Harnessing Artificial Intelligence for Enhanced Prediction of Complete Heart Block Risk via Electrocardiography

Introduction

Complete heart block (CHB), or third-degree atrioventricular block, is a critical cardiac conduction disorder characterized by the complete dissociation of atrial and ventricular activity. CHB can precipitate ventricular standstill, recurrent syncope, and sudden cardiac death, posing substantial morbidity and mortality risks if undetected or untreated. Electrocardiography (ECG) remains the cornerstone diagnostic modality for conduction system disease, yet current ECG-based risk stratification—primarily relying on the identification of bifascicular block—provides limited sensitivity and predictive accuracy for incident CHB. Recent advances in artificial intelligence (AI) within cardiology have demonstrated AI-enhanced ECG (AI-ECG) capabilities to detect various subclinical cardiac abnormalities, heralding transformative potential in prognostication and early diagnosis. Against this backdrop, the study by Sau et al. introduces AIRE-CHB, an AI-ECG-based risk estimator designed to predict incident CHB, aiming to address the limitations inherent in conventional ECG interpretation and improve clinical decision-making.

Study Background and Disease Burden

Complete heart block represents a significant clinical challenge due to its abrupt onset and potential for catastrophic outcomes. The pathophysiology involves interruption of electrical conduction through the atrioventricular node or below, often due to degenerative fibrosis, ischemia, or infiltrative disease. Timely identification of individuals at risk permits prophylactic interventions, including pacemaker implantation, to mitigate adverse events. However, standard risk stratification tools such as identification of bifascicular block on ECG lack robust prognostic performance and do not comprehensively capture subclinical conduction abnormalities that precede overt CHB. The development of AI methodologies capable of mining latent ECG signals to enhance risk prediction addresses an unmet clinical need to stratify patients more accurately and direct appropriate management strategies.

Study Design, Setting, and Participants

This prognostic cohort study pursued a two-phase design comprising development and external validation of the AI-ECG risk estimator for CHB (AIRE-CHB). The institutional derivation cohort originated from Beth Israel Deaconess Medical Center, encompassing a large dataset of 1,163,401 ECGs from 189,539 patients. External validation occurred in the UK Biobank cohort, an established volunteer-based population dataset including 50,641 ECGs from a comparable number of participants. The exposure variable was the raw digital ECG tracing, subjected to AI algorithm training and testing. The endpoint was defined as the diagnosis of new CHB occurring more than 31 days after the ECG examination, ensuring assessment of incident rather than prevalent disease.

Methodological Details

AIRE-CHB employs a residual convolutional neural network architecture optimized with a discrete-time survival loss function tailored for time-to-event prediction. This machine learning framework integrates complex temporal and spatial features from ECG signals to generate individualized risk estimation for incident CHB. The algorithm was trained on the large institutional ECG repository, incorporating censoring and competing risk considerations inherent in survival analyses. External validation tested predictive performance robustness, mortality competing risk adjustments, and comparative effectiveness against traditional ECG markers, specifically presence of bifascicular block.

Key Findings

In the Beth Israel Deaconess cohort, AIRE-CHB demonstrated excellent discrimination for incident CHB with a concordance index (C index) of 0.836 (95% CI, 0.819–0.854) and an area under the receiver operating characteristic curve (AUROC) of 0.889 (95% CI, 0.863–0.916) for prediction within one year. By contrast, bifascicular block showed a substantially inferior AUROC of 0.594 (95% CI, 0.567–0.620), signifying limited clinical utility. Notably, patients stratified within the highest risk quartile by AIRE-CHB exhibited an adjusted hazard ratio (aHR) of 11.6 (95% CI, 7.62–17.7; P < .001) for developing incident CHB compared to those in the lowest quartile, underscoring strong prognostic enrichment.

External validation in the UK Biobank cohort reaffirmed the model's performance, yielding a C index of 0.936 (95% CI, 0.900–0.972) and aHR of 7.17 (95% CI, 1.67–30.81; P < .001). These results indicate superior generalizability and robustness across diverse populations.

Expert Commentary

The introduction of AIRE-CHB marks a significant advance in leveraging AI for cardiac conduction disorder risk stratification. This model’s superior performance compared to conventional bifascicular block assessment provides a compelling argument for integration into clinical workflows, particularly for patients presenting with syncope or those under surveillance for conduction disturbances. The use of a survival loss function is innovative, permitting temporal risk prediction rather than binary classification.

However, some limitations warrant mention. First, despite robust external validation, further studies in different demographic and clinical settings are necessary to confirm reproducibility and evaluate cost-effectiveness. The black-box nature of deep learning models can hinder interpretability, potentially complicating clinician acceptance. Moreover, the study population, largely derived from an academic center and a volunteer cohort, may not fully represent real-world heterogeneity including patients with severe comorbidities or minority groups.

From a mechanistic perspective, AI may be extracting subtle temporal conduction delays, microvolt-level signal alterations, or other latent features imperceptible to human readers, reflecting early pathological remodeling of the conduction system. Thus, AIRE-CHB exemplifies the confluence of clinical cardiology and computational science, harnessing high-dimensional data analytics to enhance predictive precision.

Conclusion

The development and validation of AIRE-CHB demonstrate that AI-enhanced electrocardiography can dramatically improve risk stratification for incident complete heart block, outperforming traditional ECG markers. This novel tool has strong potential to inform clinical decision-making, particularly in populations vulnerable to high-grade atrioventricular block and sudden cardiac events. Future efforts should focus on broader validation, integration with clinical risk factors, and implementation science to optimize patient outcomes. AIRE-CHB may ultimately transform CHB risk assessment from a binary, crude classification to a continuous, individualized prediction, facilitating precision cardiology.

References

Sau A, Zhang H, Barker J, Pastika L, Patlatzoglou K, Zeidaabadi B, El-Medany A, Khattak GR, McGurk KA, Sieliwonczyk E, Ware JS, Peters NS, Kramer DB, Waks JW, Ng FS. Artificial Intelligence-Enhanced Electrocardiography for Complete Heart Block Risk Stratification. JAMA Cardiol. 2025 Aug 20:e252522. doi:10.1001/jamacardio.2025.2522. Epub ahead of print. PMID: 40833775; PMCID: PMC12368796.

Stewart S, Murphy NM. Atrioventricular Block: Epidemiology, Pathophysiology, and Clinical Management. Heart. 2020;106(17):1342-1348. doi:10.1136/heartjnl-2020-316160.

Attia ZI, Noseworthy PA, Lopez-Jimenez F, et al. An Artificial Intelligence-Enabled ECG Algorithm for the Identification of Patients With Left Ventricular Dysfunction. Circulation. 2019;140(24):1894-1904. doi:10.1161/CIRCULATIONAHA.119.039366.