Abstract 17934: Performance of an Electrocardiographic Deep Learning Model for Detecting Wall Motion Abnormalities: An Analysis of Impact From Cardiovascular Comorbidities and Race in an External Validation Study
Rayan A Ansari, neal bhatia, James Tooley, Vyom Thakkar, JUSTIN XU, Jessica Torres, Jagteshwar Tung, Mahmood Alhusseini, Paul L Clopton, Reza Sameni, Gari Clifford, Euan A Ashley, Marco V Perez, Matei Zaharia, Sanjiv M Narayan, Albert J Rogers- Physiology (medical)
- Cardiology and Cardiovascular Medicine
Introduction: Cardiac wall motion abnormalities (WMA) are key prognostic indicators for major cardiovascular events and mortality (MACE), yet poorly detected by current ECG screening such as for Q waves. Given that race and gender impact the ECG, it is also unknown if ECG screening for WMA could be applied broadly. We hypothesized that deep learning of the ECG could accurately identify WMA in diverse patients with varying comorbidities at 2 Institutions.
Methods: We developed a DL Convolutional Neural Network model on N=35,210 ECG records from Stanford University (iECG, Philips), labeled by natural language processing of echocardiography reports within 60 days ( Panel A ). The tested model was then externally validated on an independent set of N=2338 ECGs from Emory University (Muse, GE). Patients with ventricular pacing were excluded. The patient cohort was segmented by cardiovascular comorbidity (<3, >=3) and white vs. non-white race to compare model performance.
Results: The Stanford cohort (62.9±16.9 years, 47% female, 58% white, 6% black, 16% Asian, 23% >=3 comorbidities) differed from the Emory cohort (61.9±16.0 years, 48% female, 48% white, 41% black, 6% Asian, 52% >=3 comorbidities). Nevertheless, the Stanford model detected WMA in the Emory cohort with AUC 0.723. For patients with low and high comorbidity burdens, the AUC was 0.71 and 0.69, respectively. For White and Non-White patients, the AUC was 0.74/0.70 and 0.65/0.66, respectively ( Panel B ).
Conclusions: Across racial groups and patients with variable burden of comorbidities, a DL model of electrocardiograms robustly identifies WMAs in an external validation cohort despite differences in the populations. The model’s performance across different comorbidity and racial demographic groups underscores its potential for general clinical application, offering an effective tool for clinical screening and enhanced risk stratification.