Weak heart: family doctors helped by an AI that reads the ECG

Why does a weak heart often go unnoticed?

In some heart conditions, the muscle pumps blood less well. Doctors measure this with the "ejection fraction", which shows what share of the blood is sent around the body with each beat. When it is low (50% or less in this study), it is called low ejection fraction.

This condition is often underdiagnosed even though it can be treated. Confirming it usually takes an ultrasound scan of the heart, which is not ordered for everyone.

How was the trial run?

The researchers ran a "pragmatic" trial, meaning it took place in real life, during routine care. One hundred and twenty primary care teams from 45 clinics or hospitals were assigned by chance: some had access to the AI results (181 clinicians), the others worked as usual (177 clinicians).

The AI analysed the ECG, a trace of the heart's electrical activity already done as part of care. The adults involved (22,641 in total) had no known heart failure. The main outcome was a new diagnosis of low ejection fraction within 90 days of the ECG.

The trial is registered under the number NCT04000087, which lets anyone check how it was designed. Clinicians in the intervention arm could see the AI result for their patients' ECGs, while those in the control arm worked without it.

What did the trial show?

The trial met its main goal. Across all patients, diagnosis of low ejection fraction rose from 1.6% in the usual-care group to 2.1% in the AI group.

Among the 6% of patients the AI flagged as likely to have a problem, the gap was clearer: 14.5% diagnosed with usual care, against 19.5% with the AI. In this subgroup, more heart ultrasounds were done (49.6% versus 38.1%), while across all patients the number of ultrasounds stayed similar between the two groups (19.2% versus 18.2%).

In the whole group, the odds ratio was 1.32, which means the odds of a diagnosis were roughly a third higher with the AI. The trial was designed around a single main question, so its answer is easy to read.

What could this mean for patients?

The ECG is quick, inexpensive and widely available. If an AI can spot earlier, in the family doctor's office, the people whose heart deserves a closer look, more patients could reach care sooner.

The results show the tool helps reach a diagnosis in everyday practice. It does not replace the doctor's judgement or the ultrasound, which remains the confirming test.

What are the limits of this study?

The trial counts new diagnoses over 90 days. It does not say whether the patients found live better or longer: that remains to be proven by other work.

The absolute gaps are modest (0.5 percentage points across all patients) and the statistical ranges are fairly wide. The AI was not tested in every kind of practice, and the trial does not say whether these extra diagnoses led to quicker treatment.