It’s impossible to feel if your blood pressure is too high.
The imperceptibility of high blood pressure, or hypertension, has earned it the nickname “the silent killer.” High blood pressure puts individuals at risk of major cardiac events, like a heart attack or stroke, but can also damage the kidneys, arteries, brains, eyes, and essentially every organ that receives a regular supply of blood (read: all of them).
By some estimates, 1.3 billion people globally (pdf) have high blood pressure. But not all of them know it, because it takes a doctor’s visit to get a diagnosis.
So researchers based at the University of Toronto in Canada and the Affiliated Hospital of Hangzhou Normal University in China are working on a way to track blood pressure as easily as posting on social media. On Tuesday (Aug. 6), they published proof-of-concept that a smartphone app taking a short video can accurately detect blood pressure—at least, in certain kinds of users.
“We’re using a tech called a transdermal optical imaging,” says Kang Lee, a developmental psychologist at the University of Toronto and lead author of the paper. (Lee studies how children learn to lie; his work was originally geared toward a smartphone-based lie detector.) Essentially, light emitted from smartphone cameras bounces off proteins near the surface of the skin at different rates. One of these is hemoglobin, a protein in our blood that shuttles around oxygen. By measuring minute changes in hemoglobin—to the tune of 900 pictures taken over 30 seconds—it’s possible to get a reading of blood pressure.
Based on those measurements, Lee and his team trained a machine learning algorithm that produced blood pressure readings that were accurate roughly 95% of the time. At least, in the population they studied: Over 95% of participants out of a group of 1,328 were of East Asian or European descent.
In other words, the validity of this proof-of-concept paper is still in question for people with darker skin. Another protein found in skin is melanin, which is responsible for skin tone. The smartphone app should be able to tell the difference between light bouncing off melanin and light bouncing off hemoglobin. But without representative testing on individuals with darker skin tones, it’s impossible to say whether it actually can measure blood pressure with the same degree of accuracy.
Racial bias in machine learning is a problem receiving more and more attention. Earlier this year, a paper found that algorithmic bias made it harder for self-driving cars to spot pedestrians of color. And most of the images used to train algorithms that spot cancerous moles are taken of people with light skin—making these tests less effective for people of color.
A biased blood pressure algorithm could be doubly damaging for minority populations. In the US, people in the Latinx and African American communities have higher-than–average rates of hypertension. They’re also less likely to visit a health care provider’s office, where they would ordinarily be diagnosed. Theoretically, an app that can take blood pressure on a smartphone would be a way to bridge this health care gap—but there’s no way it can if the research excludes certain skin types.
Adewole Adamson, a dermatologist at the University of Texas at Austin, has studied how biases in machine learning leave out people of color. “I don’t think that it invalidates the technology,” Adamson says. “But the potential consequences of the well-meaning process of scientists trying to get the research out could have important clinical ramifications.”
Lee and his team acknowledged the fact that there was not enough skin type diversity in their sample to ensure their blood pressure app can work for everyone. Because they cast a small net for research participants, the people included reflect the populations near the researchers’ Canadian and Chinese universities. They hope to find more research partners in different parts of the world who may be able to recruit more diverse participants.
But even if this app could work on people with all skin types, Adamson also pointed out that that doesn’t mean its benefits could reach everyone.
The idea behind cheap at-home blood pressure measurements is to prompt users to see their primary care physicians if they notice a consistently high reading. (Like our sleep cycles, blood pressures have circadian rhythms that ebb and flow daily.) For the model to work, though, users would still need to have access to a smartphone and an affordable health care provider. “Folks of color in the US are the ones who have a high risk of blood pressure and may not know about it,” says Adamson. “So if this fancy test doesn’t reach those people, then they won’t actually accrue or get that benefit.”
The smartphone app Lee and his team created is not yet ready for consumer use. A version of the app, Anura, is available on the iOS and Android app stores for free, but in its current form it only detects a users’ heart rate and stress levels. Lee says it would likely take years to conduct all the testing necessary to make the app applicable for everyone, and even more for regulatory authorities like the US Food and Drug Administration to approve it for regular use.