AI-Enhanced Electrocardiogram Successfully Predicts Complete Heart Block

AIRE-CHB, an artificial intelligence-enhanced electrocardiogram (AI-ECG) designed for complete heart block (CHB), has outperformed standard ECGs in identifying the risk of incident CHB in adult patients.1

ECGs are the current gold standard for diagnosing CHB, and in the absence of reversible causes, the standard treatment is permanent pacemaker implantation. However, ECGs frequently fail to show evidence of high-grade atrioventricular (AV) block at the time of recording. Ambulatory ECG recording has been recommended in patients with syncope and suspected conduction disease, but sensitivity is often low and represents limited health-economic value.1,2

“Artificial intelligence-enhanced ECG (AI-ECG) has been recently shown to have significant potential to diagnose hidden cardiovascular disease and predict the risk of future disease,” Arunashis Sau, PhD, academic clinician at Imperial College London, and colleagues wrote. “Recently, we have developed the AI-ECG risk estimation platform (AIRE) capable of predicting mortality and cardiac events from 12 lead ECGs. Here, we build on this platform to predict incident CHB and explore the underlying biology behind these predictions.”1

Investigators examined 2 main groups – the Beth Israel Deaconess Medical Center (BIDMC) cohort, which was an unselected secondary care cohort consisting of routinely collected health record data, and the UK Biobank (UKB) cohort, which was a longitudinal study of volunteers aged 40-69 years. Of these, patients who lived near an imaging assessment center were invited for a second follow-up visit, wherein they had ECG performed.1

Analysis utilized a single ECG per participant. Incident CHB was defined as a new diagnosis of CHB at >31 days after the ECG (to exclude CHB cases diagnosed at the time of hospital presentation). Participants with pacemakers or high-grade AV block (second degree or complete) diagnosed ≤31 days after ECG were excluded, as were those with <31 days follow-up.1

The primary outcome of the study was AIRE-CHB successfully predicting CHB diagnoses based on International Classification of Diseases (ICD) codes. Sensitivity analyses were performed via pacemaker implantation, and both CHB and pacemaker implants were used as alternative endpoints. Syncope history in BIDMC was determined using ICD codes.1

The BIDMC cohort included 1,163,401 ECGs from 189,539 participants; a total of 34,938 participants (18.4%) died during follow-up. The internal testing dataset for incident CHB (BIDMC, n = 61,676; 791 events [1.3%]). AIRE-CHB predicted incident CHB with a C index of 0.836 (95% CI, 0.823-0.853) over a mean (standard deviation [SD]) follow-up of 6.66 (5.78) years. Participants in the high-risk quartile had an age- and sex-adjusted HR of 11.6 (95% CI, 7.62-17.7; P <.001) for incident CHB development compared with the low-risk group.1

Sensitivity analyses for the prediction of pacemaker implantation and CHB or pacemaker implantation yielded similar results (C index, 0.799; 95% CI, 0.779-0.819, and C index, 0.814; 95% CI, 0.798-0.829, respectively). The AUROC for prediction of incident CHB at year 1 was 0.889 (95% CI, 0.863-0.916), and the presence of bifascicular block had an AUROC of 0.594 (95% CI, 0.567-0.620; P <.001).1

External validation was performed via the UKB cohort, which consisted of 50,641 participants. Of these, 35 (0.07%) developed CHB during a mean (SD) follow-up of 4.01 (2.16) years. The C index for incident CHB prediction was 0.936 (95% CI, 0.899-0.972). Participants in the high-risk quartile had an age- and sex-adjusted HR of 7.17 (95% CI, 1.67-30.83; P <.001) for development of incident CHB compared to the low-risk group.1

Investigators also investigated a subset of the BIDMC test group who had ECGs performed in the emergency department and a history of syncope in the 30 days prior (n = 2417). Individuals with an event (n = 17) or ≥1 year of follow-up, AIRE-CHB successfully discriminated the risk of CHB diagnosis within 1 year with an AUROC of 0.832 (95% CI, 0.743-0.908). Binary AIRE-CHB predictions showed a sensitivity and specificity of 0.706 (95% CI, 0.440-0.897) and 0.835 (95% CI, 0.819-0.849), respectively.1

“We have shown AIRE-CHB has the ability to risk stratify individuals, both on a population level and more selected populations, including those with syncope,” wrote Sau and colleagues. “The AIRE platform, including AIRE-CHB, could be applied to patients presenting with syncope to identify those most likely to have intermittent CHB.”1

References

1. Sau A, Zhang H, Barker J, et al. Artificial Intelligence–Enhanced Electrocardiography for Complete Heart Block Risk Stratification. JAMA Cardiol. Published online August 20, 2025. doi:10.1001/jamacardio.2025.2522

2. Muntané-Carol G, Okoh AK, Chen C, et al. Ambulatory Electrocardiographic Monitoring Following Minimalist Transcatheter Aortic Valve Replacement. JACC Cardiovasc Interv. 2021;14(24):2711-2722. doi:10.1016/j.jcin.2021.08.039