In recent years, there’s been a veritable explosion in the number and type of health monitoring devices available in smartphones and fitness apps.
Your smartphone is likely tracking the number of steps you take, how far and fast you walk, and how many flights of stairs you climb each day. Some phones log sleep, heart rate, how much energy you’re burning, and even “gait health” (how often are both feet on the ground? how even are your steps?). And, of course, nonphone wearables and fitness gadgets are available, such as devices to measure your heart rhythm, blood pressure, or oxygen levels. The accuracy of these devices varies — and, in some instances, your skin tone may make a difference.
Generally, how accurate are health monitors?
I know from my experience with hospital monitoring devices that they aren’t always accurate. False alarms from EKG monitors often send medical staff scurrying into patient rooms, only to find the patient feeling fine and surprised about the commotion. A particularly common false alarm is a dangerous and unstable heart rhythm on a continuous heart monitor, which can be due to the motion from a patient brushing their teeth.
High-stakes devices with monitoring capability, such as defibrillators and pacemakers, are extensively tested by their makers and vetted by the FDA, so their accuracy and reliability are generally quite good.
But what about home health monitoring devices intended for consumer use that are not extensively tested by the FDA? Ever count your steps for a few minutes just to see if your phone’s tally agrees? Or climb a couple of flights of stairs to see if you are getting full credit for not taking the elevator?
The accuracy of consumer devices depends in part on what is being monitored. For example, one study assessed the accuracy of heart rate monitors and energy expenditure calculators in phones and health apps. Accuracy was quite high for heart rate (often in the range of 95%), but much less accurate for energy expenditure. Accuracy can also vary depending on who is being monitored.
Device bias: What it is and why it occurs
While no health gadget is perfect, some users get more reliable results than others. For example, if you’re wearing nail polish, a pulse oximeter — a device that clips onto the fingertip to measure blood oxygen through the skin — may not work well, because the polish interferes with proper function of the light sensor. In that situation, there’s a simple solution: remove the polish.
But in other cases, the solution isn’t simple. Increasingly, we’re recognizing that certain medical devices are less accurate depending on a person’s skin color, a phenomenon called device bias.
- Pulse oximeters. Although generally considered highly accurate and commonly relied upon in healthcare settings, their accuracy tends to be lower in people of color. That’s because the device relies on shining light through the skin to detect the color of blood, which varies by oxygen level. The amount of pigment in the skin may alter the way light behaves as it travels to blood vessels, leading to inaccurate results. The FDA has released an alert about this and other limitations of pulse oximeter use.
- Bilirubin measurement in newborns. Bilirubin is a breakdown product of red blood cells. Newborns are screened for high levels because this can cause permanent brain damage. When detected, phototherapy (light treatments) can help the baby get rid of the excess bilirubin, preventing brain damage. The screening involves examining a newborn’s skin and eyes for jaundice (a yellowing due to elevated bilirubin) and a light meter test to detect high bilirubin levels. But the accuracy of this test is lower in Black newborns. This is particularly important because jaundice is more difficult to detect in infants with darker skin, and dangerously high bilirubin levels are more common in this population.
- Heart rate monitors in smartphones. According to at least one study, smartphone apps may also be less accurate in people of color. Again, this is because the more skin pigment present, the more trouble light sensors have detecting pulsations in blood flow that reflect heartbeats.
Why device bias matters
Sometimes an error in measurement has no immediate health consequences. A 5% to 10% error rate when measuring heart rate may be of little consequence. (In fact, one could ask why anyone needs a device to monitor heart rate when you could just count your pulse for 15 seconds and multiply by 4!)
But pulse oximeter readings are used to help decide whether a person needs to be hospitalized, who requires admission to the intensive care unit, and who requires additional testing. If the oxygen level is consistently overestimated in people of color, they may be more likely to be undertreated compared with others whose readings are more accurate. And that may worsen previously existing healthcare disparities.
These examples add to the growing list of bias imbedded within healthcare, and other instances where failing to include diverse individuals has serious consequences. When you use a health device, it’s reasonable to wonder if it’s been tested on people like you. It’s also reasonable to expect people who develop medical and consumer health devices to widen the demographics of test subjects, to make sure results are reliable for all users before putting them on the market.
Sometimes a change in technology, such as using a different type of light sensor, can make health-related devices work more accurately for a wider range of people.
Or there may be no easy fix, and user characteristics will need to be incorporated into proper interpretation of the results. For example, a device could offer the user a choice of skin tones to match skin color. Then based on extensive data from prior testing of people with different skin colors, the device could adjust results appropriately.
The push to monitor our bodies, our health, and our life experiences continues to gain momentum. So we need to test and validate health-related devices to be sure they work for diverse individuals before declaring them fit for the general public. Even then, device bias won’t disappear: bodies vary, and technology has its limits. The key is to know it exists, fix what can be fixed, and interpret the results accordingly.
Follow me on Twitter @RobShmerling