the oximeter lies about darker skin

Problem statement

Pulse oximeters — ubiquitous devices used in virtually every clinical setting to measure blood oxygen saturation (SpO2) — produce systematically inaccurate readings in patients with darker skin pigmentation. Melanin absorbs the light wavelengths used by these devices, causing them to overestimate oxygen levels. This overestimation conceals hypoxemia, delays treatment, and contributes to worse clinical outcomes. Despite decades of awareness of this failure mode, no validated hardware or algorithmic correction exists that eliminates the bias across the full range of human skin tones.

Why this matters

Pulse oximeters are used on virtually every hospitalized patient and in millions of home-use contexts. In a University of Michigan cohort (January–July 2020), among patients with pulse oximeter readings of 92–96%, arterial blood gas measurement showed true oxygen saturation below 88% — occult hypoxemia — in 11.7% of measurements in Black patients versus 3.6% in White patients (Sjoding et al., NEJM 2020). A Johns Hopkins study of COVID-19 patients found pulse oximetry overestimated arterial oxygen saturation by an average of 1.2% in Black patients relative to White patients; Black patients had a 29% lower hazard of having their oxygen-treatment eligibility recognized, and those eventually recognized waited a median 1.0 hour longer (Fawzy et al., JAMA Internal Medicine 2022). Purchases and home use of pulse oximeters expanded sharply during the first year of the pandemic, and the installed base remains uncorrected.

What’s been tried and why it hasn’t worked

All manufacturers use the same basic dual-wavelength (red 660 nm / infrared 940 nm) technology, and calibration curves were historically derived from predominantly light-skinned study populations. The fundamental physics of pulse oximetry — measuring light absorption through tissue — is confounded by melanin absorption, which is not accounted for in the Beer-Lambert law models used. The FDA issued draft guidance in January 2025 recommending that premarket clinical studies enroll 150 or more participants, with at least 25% falling within each of three skin-tone groups on the Monk Skin Tone scale, assessed both subjectively (MST) and objectively (individual typology angle) — but this only affects new device submissions; the installed base of millions of devices remains uncorrected. No manufacturer has yet demonstrated a multi-wavelength or algorithmic approach that eliminates bias across all skin tones to within FDA's accuracy criterion for premarket studies (average root mean square error, Arms, of 3.0% or less for transmittance sensors). A Department of Veterans Affairs systematic review pooling accuracy studies found pulse oximeters overestimated oxygen saturation in Black patients by a pooled mean bias of 1.54% (95% CI 0.99 to 2.10) versus 0.62% (95% CI -0.08 to 1.32) in White patients — a small average difference that becomes clinically decisive near treatment thresholds — with wide patient-to-patient variation. Moving to multi-wavelength systems would require entirely new hardware, not just software updates, and manufacturers lack economic incentive to redesign a commodity product with thin margins.

What would unlock progress

A multi-wavelength optical approach (beyond the current two-wavelength design) that can distinguish melanin absorption from hemoglobin absorption would address the root cause. Alternatively, an algorithmic correction validated across the full Monk Skin Tone scale using large, diverse clinical datasets could improve accuracy without hardware changes. A rapid, low-cost skin-tone sensing module that enables adaptive calibration could bridge existing and next-generation devices.

Entry points for student teams

A student team could build a benchtop optical test system using additional wavelengths (e.g., 3-5 wavelengths) and tissue-mimicking phantoms with varying melanin concentrations to characterize the relationship between melanin levels and SpO2 measurement error. Another entry point would be developing an algorithmic correction layer that takes existing dual-wavelength signals plus a skin-tone input and outputs a bias-corrected SpO2 estimate. Teams with biomedical engineering, optics, or signal processing backgrounds would be well-suited. A scoped semester project could focus on phantom validation rather than clinical testing.