Health Equity 101: The Racial Biases of Smartwatches & Other Healthcare Tech

Smartwatches and personal health trackers have crept onto the wrists of approximately 16 percent of American adults, according to a 2019 report. Beyond looking sleek and keeping track of the time, these devices can also monitor sleep, gauge exercise, and help consumers better understand their health. Employers, insurance companies, and healthcare organizations have begun to take an interest in this data and use it to make better-informed decisions. But the key to many of these devices’ applications lies in their ability to track heart rate—and their heart rate readings are significantly more accurate for white people than for any other demographic.

To understand how unconscious bias is perpetuated, one must look at the underlying systems which support it. When a smartwatch wants to track your heart rate, it relies on optical sensors (mainly, a little green light) strapped around your wrist. As your heart beats, there’s momentarily less blood in your wrist, and therefore more light is reflected into the sensor. Over a short period of time, the smartwatch is able to ascertain your heart rate.

The problem is in the type of sensor used: primarily, photoplethysmographic lighting (known as PPG, or green light monitoring), which is simpler and cheaper than the infrared lights in a hospital’s pulse oximeter. Since green light has a shorter wavelength, it’s less able to penetrate melanin, the naturally-occurring skin pigment which causes darker skin. The darker one’s skin, the less accurate the green light’s reading.

Several studies have shown that green-light monitoring underperforms when encountering melanin. But comparatively little research has been done into how much that underperformance causes error rates in smartwatches. The only two major studies on the subject came to slightly different conclusions. One, conducted in part by Stanford University in 2017, did find statistically significant errors in heart rate tracking of men with darker skin tones. A 2020 study in Digital Medicine, however,found no statistical significance in accuracy across skin tone (though it did find that error rates ran 30 percent higher during activity as opposed to during rest). Academics are calling for more research.

If a single doctor were found to be inaccurately tracking the heart rate of their patients, someone would hopefully call for a swift change. The tech in smartwatches is measuring millions of Americans’ heart rates at once. And while an inaccurate heart rate reading may not seem catastrophic, its side effects can quickly snowball. Bias perpetuates bias. People of color already tend to receive less and lower quality healthcare.

Wearable health trackers weren’t designed to be medical-grade devices, but they’re increasingly being used that way. Fitbits are currently being used in over 350 clinical trials. As of 2018, approximately 21 percent of large employers who offered health insurance were collecting information from their employees’ fitness trackers. Insurance companies have offered lower premiums in return for access to Fitbit data.

Today, some employers have begun to incentivize their employees’ use of fitness trackers and smartwatches through extra vacation days or even lower health insurance premiums. Major companies and institutions include these devices in their wellness programs, and some employers have even tried to make them mandatory.

It’s not all black and white. Some smartwatches, like Apple Watch, employ both green light monitoring and infrared technology, thus achieving a higher rate of accuracy when at rest. Fitbit has made earnest attempts to boost the signal of its green light tech. But one is left to wonder what other inaccuracies are going unnoticed and unattended in modern healthcare technology.

Major problems of bias lurk in the adoption of AI into healthcare. A 2019 study in Science found significant bias in algorithms used to manage the health of populations. Combing through the medical records of an academic hospital, researchers learned that an algorithm was scoring white patients as having higher risk scores than Black patients who were equally sick. This meant white patients were put to the top of the line for treatment of complex conditions like diabetes, while Black patients with the same condition were not. The same flaw was identified in ten of the most widely used algorithms in the industry, which, collectively, applied to over 150 million Americans.

The biases of AI algorithms come from biased data sets: using prior medical histories and cost metrics to arrive at supposedly optimal outcomes. But historical data is not race-blind. In 2009, all but 4 percent of participants in genomic data studies were of European ancestry. When machines learn from biased data, they arrive at unconsciously biased conclusions. In the healthcare arena, this can have serious, even lethal, side effects.

In designing and deploying AI algorithms in healthcare, organizations must examine each system and its foundational data with bias in mind, and take steps to ensure fairness. Specialists in diversity, inclusion, and equity should be consulted. After deploying the algorithm, continuous monitoring is required and feedback from patients and providers is crucial. Left unaddressed, biased data will continue to reinforce the inequities of the healthcare system.

In June 2019, the American Medical Association (AMA) called on the Food and Drug Administration (FDA) to guard against bias in AI, and focus on patient outcomes. That same year, Senator Cory Booker, Senator Ron Wyden, and Rep. Yvette Clark introduced the Algorithmic Accountability Act, which would direct the Federal Trade Commission (FTC) to enforce bias assessments in the private sector. Much as with social injustice, the recognition of unconscious and systemic bias is the first step in correcting it.

Tech has a long history of racial bias, both big and small: from Kodak color film smudging the faces of those with darker skin to automated soap dispensers not recognizing hands with higher melanin counts. Today, the inconveniences of those biases have potentially lethal ramifications, whether through a failure to provide proper blood pressure readings or through a driverless car that doesn’t recognize a darker skin tone.

In health technology, the margin of error is as thin as the blade of a scalpel. As the nation begins to confront its unconscious social biases, that introspection must extend to its technological developments as well.

Matt Zbrog

Writer

Matt Zbrog is a writer and researcher from Southern California. Since 2018, he’s written extensively about trends within the healthcare workforce, with a particular focus on the power of interdisciplinary teams. He’s also covered the crises faced by healthcare professionals working at assisted living and long-term care facilities, both in light of the Covid-19 pandemic and the demographic shift brought on by the aging of the Baby Boomers. His work has included detailed interviews and consultations with leaders and subject matter experts from the American Nurses Association (ASCA), the American College of Health Care Administrators (ACHCA), and the American Speech-Language Hearing Association (ASHA).