AI-ECG-SHD is a multicentre prospective observational diagnostic accuracy study evaluating artificial intelligence-enhanced electrocardiography (AI-ECG) for the detection of structural heart disease in adults undergoing clinically indicated transthoracic echocardiography (TTE).
Participants undergo standard 12-lead ECG acquisition together with ECG recordings obtained using portable or wearable single-, 3-, and 6-lead devices. Established AI-ECG models will be evaluated for the detection of heart failure, pulmonary hypertension and valvular heart disease using TTE as the clinical reference standard, with performance compared across ECG acquisition modalities within the same prospective cohort.
AI-ECG predictions are generated offline and are not provided to treating clinicians or used to alter clinical care. The study will also compare AI-ECG performance with N-terminal pro-B-type natriuretic peptide (NT-proBNP) and evaluate the diagnostic performance of a combined AI-ECG and NT-proBNP approach.
The study aims to recruit 590 participants across six NHS hospital sites. A prespecified subgroup of 203 participants will undergo dedicated venous blood sampling for NT-proBNP; eligible same-day NT-proBNP measurements obtained as part of routine clinical care will also be included where available.
Patients awaiting a transthoracic echocardiogram (TTE) based on clinical grounds as part of their standard care
Age 18-90 years
Able to provide informed consent
Severe arrhythmia or unstable cardiovascular disease
Known severe structural heart disease, defined as any of the following: severe left ventricular systolic dysfunction (LVEF ≤35%); severe valvular heart disease according to current British Society of Echocardiography criteria; high echocardiographic probability of pulmonary hypertension according to current British Society of Echocardiography criteria; or previous structural cardiac intervention, including valve repair, valve replacement/implantation, or surgical repair of a cardiac chamber or congenital structural abnormality
Cardiac implantable electronic device in situ, including a permanent pacemaker or implantable cardioverter defibrillator
Current or recent involvement in another research study
This multicentre prospective observational diagnostic accuracy study will recruit 590 adult inpatients and outpatients aged 18-90 years undergoing transthoracic echocardiography (TTE) as part of routine clinical care. Participants will be recruited from Chelsea and Westminster Hospital, West Middlesex University Hospital, Hammersmith Hospital, St Mary's Hospital, Bristol Royal Infirmary, and Southmead Hospital. Patients with known severe structural heart disease are excluded. Patients with previously diagnosed structural abnormalities that do not meet the prespecified severe structural heart disease exclusion criteria may remain eligible for recruitment.
Cohort
Prospective
Traditional 12-lead Electrocardiogram
Other interventionA study-specific standard 12-lead ECG recording will be obtained with the participant in a semi-reclined position. A 10-second ECG will be acquired and pseudonymised for subsequent offline AI-ECG analysis. AI-ECG outputs will not be used to inform or alter clinical care.
Apple Watch Series 4 Single-lead Electrocardiogram
Other interventionA 30-second single-lead ECG will be recorded using an Apple Watch Series 4. The watch will be worn on the participant's left wrist and the participant will place their right index finger on the Digital Crown during recording. The ECG will be used for subsequent offline AI-ECG analysis and will not inform clinical care.
Eko Core 500 Digital Stethoscope 3-lead Electrocardiogram
Other interventionA 30-second 3-lead ECG will be recorded using the Eko CORE 500 digital stethoscope with the participant seated. The device will be positioned over the upper left parasternal region. The ECG will be used for subsequent offline AI-ECG analysis and will not inform clinical care.
AliveCor KardiaMobile Single- and 6-lead Electrocardiogram
Other interventionTwo 30-second ECG recordings will be obtained using the AliveCor KardiaMobile 6L device: a single-lead recording with the participant's thumbs placed on the upper electrodes, followed by a 6-lead recording with the thumbs on the upper electrodes and the lower electrode in contact with the left knee or ankle. Recordings will be used for subsequent offline AI-ECG analysis and will not inform clinical care.
Phlebotomy for N-terminal pro-B-type natriuretic peptide
Other interventionA single 10 mL venous blood sample will be collected from the prespecified subgroup of 203 participants for measurement of serum NT-proBNP. Samples will be analysed by the relevant hospital clinical biochemistry laboratory and disposed of after testing. Eligible same-day NT-proBNP measurements obtained as part of routine NHS care may also be included without an additional blood sample. NT-proBNP results collected for the study will not be used to alter clinical care.
AI-ECG model classification performance for detection of structural heart disease
Diagnostic performance of the AI-ECG models for detection of heart failure (HF), pulmonary hypertension (PH), and valvular heart disease (VHD) will be evaluated separately across conventional 12-lead, derived single-lead, and device-acquired single-, 3-, and 6-lead ECG modalities. The primary measure of model discrimination will be the area under the receiver operating characteristic curve (AUROC).
Additional AI-ECG performance metrics for detection of structural heart disease
Diagnostic performance will be evaluated separately for each AI-ECG model, disease phenotype, and ECG modality using the area under the precision-recall curve (AUPRC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score.
Comparison of AI-ECG diagnostic performance across ECG modalities
AI-ECG diagnostic performance will be compared across conventional 12-lead, derived single-lead, and device-acquired single-, 3-, and 6-lead ECG recordings obtained from the same participants. Comparisons will include AUROC and AUPRC, together with sensitivity, specificity, PPV, NPV, and F1 score, for each disease phenotype and ECG modality.
NT-proBNP diagnostic performance for detection of structural heart disease
The diagnostic performance of NT-proBNP for detection of HF, PH, and VHD will be evaluated against the same echocardiographic reference standards used for AI-ECG. Performance measures will include AUROC, AUPRC, sensitivity, specificity, PPV, NPV, and F1 score.
Combined AI-ECG and NT-proBNP diagnostic performance
AI-ECG predictions and NT-proBNP will be combined using logistic regression to evaluate the diagnostic performance of a combined digital and biochemical biomarker approach for detection of HF, PH, and VHD. Performance will be assessed using AUROC and AUPRC, with sensitivity, specificity, PPV, NPV, and F1 score also reported for binary model outputs.
Click on the lead sponsor to view all of their trials.
