A new prospective trial shows consumer wearables can flag silent, life-threatening heart structural issues before symptoms even appear.

Can a consumer smartwatch do the job of a clinic-grade ultrasound? For years, consumer wearables have been relegated to simple heart rate tracking and basic rhythm checks. This prospective validation of an AI-ECG model challenges that limitation, shifting the Apple Watch from a simple heart-rhythm tracker into a tool that can flag structural heart disease.

The clinical implications are highly disruptive. While previous research relied on clean, retrospective data, this real-world test proves that noisy, single-lead signals can reliably detect deep structural failures. It suggests we are close to a future where passive wearable monitoring acts as an active triage system for major cardiac dysfunction.

Testing the wearable AI

The WATCH-SHD study tested 596 adults at a Yale New Haven Hospital clinic who were already scheduled for routine echocardiograms. The cohort had a median age of 62 years and was 51.2% women. Among these patients, 5.1% (30 individuals) actually had severe structural heart disease, defined as severe valve disease, severe muscle thickening, or a weak pump with an ejection fraction under 40%. During their clinic visit, patients took a quick 30-second ECG on an Apple Watch, which was then processed by a noise-adapted AI model.

How the AI performed

The AI model analyzed these noisy, single-lead recordings with high accuracy, proving that clinic-grade diagnostic power can live on a wristband.

• The model achieved an overall discrimination score (AUROC) of 0.841.
• It caught 76.7% of severe structural heart disease cases.
• It correctly cleared 83.2% of healthy patients.
• The negative predictive value reached 98.5%, meaning a clean bill of health from the AI was highly reliable.
• The positive predictive value was low, sitting at 19.7%.

The screening efficiency shift

That 19.7% positive predictive value is the most critical detail for analysts to parse. It means that for every five people the AI flags, four will actually have normal hearts. In a mass consumer market, this could trigger a wave of unnecessary, expensive echocardiograms. However, the true value of this technology lies in screening efficiency, as the AI-guided strategy reduced the number of patients needed to test to find one true case by more than 60% compared to usual care.

This builds directly on earlier retrospective work, such as the multinational validation of deep learning for noisy ECGs, which proved the math worked in theory. Now we know it holds up in a live clinical setting. It also complements broader efforts in detecting structural heart disease using AI from standard clinic charts.

We must remain honest about the limitations of this trial. This was a single-center study at an active cardiology clinic, where the baseline rate of heart disease is much higher than in the general public. If deployed to millions of casual smartwatch users, the false-positive rate will balloon, potentially overwhelming clinics with worried-well patients. Still, the clinical trajectory is clear: wearables are moving from fitness novelty to serious clinical gatekeepers.

Read the full study in medRxiv.