What is the problem?
In hospital, spotting the most vulnerable patients early is hard. A person can look stable while getting worse, and busy doctors cannot watch every sign at all times.
The electrocardiogram (ECG) records the electrical activity of the heart. It holds more information than the human eye reads day to day, and AIs can now pull risk signals out of it.
The study asked whether an AI-enabled ECG could identify hospitalised patients at high risk of death, so that doctors could act sooner.
What did the researchers do?
They ran a pragmatic randomised trial, meaning one close to everyday hospital life, across several sites. Patients were assigned by chance: for some, doctors received an AI report and warning messages flagging patients at high risk of death; for others, care stayed as usual.
The main goal was to see whether the alert was associated with fewer deaths from any cause within 90 days.
In total, 39 physicians and 15,965 patients took part across several sites, and the trial is registered as NCT05118035. The comparison group received usual care without the AI report.
What were the results?
The trial met its main goal: 3.6% of patients in the alert group died within 90 days, against 4.3% in the comparison group (hazard ratio of 0.83, confidence interval 0.70 to 0.99). A hazard ratio below 1 means fewer deaths in the alerted group.
The effect appeared mostly in patients with a high-risk ECG (hazard ratio of 0.69). In that group, alerted patients received more intensive care, and cardiac deaths were 0.2% against 2.4%, but these figures rest on few events and the uncertainty is wide (confidence interval 0.01 to 0.56 for the hazard ratio).
This was a prespecified analysis, meaning it was planned before the data were examined, which makes the finding more reliable than a pattern spotted afterwards. In the high-risk group, alerted patients also received higher levels of intensive care than the control group, which may be part of how the alert helped.
Key figures
- 15,965hospitalised patients included, cared for by 39 physicians
- 3.6% vs 4.3%deaths at 90 days, with and without the AI alert
- 0.69hazard ratio for death in patients with a high-risk ECG (below 1 means fewer deaths)
What does this change for patients?
Rather than a simple measure of technical accuracy, this trial compares patients assigned by chance and looks at an outcome that matters: survival. The alert seems to help spot high-risk patients and trigger quicker care.
For a patient, nothing changes in what they should do: the alert is meant for the doctor, who keeps the decision.
The authors conclude that the alert assists in detecting high-risk patients, prompting timely clinical care and reducing mortality, while noting the full mechanism is still to be established.
What are the limits?
The authors write that the precise way the alert reduced mortality has yet to be worked out. The main result is an association, with a difference of less than one percentage point between groups, and the upper end of the confidence interval (0.99) comes close to no effect.
Subgroup results, especially on cardiac deaths, rest on few cases. More trials in other hospitals are needed to confirm.
The cardiac death result (hazard ratio of 0.07) comes with a very wide confidence interval, from 0.01 to 0.56, so it should be treated as a lead to confirm rather than a firm figure.
Heart and hospital stays: a question for the team
In this trial, an ECG-based alert was linked to fewer deaths in hospital. It depends on the hospital and is not available everywhere. A quick, well-followed ECG remains a useful reflex.
Mima can keep your ECGs, appointments and prescriptions in one place, and note day by day any breathlessness or tiredness for your cardiologist. A question: "Does my hospital use alerts to monitor inpatients?"
Frequently asked questions
Can AI reduce deaths in hospital?
In this randomised trial, an AI alert based on the ECG was associated with fewer deaths at 90 days (3.6% against 4.3%). It is an association, and the exact mechanism is still to be worked out.
What does an ECG measure?
The electrocardiogram records the electrical activity of the heart using electrodes placed on the skin. It is routinely done in hospital.
Does the doctor always follow the AI?
No: the alert is a message to the doctor, who keeps the decision. The study only reports that alerted patients with high-risk ECGs received more intensive care.
What is a pragmatic randomised trial?
Patients are assigned by chance to one of two groups, in conditions close to real hospital practice rather than in a tightly controlled setting.
The original study
Lin CS, Liu WT, Tsai DJ, Lou YS, Lin C, et al. “AI-enabled electrocardiography alert intervention and all-cause mortality: a pragmatic randomized clinical trial.” Nature Medicine, 29 April 2024. DOI : 10.1038/s41591-024-02961-4
Read the full study in Nature MedicineThis article summarises a scientific study and is not medical advice.