Harnessing ECG Artificial Intelligence for Rapid Treatment and Accurate Identification of Structural Heart Disease (HEART-AI)

Harnessing ECG Artificial Intelligence for Rapid Treatment and Accurate Interpretation, an Open Label Randomized Controlled Trial

The HEART-AI (Harnessing ECG Artificial Intelligence for Rapid Treatment and Accurate Interpretation) is an open-label, single-center, randomized controlled trial, that aims to deploy a platform called DeepECG at point-of-care for AI-analysis of 12-lead ECGs. The platform will be tested among healthcare professionals (medical students, residents, doctors, nurse practitioners) who read 12-lead ECGs. In the intervention group, the platform will display the ECHONeXT structural heart disease (SHD) scores in randomized patients to help doctors prioritize transthoracic echocardiography (TTEs) or magnetic resonance imaging (MRI) and reduce the time to diagnosis of structural heart disease. Also, this platform will display the DeepECG-AI interpretation which detects problems such as ischemic conditions, arrhythmias or chamber enlargements and acts an improved alternative to commercially available ECG interpretation systems such as MUSE.

Our primary objective is to assess the impact of displaying the ECHONeXT interpretation on 12-lead ECGs on the time to diagnosis of Structural Heart Disease (SHD) among newly referred patients at MHI. We will compare the time interval from the initial ECG to SHD diagnosis by transthoracic echocardiogram (TTE) or magnetic resonance imaging (MRI) between patients in the intervention arm (where ECHONeXT prediction of SHD and TTE priority recommendation are displayed) and patients in the control arm (where ECHONeXT prediction and recommendation are hidden).

The main secondary objective is to evaluate the rate of SHD detection on TTE or MRI among newly referred patients. We also aim to assess the delay between the time of the first ECG opened in the platform and the TTE or MRI evaluation among newly referred patients at high or intermediate risk of SHD.

By integrating an AI-analysis platform at the point of care and evaluating its impact on ECG interpretation accuracy and prioritization of incremental tests, the HEART-AI study aims to provide valuable insights into the potential of AI in improving cardiac care and patient outcomes.

Study Overview

• Status: Enrolling by invitation
• Conditions: Structural Heart Disease; Structural Heart Abnormality
• Intervention / Treatment: Other: ECHONEXT

Detailed Description

The HEART-AI (Harnessing ECG Artificial Intelligence for Rapid Treatment and Accurate Interpretation) study primarily aims to assess the effect of displaying the ECHONeXT interpretation on the time interval from the initial ECG to the rate of Structural Heart Disease (SHD) diagnosis on transthoracic echocardiograms or magnetic resonance imaging.

We will achieve this by comparing the time between the first ECG and diagnosis of SHD on TTE or MRI between the intervention group, where the ECHONeXT interpretation is displayed to users, and the control group, where it is not displayed, thereby quantifying the influence of AI-supported diagnostics on clinical decision-making and patient management strategies.

For the purpose of the study, SHD will be defined as presence of any of the following on TTE or MRI:

• LVEF ≤ 45%
• Mild, moderate or severe RV Dysfunction

• The presence of one or multiple valvulopathies in this list:

  • Moderate-to-severe pulmonary regurgitation
  • Moderate-to-severe tricuspid regurgitation
  • Moderate-to-severe mitral regurgitation
  • Moderate-to-severe aortic regurgitation
  • Moderate-to-severe aortic stenosis
• Moderate or severe pericardial effusion (Tamponade or any effusion > 1 cm)
• LV wall thickness ≥ 1.3 cm
• Apical cardiomyopathy
• Pulmonary hypertension as defined using the systolic pressure of the pulmonary artery greater or equal to 25 mm Hg on TTE.

• Study Type: Interventional
• Enrollment (Estimated): 16160
• Phase: Not Applicable

Contacts and Locations

Study Locations

• Canada
  • Quebec
    • Montreal, Quebec, Canada, H1T1C8
      • Montreal Heart Institute

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• Users

  1. Users who are providing clinical care and who read ECGs as part of their practice.
  2. Users who have provided informed consent to participate in the study.
  3. Users who have completed the required training on the use of the DeepECG platform.

ECG

1. 12-lead ECGs recorded during the study period at the Montreal Heart Institute.
2. ECGs of adequate technical quality for interpretation, as determined by the recording software and visual inspection.

Patients

1. Patients aged 18 years or older

Additional Inclusion criteria for the randomization part of the study

1. Outpatients or patients who presented at the ambulatory emergency department. The location will be determined according to the ECG where it was recorded which is entered by the ECG technician. These locations will be included for the eligibility of the randomization:

   a. locations_to_keep = ['21_URGENCE AMBULATOIRE', '1_CARDIOLOGIE GENERALE', "17_CLINIQUE D'ARYTHMIE"]

2. New patients without a prior formal evaluation by a cardiologist or internal medicine specialist for suspected or provisionally identified cardiac conditions, including:

   1. Arrhythmia
   2. Heart Failure
   3. Coronary Artery Disease
   4. Valvular Heart Disease
   5. Cardiomyopathy
   6. Other cardiac conditions

3. Patients with previous TTE or MRI:

   1. Have no documented history of any cardiac condition
   2. No transthoracic echocardiogram or MRI in the last 24 months (from any center)

Exclusion Criteria:

Users

1. Users who are unable to commit to the duration of the study (approximately 1 month minimum) or adhere to the study protocol.

Additional Exclusion criteria for the randomization part of the study ECG

1. ECG with too many artefacts or without any QRS visible as interpretated by the MUSE GE algorithm.

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Diagnostic
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: None (Open Label)

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
No Intervention: No ECHONEXT interpretation; Not displaying the ECHONEXT algorithm interpretation.
Experimental: ECHONEXT interpretation; The ECHONeXT algorithm was trained to predict the presence of SHD on TTE using a single 12-lead ECG. It was developed at Columbia hospital, released as open-weights and validated at the MHI. It was trained on 800,000 ECG and TTE pairs.	Other: ECHONEXT; ECHONEXT Artificial intelligence algorithm

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Assess the effect of displaying the ECHONeXT interpretation on the time to diagnosis of Structural Heart Disease (SHD)	Time interval from the first ECG opened in the platform to SHD diagnosis on TTE or MRI, calculated as: Date of SHD diagnosis on TTE - Date of access of the first ECG where an ECHONeXT interpretation was available and a user consulted the ECG	18 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Evaluate the effect of displaying the ECHONeXT interpretation on the delay between the ECG and the TTE evaluation for patients at high or intermediate risk of SHD	Delay between the time of the first ECG opened in the platform and the TTE calculated as: Date of TTE evaluation - Date of access of the first ECG where an ECHONeXT interpretation was available and a user consulted the ECG	18 months
Assess the effect of displaying the ECHONeXT interpretation on the rate of SHD diagnosis on TTE	Diagnosis of SHD (Yes/No) on TTE	18 months
Assess the agreement of the users with the ECG-AI algorithm's interpretations	Agreement (Yes/No) of the user with the ECG-AI algorithm's interpretation. Agreement is defined as the user clicking on "thumbs up" on the platform.	18 months
Determine the acceptability and usability of the DeepECG platform in clinical practice based on the end-of-study survey	Questions of the end-of-study survey on the usability and appreciation of the DeepECG platform and the ECHONeXT interpretation	18 months
Determine the primary endpoint stratified according to the presence of a previous TTE > 24 months or no previous TTE (brand new patients)	Questions answer on the pre-ECG questionnaire	18 months

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Describe the engagement of users and the overall utilization of the DeepECG platform algorithm in the clinical setting	Number of ECGs accessed per user Number of days per user with at least 1 ECG accessed using the platform over total number of days the user is in the study (i.e. has access to the platform).	18 months
Compare the TTE priority classification assigned by the user between the intervention and the control group	TTE priority classification (A, B, C, D, E, etc.) assigned by the user on the post-ECG questionnaire to the first ECG recording where an ECHONeXT interpretation was available and a user consulted the ECG	18 months
Compare the TTE priority classification assigned by the user between the intervention and the control group stratified by location (emergency vs outpatient	TTE priority classification (A, B, C, D, E, etc.) assigned by the user on the post-ECG questionnaire to the first ECG recording where an ECHONeXT interpretation was available and a user consulted the ECG.	18 months
Review additional qualitative feedback and insights captured after reading an ECG	Narrative description put in the "other" field of the post-ECG questionnaire	18 months
Assess the agreement of the user with the ECG-AI algorithm's interpretations according to practice type, number of years in practice, age of user and familiarity with AI tools (based on end of study questionnaire)	Agreement (Yes/No) of the user with the ECG-AI algorithm's interpretation. Agreement is defined as the user clicking on "thumbs up" on the platform.	18 months
Describe the agreement of the user with the ECG-AI algorithm's interpretation according to the diagnosis category of the ECG-AI (ischemic, arrythmia, chamber enlargement, structural heart disease, other)	Agreement (Yes/No) of the user with the ECG-AI algorithm's interpretation. Agreement is defined as the user clicking on "thumbs up" on the platform.	18 months
Evaluate the effect of displaying the ECHONeXT interpretation on the delay between the ECG and the TTE evaluation, by subgroups defined by the risk level of SHD (low/ intermediate/ high)	Delay between the time of the first ECG opened in the platform and the TTE calculated as: Date of TTE evaluation - Date of access of the first ECG where an ECHONeXT interpretation was available, and a user consulted the ECG	18 months
Evaluate the model performance after applying continual learning	We will re-train the DeepECG models using examples that were downvoted by users or that users " bookmarked " in addition to the previous ECGs that were used for training the model. Model performance will be compared using the DeLong test for the AUC and AUPRC before and after retraining the model. Users will not be exposed ot this new model.	18 months
Describe the agreement of the user with the ECG-AI algorithm's interpretation in the two subgroups.	Agreement (Yes/No) of the user with the ECG-AI algorithm's interpretation Agreement is defined as the user clicking on "thumbs up" on the platform.	18 months
Sensitivity and specificity of ECHONeXT to detect SHD on TTE	Assess the sensitivity and specificity of ECHONeXT to detect SHD on TTE	18 months

Collaborators and Investigators

• Sponsor: Montreal Heart Institute

Publications and helpful links

Helpful Links

• Related Info

Study record dates

Study Major Dates

• Study Start (Actual): April 16, 2025
• Primary Completion (Estimated): January 31, 2026
• Study Completion (Estimated): January 31, 2027

Study Registration Dates

• First Submitted: June 12, 2024
• First Submitted That Met QC Criteria: June 12, 2024
• First Posted (Actual): June 17, 2024

Study Record Updates

• Last Update Posted (Actual): July 31, 2025
• Last Update Submitted That Met QC Criteria: July 28, 2025
• Last Verified: July 1, 2025

More Information

Terms related to this study

• Keywords: magnetic resonance imaging; artificial intelligence; transthoracic echocardiogram
• Additional Relevant MeSH Terms: Cardiovascular Diseases; Heart Diseases
• Other Study ID Numbers: HEART-AI-001

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No