Abstract 18643: Deep Learning Analysis and Prognostic Significance of ECG Features in Systemic Sclerosis
Circulation · 02.11.2026 · 154 (Suppl_1)
DOI: 10.1161/circ.154.suppl_1.18643 ↗Kurz gefasst
Fachartikel in der Zeitschrift Circulation, erschienen am 02.11.2026. Verfasst von Rayyan Islam und 10 weiteren Autor:innen.
Abstract (Originalsprache)
Background
Cardiac involvement is a major cause of death in systemic sclerosis (SSc), but the ECG manifestations of SSc-related cardiac disease and their prognostic significance remain incompletely characterized. We use deep learning and survival analysis to characterize a learned SSc ECG signature and identify prognostic serial ECG abnormalities.
Methods
We analyzed 38,271 ECGs from 3,369 patients at Johns Hopkins: 878 with SSc and 2,491 age, sex, and race-matched controls without cardiac or respiratory disease. A waveform-only 1D residual CNN was trained on raw 12-lead waveforms to classify SSc from controls. We related predictions to ECG abnormalities and waveform morphology, with integrated gradients used to localize waveform attribution. In survival analysis, multivariable time-varying Cox models assessed serial ECG abnormalities against all-cause mortality.
Results
Within ±90 days of diagnosis, the CNN distinguished SSc from matched controls with AUROC 0.851 (95% CI 0.771-0.909). Performance attenuated with temporal distance but remained discriminative >2 years before and after diagnosis (AUROC 0.658 and 0.715, respectively). AUROC was higher in pulmonary hypertension, interstitial lung disease, and arrhythmia-positive strata (0.81-0.86) but remained discriminative in negative strata (0.71-0.75), suggesting a broader SSc ECG signature not explained solely by these complications. Model-predicted SSc probability was higher in feature-positive strata for right-heart and conduction abnormalities, including right ventricular hypertrophy (RVH, AUROC 0.90), complete right bundle branch block (RBBB, 0.85), left anterior fascicular block (LAFB, 0.82), and right-axis deviation (0.78). Integrated-gradient analysis localized model attribution to QRS morphology, with greatest saliency in V1-V2 and aVR. In survival analysis, 247 (28%) of the SSc patients died; RV hypertrophy (HR 2.60), pulmonary hypertension (2.55), atrial fibrillation (2.45), complete RBBB (2.13), low QRS voltage (1.74), right-axis deviation (1.50), and ventricular rate (1.25) were independently associated with mortality; all p
Zitieren
APA
Islam, R., Zaeske, A., Woods, A., Hsu, S., Lammi, M., Hummers, L., Mathai, S., Hassoun, P., Wigley, F., Shah, A., & Mukherjee, M. (2026). Abstract 18643: Deep Learning Analysis and Prognostic Significance of ECG Features in Systemic Sclerosis. Circulation, 154(Suppl_1). https://doi.org/10.1161/circ.154.suppl_1.18643
Vancouver
Islam R, Zaeske A, Woods A, Hsu S, Lammi M, Hummers L, et al. Abstract 18643: Deep Learning Analysis and Prognostic Significance of ECG Features in Systemic Sclerosis. Circulation. 2026;154(Suppl_1). doi:10.1161/circ.154.suppl_1.18643
BibTeX
@article{Islam2026,
author = {Islam, Rayyan and Zaeske, Anna and Woods, Adrianne and Hsu, Steven and Lammi, Matthew and Hummers, Laura and Mathai, Stephen and Hassoun, Paul and Wigley, Fredrick and Shah, Ami and Mukherjee, Monica},
title = {Abstract 18643: Deep Learning Analysis and Prognostic Significance of ECG Features in Systemic Sclerosis},
journal = {Circulation},
year = {2026},
volume = {154},
number = {Suppl_1},
doi = {10.1161/circ.154.suppl_1.18643}
}
EUVIMED