and medical records to evaluate inappropriate use of antiarrhythmic drugs (AAD), beta-blockers (BB), calcium channel blockers (CCB), anticoagulation, or resource utilization of cardiology and/or electrophysiology (EP) consultation. RESULTS: A total of 4969 ECGs were screened with diagnoses of AF (2282), IVCD (296), PAC (972), SB (895), and ST (638). Among these, 101 ECGs (2.0%) were misinterpreted. Wrongly diagnosed AF (58.4%) was the most common followed by wrongly diagnosed PAC (14.9%), wrongly diagnosed ST (12.9%), wrongly diagnosed IVCD (7.9%), and wrongly diagnosed SB (6.0%). Patients with misinterpreted ECGs were aged 76.6 ± 11.6 years with male (52.5%) predominance and hypertension being the most prevalent (83.2%) comorbid condition. The misinterpretation of ECGs led to the inappropriate use of BB (19.8%), CCB (5.0%), AAD therapy (7.9%), anticoagulation (6.9%) in patients with wrongly diagnosed AF, as well as inappropriate resource utilization including cardiology (41.6%) and EP (8.9%) consultations. CONCLUSIONS: Misinterpretation of ECGs may lead to inappropriate medical therapies and increased resource utilization. Therefore, it is essential to encourage physicians to carefully examine AI interpreted ECG's, especially those interpreted as having AF.
Shaik et al. (Mon,) conducted a observational in ECG interpretation (n=4,969). AI ECG interpretation was evaluated on Rate of ECG misinterpretation. AI misinterpretation occurred in 2.0% of 4969 ECGs, most commonly as atrial fibrillation (58.4%), leading to inappropriate medical therapies and increased resource utilization.