When Carolyn Hendrickson checked the data from her intensive care patients, there was no doubt about it. A clear pattern emerged. The pulse oximeters, those small clips that doctors and nurses rely on to monitor blood oxygen, weren’t giving everyone the same quality of information. In some patients, especially those with darker skin, the devices were misreading oxygen levels.

The errors were significant enough that they could mean the difference between timely treatment and a missed emergency.

The study, presented at the American College of Cardiology’s Annual Scientific Session, examined 1,760 paired measurements of blood oxygen levels from 631 critically ill patients at Zuckerberg San Francisco General Hospital. Each patient’s oxygen levels were measured two ways: with a pulse oximeter and with a blood gas analysis — a much more reliable but also more invasive method.

Researchers measured each participant’s skin color using both a subjective scale (the Monk Skin Tone Scale). They also made objective melanin measurements with a non-invasive device called a spectrophotometer. According to the objective measurements, 53% of patients were classified as having medium pigmentation, 33% were classified as having light pigment and 14% were classified as darkly pigmented.

The findings revealed a consistent problem. Pulse oximeters underestimated blood oxygen levels overall. But, for patients with darker skin, the margin of error often flipped in the opposite, more dangerous direction — overestimating oxygen levels and potentially delaying care.

“Although pulse oximeter bias on average was negative for all people, it was less negative in the darkly pigmented people than in the people with lighter pigment, meaning that pulse oximeters do not perform the same across different skin pigment categories,” said Carolyn Hendrickson, MD, associate professor of medicine at the University of California San Francisco and the study’s first author. “We also found that the proportion of positive bias — the one that goes in the worrisome direction meaning that someone might have dangerously low oxygen saturation that is not detected with a non-invasive monitor — was higher in patients with dark skin pigment compared to those with medium and light skin pigment.”

Why is this happening?

The problem lies in the very way pulse oximeters work.

These devices shine light through the skin and measure how much is absorbed. Melanin — the pigment that gives skin its color — also absorbs light. And higher melanin levels, found in darker skin, can interfere with the device’s readings.

This problem has been discussed in the past.

In 2022, a study from Harvard Medical School and Brigham and Women’s Hospital found that Black, Hispanic, and Asian patients in intensive care units were more likely to have falsely elevated oxygen readings than White patients. As a result, they received less supplemental oxygen, revealing a direct line between device inaccuracy and unequal treatment.

“It’s important to keep in mind that pulse oximeters give us an estimate, but it’s more than just a number. We use that estimate to make clinical decisions, such as how much supplemental oxygen to give a patient,” said corresponding author Eric Gottlieb, HMS clinical fellow in medicine at Brigham and Women’s, at the time. “It has real meaning for the patients that we care for because we can track back racial disparities in treatment to these differences in measurements.”

This is all the more important in the age of AI. When we’re training algorithms to respond to real, clinical data, it’s crucial to understand such biases.

It’s a problem in wearables too

Pulse oximeters aren’t the only devices where light sensors may not perform equally across populations.

Consumer wearables like the Apple Watch and Fitbit also use light-based sensors to track heart rate and, increasingly, blood oxygen. But a recent study led by Jessica Ramella-Roman of Florida International University showed that these devices struggle to deliver consistent readings for people with darker skin or those with obesity.

The problem was a similar one. Skin thickness, melanin, and blood flow can all distort the light signals wearables rely on. That means people at higher risk for cardiovascular disease could be the ones least likely to get accurate data. These are some of the people who benefit most from monitoring.

Oximeters as a Limiter

The study comes with limitations. For instance, all pulse oximeters in this study came from one manufacturer. Also, most participants had good oxygenation; readings could be different in low-oxygenation patients. But despite this, this is all the more concerning as people are increasingly using such devices as home.

Medical-grade pulse oximeters, like the one studied in Hendrickson’s research, are held to regulatory standards. But consumer devices are not. Even among hospital-grade devices, most validation studies have been done on healthy volunteers — not critically ill patients — and rarely include diverse skin tones.

The team suggests that future oximeters could include visual indicators — like warning lights or uncertainty alerts — when readings might be less reliable.

Meanwhile, researchers are calling for more inclusive testing that considers pigmentation, obesity, and other factors that affect sensor performance.