AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
Association Between Electrocardiogram-Based Artificial Intelligence Sex Discordance and Frailty
By Kim JY, Akbilgic O, Karabayir I
Background: Frailty is a multidimensional syndrome characterized by reduced physiological reserve and increased vulnerability to adverse outcomes. Frailty indices such as the electronic frailty index (eFI) enable scalable assessment using electronic health records. However, they depend on accurate, complete, and structured data, limiting their ability to capture changes in frailty over time. Electrocardiography (ECG) offers a complementary, objective, low-cost, and widely available physiologic signal. When combined with artificial intelligence (AI), ECGs can reveal latent biological information beyond conventional interpretation, including age and sex, both of which are strongly associated with frailty. Yet, whether discordance between AI-predicted and recorded sex reflects physiological alterations associated with frailty remains unknown. Hypothesis: We hypothesized that the discrepancy between ECG-AI predicted and recorded sex reflects underlying frailty and provides information beyond standard demographics. Methods: We previously developed and validated a 1D convolutional neural network (1D-CNN) for ECG-based sex classification using ~3.7 million 12-lead ECGs from Atrium Health Wake Forest Baptist. The model achieved an AUC of 0.95 in an holdout cohort. For each ECG, it generated a probability of female sex, and sex discordance was defined as the absolute difference between the probability and recorded sex. In this study, we analyzed a subset of the holdout cohort with both a digital 12-lead ECG and an eFI assessment performed within 15 days, averaging multiple eFI values per ECG when available. Clinical frailty was defined as eFI>0.21. Associations between sex discordance and frailty were assessed using unadjusted and multivariable logistic regression adjusted for age and recorded sex, with subgroup analyses by age and sex and a sensitivity analysis restricting the ECG-eFI interval to ±7 days. Results: Among 9,275 ECG-eFI pairs, frailty prevalence was 29.7%. Mean (SD) age was 71.5 (9.5) years, 49.1% were male, and 15.4% were Black. In unadjusted analyses, sex discordance was associated with increased odds of frailty (OR 2.03, 95% CI 1.70-2.43; p<0.001). This association remained significant after multivariable adjustment (adjusted OR 2.03, 95% CI 1.69-2.45; p<0.001). The adjusted model improved fit over a demographics-only model (pseudo-R ² 0.034 vs 0.029). The association was consistent across subgroups by age (<65: OR 3.33; ≥65: OR 1.91) and sex (male: OR 1.70; female: OR 2.45). In the ±7-day sensitivity analysis, results were consistent (adjusted OR 2.05) (Table 1). Conclusion: ECG-AI sex discordance was independently associated with frailty and may serve as a novel, complementary, and objective marker of biological vulnerability. Further work is needed to define its prognostic and clinical utility for frailty assessment.