Electrocardiogram-based mortality prediction in patients with COVID-19 using machine learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Infectious Disease Epidemiology

Background:

• Electrocardiograms (ECGs) are routinely used in COVID-19 patient care.
• No prior studies have assessed the value of ECG-based machine learning for predicting COVID-19 mortality.

Purpose of the Study:

• To evaluate the added value of ECG-based machine learning models in predicting in-hospital mortality among COVID-19 patients.
• To compare the performance of logistic regression, LASSO, and deep neural network (DNN) models using ECG data.

Main Methods:

• Utilized data from the CAPACITY-COVID registry, including 882 COVID-19 patients from seven Dutch hospitals.
• Developed three predictive models: logistic regression (baseline), LASSO (with ECG features), and DNN (with raw ECG waveforms).
• Externally validated models using data from two additional hospitals.

Main Results:

• All three models demonstrated comparable performance in predicting mortality (AUCs ranging from 0.73 to 0.77).
• Key predictors identified by the LASSO model included age, low QRS voltage, ST depression, premature atrial complexes, sex, increased ventricular rate, and right bundle branch block.
• Deep neural networks (DNNs) using raw ECG data performed comparably to models relying on manual feature annotation.

Conclusions:

• ECG-based prediction models can aid in the initial risk stratification of COVID-19 patients.
• Several ECG abnormalities are significantly associated with in-hospital all-cause mortality in COVID-19.
• Pre-trained DNNs offer a viable alternative to manual ECG annotation for predictive modeling.