Artificial Intelligence-Enabled ECG for Diastolic Dysfunction in Congenital Heart Disease: A Novel Risk Stratification Tool

Comprehensive Summary

O’Sullivan et al. evaluated an AI-enabled ECG model for diastolic dysfunction in 6,741 adults with congenital heart disease (median age: 37 years; 49% female) followed for a median of 10 years. The model classified 65.8% of patients as grade 0, 4.0% as grade 1, 19.7% as grade 2, and 10.5% as grade 3. Higher grades were associated with increasing disease complexity, higher heart failure prevalence (6.7% in grade 0 vs. 24.8% in grade 3), as well as rising NT-proBNP (129 to 763 pg/mL). AI-ECG grades correlated well with echocardiographic strain and invasive filling pressures, with pulmonary artery wedge pressure rising from 11 to 16 mmHg, and grades 2 and 3 independently predicted mortality (HR: 1.38 and 1.63 respectively). Overall, the findings support AI-ECG as a scalable adjunct for diastolic assessment and risk stratification in adult congenital heart disease in circumstances conventional methods are frequently limited.

Outcomes and Implications

This study demonstrates that meaningful hemodynamic and prognostic information can be extracted from routine ECGs in a population with highly heterogeneous anatomy and physiology. The alignment of AI-ECG grades with invasive pressures and long-term outcomes suggests potential value for identifying patients at higher risk who may benefit from closer surveillance, earlier referral, or invasive evaluation. At the same time, the observed variability in performance across congenital subtypes emphasizes that ECG-based AI should complement, rather than replace, anatomy-specific imaging and clinical judgment, and highlights the need for prospective and lesion-specific validation before broad clinical adoption.