A Computer Alert Got More Patients to a Heart Team. The Real Question Is What Happened Next.

One condition targeted by the system was aortic stenosis, a narrowing of the heart’s aortic valve.

Aortic stenosis becomes more common with age. Severe disease can cause shortness of breath, chest discomfort, dizziness, fainting and other serious problems.

But symptoms can develop gradually.

An older person who becomes less active may simply think, “I’m getting older.”

Sometimes that’s true.

Sometimes it isn’t.

An echocardiogram—an ultrasound examination of the heart—can reveal significant valve disease. The challenge is making sure an important finding leads to appropriate follow-up.

That’s the gap the ALERT system was designed to address.

The ALERT trial was a cluster-randomized study involving 765 clinicians and 2,016 echocardiograms across 35 hospitals in five U.S. health systems.

The system used was Tempus Next, an AI-enabled clinical platform. It analyzed information from echocardiogram reports and medical records to identify patients who might meet guideline-based criteria for significant valve disease but did not appear to have an appropriate treatment plan.

The system then generated an electronic notification to the clinician.

This distinction matters.

The trial was not testing whether an AI system could independently diagnose someone or decide that the person needed a new heart valve.

The automated system identified possible cases using clinical information and criteria, and the resulting alert put the issue in front of a human clinician.

The clinician still had to decide what happened next.

This is where the study produced a meaningful result.

Within 90 days, patients whose clinicians received the alerts were more likely to have a multidisciplinary heart-team evaluation:

22.7% vs. 17.9% — an absolute difference of 4.8 percentage points.

They were also more likely to undergo valve intervention:

13.4% vs. 9.6% — an absolute difference of 3.8 percentage points.

So the alert helped move more patients into the evaluation and treatment pathway.

But look at those numbers carefully.

Even with the alerts, only 13.4% underwent a valve intervention during the study period.

The computer didn’t suddenly identify a huge group of people who all needed immediate procedures.

It nudged more patients toward the next appropriate step.

That’s a much more accurate—and arguably more interesting—description of what happened.

The ALERT trial’s main outcome was a 90-day measure of the care process: whether patients received a valve intervention or a multidisciplinary heart-team evaluation.

Those are meaningful outcomes because getting the right patient to the right specialist is an important part of good care.

But they aren’t the same thing as showing that patients ultimately:

• lived longer,
• had fewer heart attacks or strokes,
• felt better,
• stayed out of the hospital longer, or
• had a better quality of life.

The trial did not establish those things.

So it would be premature to say that the technology improves patient outcomes.

What we can say is that it improved the process of getting some patients evaluated and treated.

That’s still worth paying attention to.

Sort of.

Tempus describes Tempus Next as an AI-enabled platform, and the ALERT study used that platform to analyze echocardiogram information and generate notifications.

But the particular clinical trigger wasn’t an autonomous AI diagnosis.

The system used echocardiographic findings and guideline-related clinical criteria to identify possible cases, after which clinicians reviewed the information and decided what to do.

So this isn’t a story about a computer replacing a cardiologist.

It’s a story about using automation to make it harder for an important finding to disappear into a medical record.

That may ultimately prove to be one of the more useful applications of AI in healthcare.

The commercial relationships are worth knowing.

The ALERT trial was funded by Tempus AI and Medtronic. Tempus developed the Tempus Next platform used in the study. Medtronic sells structural-heart products, including TAVR systems.

That doesn’t mean the results are wrong.

It does mean readers should know who was involved financially when evaluating claims about a technology that can identify patients who may ultimately be candidates for valve procedures.

The system itself was designed to be device-agnostic, meaning clinicians were not required to use a particular manufacturer’s valve.

That’s exactly the sort of context a health watchdog should provide.

TAVR, or transcatheter aortic-valve replacement, is a minimally invasive procedure used to replace a diseased aortic valve in appropriate patients.

Recent research has also challenged the old assumption that patients with severe aortic stenosis but no symptoms should always simply be watched.

In the EARLY TAVR randomized trial, 901 patients with asymptomatic severe aortic stenosis were assigned to early TAVR or clinical surveillance.

The combined outcome of death, stroke or unplanned cardiovascular hospitalization occurred in:

26.8% with early TAVR vs. 45.3% with surveillance.

But here’s the part that’s easy to miss.

Death occurred in 8.4% vs. 9.2%.

Unplanned cardiovascular hospitalization occurred in 20.9% vs. 41.7%.

So much of the difference in the combined outcome came from fewer cardiovascular hospitalizations, rather than a large difference in deaths.

That doesn’t make the finding unimportant. It simply means the results shouldn’t be presented as though early TAVR had been shown to dramatically reduce mortality.

It didn’t.

And EARLY TAVR was answering a different question from ALERT. EARLY TAVR studied when to replace the valve. ALERT studied whether an automated alert could help get patients into the evaluation and treatment pathway.

Finding more disease isn’t automatically the same thing as making people healthier.

That matters particularly when the disease is found before symptoms appear.

The more aggressively a healthcare system searches for abnormalities, the more important it becomes to know which findings actually benefit from treatment and which are better monitored.

That’s why the next step in research matters so much.

We need to know whether getting more patients through the evaluation pathway ultimately produces better outcomes for patients, not simply more activity within the healthcare system.

The useful lesson isn’t that older adults should ask for an AI scan.

It’s simpler than that.

If an echocardiogram finds significant valve disease, the finding deserves appropriate follow-up.

And if an older person develops unexplained shortness of breath, dizziness, fainting, chest discomfort or a noticeable decline in exercise tolerance, those symptoms shouldn’t automatically be dismissed as “just getting older.”

Sometimes there is a treatable heart problem underneath.

Technology may help healthcare systems find more of those problems.

But the technology isn’t the endpoint.

The endpoint is what happens to the patient afterward.

The ALERT trial found that an automated electronic alert helped more patients with significant valve disease reach a heart team and increased valve interventions by about 4 percentage points. That’s a real improvement in the care process—but it isn’t proof yet that the technology helps people live longer or feel better.

The system used an AI-enabled platform, but the alert depended on clinical findings and criteria and still required human judgment. The study was funded by Tempus AI and Medtronic, which is important context for readers to know.

The promising part isn’t that a computer can replace a doctor. It’s that a computer may help a doctor miss fewer important findings. The real test is whether that leads to better outcomes for patients.