ESC Congress 2026 Abstract · Oct 26, 2026
Comparison of AI-assisted ECG and clinical risk scores in selecting individuals with type 2 diabetes for screening for asymptomatic heart failure
By Wegrzyn-Witek A, Soh C, Kosmala W et al.
Background: Early identification of stage B heart failure (SBHF) in individuals with type 2 diabetes mellitus (T2DM) may allow timely initiation of cardioprotective therapies and help prevent progression to symptomatic heart failure. However, cardiovascular risk in T2DM is heterogeneous, and a selection process for echocardiography would be desirable. Purpose: To evaluate whether conventional clinical risk scores (ARIC-HF for the general population and WATCH-DM for individuals with diabetes) or an artificial intelligence–enhanced electrocardiogram (AI-ECG) could improve selection of patients for echocardiographic screening. Methods: Clinical data, 12-lead AI-ECG and 2D echocardiography (2DE) were obtained prospectively in patients with T2DM, age >65 years enrolled in the CD-AIM trial. Risk of HF was quantified using the ARIC-HF and WATCH-DM risk scores. The AI-ECG score was derived from a continuous wavelet transform signal processing to provide new frequency and energy information to detect cardiac relaxation abnormalities. SBHF was defined as impaired global longitudinal strain (GLS<16%) or diastolic dysfunction (DD), defined as E/e’≥14 and LAVi >34 ml/m². Results: Among 176 participants, ARIC-HF ranged from 0.03% to 25.52%, and WATCH-DM from 1 to 20. SBHF was identified in 46 (26.1%) patients. Impaired GLS was present in 38 (21.6%) participants, and DD in 46 (26.1%). Using optimised thresholds (ARIC-HF >2.2%, WATCH-DM ≥7, AI-ECG ≥0.61), all three tools showed high sensitivity but limited specificity (figure). ARIC-HF demonstrated the strongest diagnostic performance, with AUC values of 0.61–0.67 across endpoints, outperforming both WATCH-DM and AI-ECG (AUC 0.54–0.57). Conclusion: ARIC-HF provided the strongest risk stratification among the three approaches, although its discrimination remained moderate. None of the investigated tools demonstrated sufficient accuracy for reliable SBHF detection alone. The limited diagnostic performance of AI-ECG indicates that further refinement is required before it can be reliably implemented as an independent screening tool.ARIC, WATCH and ECG