A deep learning model using resting 12-lead ECGs predicted whether patients would achieve their age-predicted peak heart rate during stress testing with an AUROC of 0.83 (95% CI 0.81-0.85).
Observational (n=6,986)
No
Does a deep learning model using resting 12-lead ECGs predict peak heart rate and appropriate stress response in patients undergoing stress testing?
A deep learning model can predict peak heart rate and chronotropic incompetence from a resting 12-lead ECG, potentially streamlining stress testing and patient management.
Effect estimate: AUROC 0.83 (95% CI 0.81-0.85)
Introduction: Cardiovascular stress testing is crucial for the evaluation of ischemic cardiomyopathy and inducible arrhythmias. Inappropriate heart rate (HR) response during stress, or chronotropic incompetence, is associated with sinus node disease, conduction system abnormalities, and decreased functional capacity. However, inappropriate exercise tolerance often presents during stress testing, necessitating early termination. Early identification of those who are unable to complete testing due to exercise intolerance or chronotropic incompetence could streamline subsequent testing and management. This study investigates the feasibility of using deep learning models to predict peak HR using resting 12-lead electrocardiogram (ECG). Research Questions: Can a deep learning model effectively learn from resting 12-lead ECG waveforms to: (1) classify whether a patient's peak heart rate will exceed predicted peak HR defined as (230-age) × 0.8, and (2) predict the peak heart rate achieved during a stress test? Methods: A total of 7,625 stress test records were obtained from a single institution, from which 6986 samples (4893 training/validation, 2093 test) were included. Preprocessing involved extracting 12-lead ECG signals, identifying the resting waveform and the peak HR during stress. All 12 standard leads were required for inclusion, and ECG waveforms were padded to a uniform sequence length. The processed data was used to train two separate convolutional neural networks for predicting appropriate stress response and the peak HR. Results: The training/validation set had a mean peak HR of 137.5 beats per minute (SD = 34.9) and the testing set had a mean peak HR of 137.0 (SD = 35.3). The proportion of samples that met age-based peak HR threshold in the training/validation and testing set is 64.7% and 65.9% respectively. The model achieved an AUROC of 0.83 (95% CI 0.81-0.85) and an F1-score of 0.85 for the classification task. A separate model with similar architecture predicted peak HR with an R-value of 0.69 (95% CI 0.67-0.72) and root mean square error of 26.2 beats per minute (95% CI 25.3-27.2). Conclusion: Our findings show that the resting ECG can be leveraged for the prediction of HR response prior to stress testing. Successful models could offer clinicians a valuable non-invasive tool for early risk stratification, guiding patient management in the evaluation of ischemic heart disease, and identifying those at risk for chronotropic incompetence.
Liu et al. (Mon,) conducted a observational in Patients undergoing stress testing (n=6,986). Deep learning model using resting 12-lead ECG was evaluated on Classification of whether peak heart rate exceeds predicted peak HR and prediction of peak HR (AUROC 0.83, 95% CI 0.81-0.85). A deep learning model using resting 12-lead ECGs predicted whether patients would achieve their age-predicted peak heart rate during stress testing with an AUROC of 0.83 (95% CI 0.81-0.85).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: