Why Pulse Oximeters Miss Low Oxygen in Darker Skin

A number on a pulse oximeter looks like a direct measurement. It is not. The device sends light through or reflects light from tissue, detects a pulsating optical signal, and uses a calibration model to estimate arterial oxygen saturation. The displayed SpO2 is a model output, while SaO2 from arterial blood analyzed by a co-oximeter is the reference used in accuracy studies.

That distinction matters near a treatment threshold. If a device displays 94 percent while the arterial value is below 88 percent, a patient can have clinically important low oxygen that the screen does not reveal. Researchers call this occult hypoxemia. Multiple clinical studies have found it more often among patients identified as Black, and prospective work links the problem more directly to darker skin pigmentation, particularly when peripheral blood flow is poor.

Pulse oximetry remains useful. The safer response is not to discard it, but to understand its uncertainty, improve device testing, and avoid letting one reassuring number overrule symptoms, trajectory, or a need for confirmatory assessment.

How the device creates an estimate#

Oxygenated and deoxygenated hemoglobin absorb red and infrared light differently. A typical fingertip device emits light at two wavelength bands and measures how much reaches the detector, and it isolates the changing signal created by each arterial pulse from more constant absorption by tissue, venous blood, bone, and skin.

The ratio of those pulsatile signals is mapped to saturation through a calibration curve developed from human testing. In controlled desaturation studies, healthy volunteers breathe adjusted gas mixtures while simultaneous arterial samples provide reference values.

The device is therefore not solving oxygen saturation from first principles. It applies an empirical relationship learned from selected participants, sensor sites, saturation ranges, and laboratory conditions. If those conditions or people do not represent intended clinical use, performance can differ.

Where pigmentation enters the signal#

Melanin absorbs light and can change the optical path through skin. An algorithm must distinguish that background absorption from the pulsatile arterial component. Device wavelength, emitter variation, detector response, signal processing, sensor geometry, and calibration population all affect how well it succeeds.

Darker skin does not make pulse oximetry impossible. Performance varies across devices, sensors, saturation ranges, and clinical conditions. Some laboratory studies of particular systems have reported little disparity under controlled conditions. Real-world studies across heterogeneous devices have often found more upward error and more occult hypoxemia in darker-skinned groups; that variation is a reason to test each system well, not a reason to deny the broader safety signal.

Race is an imperfect proxy for skin tone#

Many retrospective studies used recorded race or ethnicity because direct pigmentation measurements were unavailable. Race is a social classification and does not map neatly to melanin. Skin tones overlap within every racial group, and device bias acts through physical and technical pathways rather than a race label itself.

Those studies still matter because they identify inequity in actual care. Their limitation points toward better research: measure pigmentation at relevant sites with validated visual and instrumental methods, while also reporting race and ethnicity to examine social patterns. FDA's January 2025 draft proposes both the Monk Skin Tone scale and colorimetry measures for clinical performance studies. It explicitly notes that a skin-tone scale is not a substitute for racial and ethnic diversity.

What occult hypoxemia means#

Definitions vary, but influential studies compared an SpO2 in a seemingly acceptable band, often 92 to 96 percent, with an arterial saturation below 88 percent. The danger lies in discordance that changes recognition or action.

An overreading can delay supplemental oxygen, arterial testing, escalation of care, or eligibility for a treatment tied to an oxygen threshold. A 2022 cohort study in COVID-19 associated oximeter discrepancies with delayed or unrecognized treatment eligibility among some racial and ethnic groups. Not every one-point error changes care, and not every arterial sample is timed closely enough to the pulse reading for a perfect comparison. Clinical impact depends on direction, magnitude, decision threshold, duration, and patient condition.

Low perfusion compounds the problem#

A pulse oximeter needs a strong pulsatile signal. Cold hands, shock, vasoconstriction, peripheral vascular disease, edema, or probe pressure can reduce blood flow at the sensor, and the device then estimates from a weaker signal relative to background tissue absorption.

In a prospective controlled study, missed hypoxemia became more frequent as perfusion fell, with the greatest frequency in the darkest pigmentation group under low-perfusion conditions, which is clinically important because critically ill patients can have both impaired circulation and low oxygen. Warming the hand when safe, repositioning the sensor, checking signal quality, and using another appropriate site or method may improve information. Repeating a flawed setup without changing it does not create an accurate average.

Error tends to matter more at lower oxygen levels#

Consumer devices are commonly used in the high nineties, where many readings cluster. Clinical danger often occurs lower on the curve, where accuracy can worsen and a small upward bias crosses an action threshold.

The oxygen-hemoglobin dissociation curve also becomes steeper as saturation falls. A modest saturation change can correspond to a larger change in oxygen tension. SpO2 does not measure ventilation, carbon dioxide, blood pH, hemoglobin concentration, or tissue oxygen delivery. A person can have a satisfactory saturation and still be critically ill from ventilation failure, anemia, low cardiac output, or another cause. Oxygen saturation answers one physiologic question.

Accuracy specifications are often misunderstood#

Pulse oximeter studies commonly summarize the difference between SpO2 and reference SaO2 with accuracy root mean square, or Arms. FDA's 2025 draft proposes an overall Arms below 3 percent for the tested saturation range, along with separate criteria for comparable performance across pigmentation groups.

An Arms of 3 percent is not a promise that every reading lies within plus or minus 3 percentage points. It is a population summary combining systematic bias and random variation. Individual errors can be larger.

Look for bias plots across the saturation range, limits of agreement, subgroup results, failures, sensor site, motion and perfusion conditions, and how many participants there were and how varied they were. One overall statistic can hide clinically important pockets of error.

The 2025 FDA proposal changes the test population#

The January 2025 document was draft guidance at the July 2026 research date. It contains nonbinding recommendations and is not yet a final implementation standard.

For controlled desaturation testing, FDA proposes a diversely pigmented group of at least 150 participants, much larger than older minimums. The draft recommends distributing participants across Monk Skin Tone groups, measuring pigmentation visually and by colorimetry, and prespecifying criteria for nondisparate performance.

FDA also proposes clearer labeling and a public list of medical-purpose devices that have submitted data supporting comparable performance across skin tones. The proposal applies to pulse oximeters for medical purposes; many products marketed only for general wellness have not undergone the same premarket review.

Medical-purpose and wellness devices are not equivalent#

A fingertip product can look similar to a cleared medical device while being sold for sports, aviation, or general wellness. FDA's draft explains that certain wellness oximeters are not reviewed before sale under the medical-purpose framework.

Price, app design, and a professional-looking display do not establish clinical accuracy. Before you buy, check the intended use, the authorization status, the model-specific labeling, the tested population, the saturation range, and the stated limitations. And note that even a cleared device is still an estimator with known constraints; clearance does not promise a correct result in every person or every condition.

Other sources of inaccurate readings#

Motion can make venous or tissue changes resemble an arterial pulse. Nail polish and artificial nails can alter light transmission. Bright ambient light, sensor misalignment, skin thickness, edema, and poor fit can interfere.

Abnormal hemoglobin species create another limitation. Standard two-wavelength pulse oximetry cannot reliably distinguish carboxyhemoglobin or methemoglobin from the usual oxygenated and deoxygenated forms. Carbon monoxide poisoning can therefore produce a falsely reassuring saturation. FDA also lists skin temperature, poor circulation, and current tobacco use among factors that can affect readings. The influence is device- and context-specific, so a universal correction factor is not safe.

Why subtracting a fixed number does not work#

It may be tempting to lower every reading in a person with darker skin by two or three points. The observed error is not constant. It varies by device, sensor, true saturation, pigmentation, perfusion, motion, and other conditions.

A blanket correction can create false low readings for some people while still missing larger overreadings in others. It also transfers responsibility from device evaluation to the patient. Better practice is to hold the number as uncertain, look at the signal quality and the trend, and confirm with a direct test when the decision matters or the reading argues with the clinical picture.

Trends help, but only if conditions are comparable#

A fall from a person's stable baseline can be important even if each reading has error. Yet trends are interpretable only when the same device, site, position, perfusion, and technique are reasonably consistent.

Switching sensors, taking a reading immediately after cold outdoor air, or accepting a rapidly fluctuating display can create an artificial trend. Let the device settle, and write down the context, whenever you are tracking readings over time. A stable false-high bias can still conceal low oxygen, so a trend does not replace confirmation when symptoms or risk are concerning.

Safer bedside interpretation#

Clinicians can compare the oximeter pulse with the measured pulse, inspect the waveform or signal indicator, check probe fit, improve peripheral warmth and flow when appropriate, and repeat at another suitable site. They can ask whether skin tone, low perfusion, motion, or dyshemoglobinemia could alter the result.

Arterial blood gas with co-oximetry provides a direct reference when precision is necessary, although arterial sampling is invasive and itself requires correct collection and handling. Venous blood gas does not directly substitute for arterial oxygenation. Decisions should integrate respiratory effort, mental status, color, circulation, vital signs, disease trajectory, and response to therapy. A device reading supports assessment; it does not veto it.

What home users should do#

Follow your model's instructions, use a warm relaxed hand, take off nail products where you can, stay still, and wait for the signal to settle. Write down the number and how you felt at the time. Use the action plan you were given for your own condition rather than a threshold borrowed from someone else.

FDA advises people not to rely on a pulse oximeter alone. If you have worsening shortness of breath, chest pain, confusion, bluish or gray discoloration, an inability to stay awake, or other severe symptoms, get urgent evaluation even if the display looks reassuring. Do not change prescribed oxygen flow or treatment solely from an unverified consumer reading unless a clinician's plan specifically directs it.

Designing devices for equitable performance#

Inclusive calibration is necessary but not sufficient. Manufacturers need adequate representation across pigmentation and saturation levels, direct pigmentation measurement, standardized sensors, device-specific analysis, and transparent labeling.

Testing should include conditions that challenge clinical performance, such as low perfusion and motion, as well as pediatric populations when intended. Postmarket studies can identify problems absent in healthy volunteer laboratories.

Equity is a performance requirement. If error systematically crosses treatment thresholds more often in one population, an acceptable overall average can still conceal preventable harm.