Abstract 18941: Pediatric Electrocardiogram-Based Deep Learning to Predict Left Ventricular Dysfunction and Remodeling

• Published in: Circulation (New York, N.Y.) Vol. 148; no. Suppl_1
• Main Authors: Mayourian, Joshua, La Cava, William G, Ghelani, Sunil J, Mannix, Rebekah, Bezzerides, Vassilios J, Pu, William T, Geva, Tal, Dionne, Audrey, Alexander, Mark E, Triedman, John K
• Format: Journal Article
• Language: English
• Published: 07.11.2023
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.148.suppl_1.18941

Abstract only Background: Artificial intelligence-enhanced ECG algorithms show promise to detect a range of adult phenotypes, inclusive of ventricular dysfunction and remodeling. However, it remains unclear whether similar approaches provide meaningful and generalizable predictions in children. Methods: We trained a convolutional neural network on paired ECG-echos (≤ 2 days apart) from patients ≤ 18 years old without major congenital heart disease to detect human expert classified greater than mild left ventricular (LV) dysfunction, hypertrophy, and dilation (individually and as a composite outcome). Model performance was assessed by calculating area under the receiver operating curve (AUROC) in an internal test set and external setting (emergency department). Results: The main cohort comprised of 119,787 ECG-echo pairs (61,722 patients; median age 8.3 [IQR, 2.9-13.8] years; 55% male). Training, internal test, and external test sets comprised of 92,377, 24,343, and 3,067 ECG-echo pairs with 8.2%, 7.7%, and 10.0% composite outcome rates, respectively. During internal testing, AUROCs of 0.84 [95% CI, 0.83-0.85], 0.87 [95% CI, 0.85-0.88], 0.86 [95% CI, 0.84-0.87], and 0.84 [95% CI, 0.83-0.86] were achieved for the LV composite outcome, dysfunction, hypertrophy, and dilation, respectively (Figure). During external testing, AUROCs of 0.80 [95% CI, 0.77-0.83], 0.84 [95% CI, 0.80-0.88], 0.77 [95% CI, 0.72-0.82], and 0.86 [95% CI, 0.82-0.89] were achieved, respectively (Figure). Similar model performance was achieved when using quantitative z-score cutoffs for each outcome. Exploratory saliency mapping provides insight into notable ECG components that influence model predictions. Conclusions: Our artificial intelligence-enhanced pediatric ECG algorithm shows promise to inexpensively screen for and diagnose LV dysfunction or remodeling in children. Prospective trials to guide model implementation for supporting clinical decision making are warranted.