Back to Bigger Picture

BIGGER PICTURE Better care

An ECG alert trial ended with fewer deaths after 90 days

An ordinary ECG can contain patterns linked to a patient’s wider risk. A Taiwanese trial tested a practical use for that signal: send the treating doctor an alert, then measure whether patients fare better.

AI-generated illustration: An ECG-based alert prompts the clinical team to review a patient.
An ECG-based alert prompts the clinical team to review a patient.AI-generated illustration of the idea, not a photograph of the actual people, equipment or event.
Patients analysed
15,965
90-day mortality
3.6% with alerts versus 4.3% with usual care
Alert recipients
39 participating physicians

How the idea works

  1. Analyse a clinical ECG

    A learned model assigns risk from an already ordered ECG.

  2. Reach the treating doctor

    High-risk results trigger a report and direct message.

  3. Investigate and treat

    Clinicians choose further assessment and care.

Start with a test clinicians already order

The trial enrolled emergency and inpatient patients receiving an ECG for clinical reasons at two hospitals. A learned model assessed mortality risk from the trace. It did not prescribe treatment: the intervention was a report and a short warning message when a patient crossed the high-risk threshold.

Sources: [1]

A scarce resource is timely attention

Doctors remained responsible for assessment and treatment. Among patients classified as high risk, the alerted group received more intensive care and more cardiac investigations. The authors could not fully establish which response caused the improvement; increased clinical attention was a plausible part of the pathway.

Sources: [1]

The trial measured deaths, not just prediction accuracy

The randomized analysis included 15,965 patients. Within ninety days, mortality was 3.6% in the intervention group and 4.3% with usual care, a difference of 0.7 percentage points. The reported hazard ratio was 0.83. This is a patient-outcome result, rather than a score on archived ECGs.

Sources: [1]

The workflow must travel with the model

The result comes from a specific Taiwanese hospital setting and alert threshold. It cannot guarantee the same effect where ECG ordering, staffing or responses differ. The transferable opportunity is to test a complete chain: an existing measurement, a targeted warning and a team able to investigate it.

Sources: [1]

The original sources

  1. Nature MedicineAI-enabled electrocardiography alert intervention and all-cause mortality: a pragmatic randomized clinical trial
  2. National Defense Medical CenterPrimary trial author-hosted full text

What to explore next

A therapy chatbot was tested for eight weeks. What changed?The speech recogniser learns the speaker
Explore all 99 applications