- Design
- Systematic review and meta-analysis of 3 randomised trials
- Population
- 37,467 adults without known heart failure having routine ECGs
- Primary outcome
- New echo-confirmed ejection fraction of 50% or less
- Effect
- RR 1.36 (95% CI 1.12–1.64), I² 0%; number needed to screen 204
This meta-analysis pooled three randomised trials, two individually randomised and one cluster trial, covering 37,467 adults without known heart failure who had an ECG in routine care. In the intervention arms, an artificial-intelligence algorithm flagged ECGs suggesting a low ejection fraction to the treating clinician.
New diagnoses of an ejection fraction of 50% or less, confirmed by echo, rose by about 36% with screening, and the result was consistent across trials. About 200 people needed screening to find one extra case. Echo referral rates did not rise overall, which suggests the tool redirected echoes towards higher-yield patients rather than adding to the queue.
The weak point is positive predictive value: only one in three to one in five flagged patients turned out to have a low ejection fraction. Three trials is a small base, they used a narrow range of algorithms, and none showed that earlier diagnosis improved outcomes.
For practice: a flag is a reason to arrange an echo, not a diagnosis. Health systems weighing these tools should expect most flags to be false and plan echo capacity accordingly.
- Treat an AI-ECG flag for low ejection fraction as a prompt for echocardiography, not a diagnosis of heart failure.
- Expect most flags to be false positives: positive predictive value was 20–34% in the trials.
- Ask about breathlessness, oedema and fatigue in flagged patients before booking the echo, to set its urgency.
- Start guideline-directed therapy only once the echo has confirmed reduced function.
Why it matters
A test already done on most patients could find treatable heart failure years earlier, if the false-positive load is managed.
Don't overread it
More diagnoses is not the same as better outcomes; no trial measured hospitalisation or death.
The statistics, in plain English
A risk ratio of 1.36 means 36% more new diagnoses with screening, and I² of 0% means the three trials agreed. A number needed to screen of 204 means about 200 ECGs read by the algorithm to find one extra case. The pooled data cannot say whether those extra diagnoses led to fewer admissions or deaths.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for cardiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free