Pulse Oximeters and the Mathematics of Invisible Hypoxia
How a Measurement Standard Hid Harm in Plain Sight
Maths Casualties - Case 004
Case Summary
§Pulse oximeters are ubiquitous medical devices used to estimate blood oxygen saturation. They are small, inexpensive, and trusted; especially in emergency rooms, ICUs, and during COVID-19.
§For decades, however, these devices have been systematically less accurate for patients with darker skin pigmentation.
§The result is a phenomenon known as occult hypoxemia: dangerously low blood oxygen levels that are not detected by the device, delaying care and treatment.
§This was not a recent discovery. It was known, documented, and tolerated.
§The harm was not caused by a single faulty device, but by measurement standards that defined acceptable error in ways that systematically excluded certain populations.
The Measurement System (What Pulse Oximeters Actually Do)
Pulse oximeters do not measure oxygen directly.
They estimate oxygen saturation by:
emitting light at specific wavelengths,
measuring absorption through tissue,
and applying an empirical calibration curve.
Mathematically, the device estimates:
where is a ratio of absorbed light at two wavelengths.
This function is not derived from first principles.
It is fit using experimental calibration data.
Where the Mathematics Goes Wrong
1. Calibration as an Empirical Shortcut
§Calibration curves are built by:
§testing devices on healthy volunteers,
§inducing controlled hypoxia,
§correlating optical readings with arterial blood gas measurements.
§Historically:
§calibration cohorts were overwhelmingly light-skinned,
§skin pigmentation was treated as noise,
§and error tolerances were averaged across populations.
§This is not an oversight. It is a design decision.
2. Optical Absorption Is Not Skin-Neutral
§Melanin absorbs light in the same wavelengths used by pulse oximeters.
§This means:
§darker skin alters signal intensity,
§changes the light ratio RRR,
§and shifts the inferred oxygen saturation upward.
§Mathematically, the model assumes:
§R→SO2R \rightarrow S_{\text{O}_2}R→SO2
§is invariant across skin pigmentation.
§Empirically, it is not.
3. Error Bounds That Hide Asymmetry
§Regulatory standards historically allowed:
§±2–3% average error across test populations.
§But average error can mask subgroup failure.
§A model can meet regulatory accuracy thresholds while:
§systematically overestimating oxygen saturation in darker-skinned patients,
§especially in the clinically dangerous low-oxygen range.
§This is a textbook case of aggregation hiding harm.
How Harm Manifested Clinically
§Studies have shown that Black patients:
§are more likely to experience occult hypoxemia,
§are less likely to receive supplemental oxygen or escalation of care,
§and may experience worse outcomes as a result.
§During COVID-19, reliance on pulse oximeters:
§This was not a rare edge case.
It was a population-scale measurement failure.
Institutional Responsibility and Delay
§The limitations of pulse oximeters were:
§published repeatedly over decades,
§acknowledged in technical literature,
§known to device manufacturers.
§Yet:
§devices remained clinically central,
§labeling minimized the issue,
§and regulatory standards lagged behind evidence.
§Correction came slowly, and largely after public attention intensified.
Evidence Base
Longstanding evidence
§Multiple studies (1990s–2010s) documenting skin-tone-related inaccuracies
Recent confirmation
Regulatory response
§FDA safety communications
§NIH initiatives to revise standards
§Calls for improved calibration diversity
§This is not speculative and not disputed.