From the Ward to the Wrist: How Sumbul Desai Is Rewriting Preventive Care | Department of Medicine News

A First-in-the-World Hypertension Predictor

Desai’s newest headline-making project is a wrist-based predictor for possible hypertension. Apple Watch uses its optical heart sensor to measure pulse wave signals beat by beat, then applies machine learning tuned over years of studies totaling over 100,000 participants and a 2,000 person study conducted with traditional cuffs for ground truth, to estimate whether someone’s typical blood pressure pattern sits in stage 1 or 2 hypertension.

One unique challenge was deciding how to balance accuracy with usefulness in the clinic. At consumer scale, false positives can erode trust and flood clinics. So Desai’s team dialed toward “high specificity,” which roughly translates to the idea that when a user is notified, it should really mean something, and lower sensitivity be accepted as the price of credibility. After a notification, users are guided to log home blood-pressure readings for seven days before seeing a clinician, aiming to replace the usual “come back with a log” visit.

Critics will focus on sensitivity, Desai knows. But she returns to the counterfactual: most undiagnosed people currently get no medical alert or notifications. “Better than zero,” she says, is a defensible first step, “if you’re transparent about performance, publish methods, and design the experience to be clinically useful.”

Science You Can Cite 

Apple may very well be famous for secrecy; medicine, by necessity, is not. Desai knew it was important for the company to “show its work” when it came to health. That started with the Apple Heart Study in collaboration with Stanford, a massive investigation of irregular rhythm notifications that enrolled hundreds of thousands in months and widely published its results. Since then, Apple’s health group has paired regulated features with public white papers and, when appropriate, peer-reviewed publications.

None of this is academic theater. It’s a precondition, Desai argues, for clinical adoption. If physicians can’t see how a feature performs, across age, skin tones, and geographies, they won’t trust it. And if users get numbers they can’t act on, they’ll tune out. 

The Principles Behind Every Product

Desai outlined the three tests every feature must pass:

Scientific validity: The team looks at mountains of peer-reviewed publications, and months and years of testing and trials for new features. Every feature is tested in feasibility studies and then validated in large, diverse cohorts.

Actionability: Insights and data must lead to a behavior change, decision, or clinical conversation. Otherwise they’re just noise.

Privacy by default: Processing happens on-device; Apple doesn’t see health data from these features. 

Those principles are why some popular asks don’t ship: if a metric can’t be interpreted reliably in everyday life, or would provoke anxiety without a clear next step, her team says no.