Artificial Intelligence (AI) has shown promise in augmenting ECG analysis. We previously identified QRS amplitude diminution as a predictor of mortality in COVID-19 on follow-up ECG; but ECG data would be most useful clinically if predictive upon admission. To assess whether QRS amplitude on the admission ECG predicts mortality in COVID-19 utilizing AI processing techniques. We performed a retrospective analysis of patients admitted with laboratory confirmed SARS-CoV-2 between March 5 and July 7, 2020 (n=4,709). Patients were excluded if the ECG was not acquired within 72 hrs of admission (n=1,692). Low QRS Amplitude (LoQRS) was defined by a composite of QRS amplitude <5mm in the limb leads AND/OR <10mm in the precordial leads (a composite of V1-V3 and V4-V6). Among 3,012 patients, 373 (14.1%) met criteria for LoQRS (Figure). Compared to patients without LoQRS, these patients had a higher risk of in-hospital mortality (17.4% vs. 10.6%, p<0.001), ICU admission (31.9% vs 22.2%, p<0.001), and mechanical ventilation (26.3% vs 17.5%). Low QRS amplitudes were noted in both limb and precordial leads. In multivariable models, LoQRS was independently associated with mortality (OR 1.55, 95% CI 1.2-2.00, p<0.001) as were age (OR 1.05, 95% CI 1.04-1.06), BMI (OR 1.02, 95% CI 1.01-1.03), and CKD (OR 1.49, 95% CI 1.06-2.12). LoQRS also independently predicted ICU admission (OR 1.7, 95% CI 1.32-2.18) and mechanical ventilation (OR 1.6, 95% CI 1.23-2.00). QRS amplitude on the presenting ECG independently predicts mortality, ICU admission, and mechanical ventilation in hospitalized patients with COVID-19, and may allow initial risk stratification.
Read full abstract