Reversal of Heart Age in Adults with Obesity After Initiation of Semaglutide Detected by Serial Electrocardiography

Semaglutide lowers cardiovascular events in adults with obesity, but whether it slows biological aging of the heart, and whether any such effect is explained by weight loss, is unknown. We developed and validated a hybrid convolutional and transformer deep learning neural network model that estimates age from a routine 12-lead electrocardiogram (ECG), using a training set of 1,019,079 adult ECGs and a held-out test set of 99,994 adult ECGs. The model estimated age with a mean absolute error (MAE) of 5.80 years and explained 82% of the variation in age (R² 0.817), the best reported performance for any ECG age model to date. On an independent validation population of 802,331 patients, the model achieved an MAE of 6.92 years and R² of 0.751. We defined relative ECG age as ECG-predicted age minus chronologic age. We then compared its change from the last ECG before treatment between new users of semaglutide and patients undergoing bariatric surgery, and between patients who initiated semaglutide versus tirzepatide with at least 6 months on treatment. Patients were matched 1:1 on sex, age and year of treatment start, and on a propensity score built from baseline BMI, weight, ECG age, number of prior ECGs, eight comorbidities and race (2,549 semaglutide-bariatric surgery pairs; 950 semaglutide-tirzepatide pairs). At 12 months, bariatric surgery patients had lost an average of 25.6% of their body weight whereas semaglutide patients lost only 4.8% of body weight. However, relative ECG age on semaglutide decreased significantly below baseline at every window from 6 months onward, by −0.73 years (95% confidence interval [CI] −1.29 to −0.17) at 12 months and −1.23 years (95% CI −2.04 to −0.42) at 15 months, whereas relative ECG age did not significantly change at any evaluated time point after bariatric surgery. The difference in relative ECG age change (semaglutide minus bariatric surgery) was a statistically significant −1.28 years (95% CI −2.48 to −0.09; P=0.035; 304 vs 148 patients) at 12 months and −1.86 years (95% CI −3.19 to −0.53; P=0.006; 179 vs 95) at 21 months. Compared to tirzepatide, which was also associated with more average weight loss than semaglutide (−11.2% versus −7.1% at 12 months), the largest difference in relative ECG age (semaglutide minus tirzepatide) was −1.19 years at 15 months and was not significant (95% CI −3.27 to +0.88, P=0.26). Over 5 years, mortality was lower in semaglutide patients than in their matched bariatric surgery patients (hazard ratio [HR] 0.44, 95% CI 0.34 to 0.59), and this difference in mortality accounted for the lower rate of major adverse cardiovascular events (MACE; death, heart attack or stroke) on semaglutide (HR 0.73, 95% CI 0.58 to 0.91). Taken together, relative ECG age fell after initiation of semaglutide or tirzepatide but not meaningfully after bariatric surgery, despite the exact opposite magnitude of weight loss achieved, suggesting electrocardiographic heart age may reflect a hitherto unappreciated, weight-loss-independent dimension of GLP-1 pharmacology not associated with currently monitored standard clinical cardiovascular outcomes.