Back to Bigger Picture

BIGGER PICTURE Earlier diagnosis

A routine ECG offered a clue to heart weakness nobody had diagnosed

An ECG records electrical activity; an ultrasound shows how well the heart pumps. EAGLE connected the two, looking for people whose routine trace suggested hidden weakness worth checking.

AI-generated illustration: An ECG waveform becomes a clue for further heart-function assessment.
An ECG waveform becomes a clue for further heart-function assessment.AI-generated illustration of the idea, not a photograph of the actual people, equipment or event.
Adults in the trial
22,641
Additional diagnoses
5 per 1,000 screened
Participating sites
45

How the idea works

  1. Record an ECG

    Electrodes measure the heart's electrical activity.

  2. Flag possible weak pumping

    Learned patterns identify candidates for review.

  3. Confirm by ultrasound

    Clinicians arrange imaging and interpret the result.

The ECG was already being taken

Mayo Clinic enrolled patients receiving ordinary ECGs in routine care. A convolutional neural network looked for patterns associated with reduced ejection fraction: the proportion of blood pumped out of the heart chamber with each beat. The possibility was finding a clue without first ordering an ultrasound for everyone.

Sources: [1]

A flag, then a different test

Primary-care teams were randomly assigned to receive AI alerts or continue usual care. A positive flag appeared in the health record and could prompt echocardiography. The second test mattered: AI was a route toward confirmation, not permission to diagnose heart failure from an automated ECG label alone. Clinicians decided how to act.

Sources: [1][2]

Five more diagnoses in every thousand

Across 22,641 adults, the intervention produced five additional low-ejection-fraction diagnoses per thousand patients. The overall proportion receiving an ultrasound remained similar, while testing became more targeted among AI-positive patients. This was a workflow result from clinics, including rural sites, rather than a retrospective accuracy contest on stored recordings.

Sources: [1][2]

Finding weakness is the first step

The trial established additional detection, not improved survival. Benefit after detection depends on confirmation, appropriate treatment and follow-up. It also illustrates why a modest absolute increase can matter: a common, inexpensive test can become a doorway to selective specialist investigation without turning every routine appointment into an expensive imaging programme.

Sources: [1][2]

The original sources

  1. Mayo ClinicTrial demonstrates early AI-guided detection of heart disease in routine practice
  2. Nature MedicineArtificial intelligence–enabled electrocardiograms for identification of patients with low ejection fraction

What to explore next

A green box that helped doctors spot more polypsThe cameras watching for a wildfire's first smoke
Explore all 99 applications