Light sensors on wearables struggle with dark skin and obesity

Apple Watch | Photo by Amelia Holowaty Krales / The Verge

Light sensors used to track heart rate in wearable devices like the Apple Watch Series 5 and the Fitbit Versa 2 don’t work as well on darker skin or people with obesity, according to modeling done in a new study.

That’s a problem for features currently built into devices — but it’s also a bad sign for efforts to use light sensors for new applications in wearables, like monitoring blood pressure, says study author Jessica Ramella-Roman, an associate professor studying bioimaging sensors at Florida International University.

“The architecture of the device has to change,” she says.

The study looked at the photoplethysmography (PPG) signal, a technique that uses changes in how light is reflected to measure blood flow, on three wearable devices: the Apple Watch Series 5, Fitbit Versa 2, and Polar M600. The research team used a model that simulates how light moves through tissue to show how the sensors on those devices would behave with different skin properties. Darker skin has higher melanin and absorbs more light, while skin from people with obesity tends to be thicker, have less water, and have less blood flow than skin from people who are not obese. While previous research on accuracy and bias in wearables has focused on skin tone, Ramella-Roman says that many studies haven’t included a lot of people with obesity despite these physiological differences.

“That’s why we think it’s necessary to focus there,” she says.

The PPG signal didn’t change much as skin tone changed, the researchers’ model found: it varied less than 10 percent across the devices. But modeling obesity, both alone and together with skin tone, caused up to a 60 percent variation in the signal. Signal loss appeared to be because of the changes in skin thickness in people with obesity, the authors said in the study. There were changes in the peak of the PPG signal, which is used to calculate heart rate but whose signal strength should not change based on the heart rate value. There were also changes in the shape of the signal — which different groups are using as a way to track blood pressure.

“As we increased the BMI level and increased the skin tone, the signal diminished, and then other features started disappearing as well,” she says. The Fitbit, which had fewer sensors, had a more dramatic loss of signal than the Apple Watch, the study found.

This study only modeled how these wearable devices would detect signals in the lab, Ramella-Roman stressed. The team still has to check the devices on actual people in order to confirm the findings. They’re in the process of doing that study now and have enrolled around 100 people so far, she says.

But the problems revealed by analyses like this complicate projects looking to use wearables to track cardiovascular health for underserved groups, Ramella-Roman says. The new study indicates researchers have to be careful using PPG, particularly for projects aiming to use the devices to help monitor people at higher risk of cardiovascular issues — which includes people with obesity. “The holy grail in this context is looking at blood pressure,” she says. “But many, many studies looking at blood pressure use some combination of PPG and other modalities.”

The findings also suggest that other devices that use light sensors and PPG, like the blood oxygen measures in hospitals or doctors’ offices, might not work as well for people with obesity. There are very few studies that evaluate the devices in those groups.

The good news is it should be possible to adjust devices (like by moving around the spacing of the sensors) so that they are more accurate on people with darker skin or with obesity without compromising accuracy for other groups, Ramella-Roman says.

“The way the systems were designed initially maybe didn’t account for these individuals, but they could definitely make changes,” she says. “I don’t see anything that would limit that.”