When a Pulse Oximeter Reads 92%, Why Can Skin Pigmentation Affect Whether the Number Is Trustworthy?

When Numbers Start Making Decisions · Season Three, “The People the Average Does Not See” · Article 4

1. The same display can carry different uncertainty

A pulse oximeter clips onto a finger and produces a number within seconds. If it displays an oxygen saturation of 92 per cent, clinicians may increase observation, provide oxygen, investigate deterioration or confirm the result with an arterial blood gas. Patients at home may treat the same number as a signal to seek urgent help.

The number looks direct. It is actually an estimate of arterial oxygen saturation produced by light passing through tissue. The Therapeutic Goods Administration has warned that pulse oximeters have important limitations and that accuracy can be affected by skin pigmentation. Evidence indicates that readings in people with darker skin can systematically overestimate oxygen saturation, particularly in the clinically important range. The TGA notes that a reading around 88 to 94 per cent can differ from the true value by several percentage points and should be interpreted in context. TGA: Limitations of pulse oximeters and the effect of skin pigmentation

The display may therefore be technically produced exactly as designed while giving unequal reassurance. The problem is not simply that every device has error. It is that the direction and distribution of error may differ between groups, precisely where a threshold can influence care.

2. SpO2 is an estimate, not the oxygen in the blood itself

Pulse oximeters emit red and infrared light and estimate saturation from differences in absorption between oxygenated and deoxygenated haemoglobin. The device reports peripheral oxygen saturation, SpO2. It is not directly sampling arterial blood.

Motion, poor circulation, cold hands, nail products, probe placement, ambient light, device quality and abnormal haemoglobins can all affect readings. The displayed whole number also hides a range of uncertainty. A stable “92” does not mean that the underlying arterial saturation is known to be exactly 92.0 per cent.

Skin melanin also absorbs light. Device calibration and algorithms are expected to separate the physiological signal from other optical effects, but performance depends on the people and conditions represented in development and validation. If darker pigmentation is inadequately represented, average accuracy can conceal systematic error.

This is a measurement problem with an ethical consequence. A device need not identify race or intend discrimination to produce unequal risk. If overestimation delays recognition of hypoxaemia more often in one population, a neutral-looking instrument has different practical authority across bodies.

3. Why 92 per cent matters without being a universal verdict

Oxygen thresholds depend on clinical context. A reading that is concerning for one patient may be expected for another with chronic respiratory disease and an individually prescribed target range. Symptoms, respiratory rate, work of breathing, circulation, trend and diagnosis all matter.

Australian clinical guidance for chronic obstructive pulmonary disease emphasises assessment and appropriate oxygen targets rather than blind correction to one universal number. Australian Commission on Safety and Quality in Health Care: COPD Clinical Care Standard Resuscitation guidance likewise uses targeted oxygen therapy and clinical assessment. ANZCOR Guideline 11.6.1

Yet thresholds remain useful. They make deterioration visible and support consistent escalation. The mistake is to treat the threshold as if it erased measurement uncertainty and patient difference. A number near a decision boundary should prompt more attention to the conditions under which it was produced, not less.

When the known bias tends towards overestimation, a darker-skinned patient with a displayed 92 may have a lower true saturation than the display suggests. The safe response is not to apply an invented universal correction. It is to reduce confidence in a lone reading, assess the patient and confirm when the clinical consequence warrants it.

4. Average device accuracy can hide unequal failure

A manufacturer can report acceptable overall performance while a subgroup experiences more clinically significant error. If a validation population is weighted towards lighter skin, a summary error measure can be dominated by that group. Even when subgroup participants are included, broad categories and subjective descriptions may fail to characterise pigmentation adequately.

The relevant audit is distributional. How often does the device overestimate by one, two, three or more percentage points at low saturation? Does the error differ by pigmentation, perfusion, age or setting? What proportion of patients would cross a clinical threshold if arterial saturation were used? A mean error alone cannot answer those questions.

Direction matters as much as size. Random error creates noise in both directions. Systematic overestimation creates false reassurance. In a triage context, false reassurance can delay treatment or discharge someone whose condition is worse than it appears.

Manufacturers and regulators should therefore examine subgroup performance in the ranges where decisions change. Labels and clinical education should communicate residual limits. Procurement should not treat all approved devices as interchangeable where independent evidence shows meaningful performance differences.

5. The patient must remain visible beside the monitor

Good clinical practice already combines numbers with observation. Cyanosis, breathlessness, confusion, chest pain, respiratory effort and deteriorating vital signs can contradict a reassuring display. Repeated readings and trends can be more informative than one isolated value. Another probe site or device may identify a technical problem, while arterial blood gas analysis can provide a more direct measurement when precision is necessary.

The Australian Commission’s safety material on documenting oxygen therapy stresses recording the device, flow, target and response so that treatment is interpretable rather than reduced to a saturation value. Documenting oxygen therapy and response safely

This is not permission to dismiss devices whenever intuition disagrees. Clinical judgement has its own biases, including unequal assessment of pain and severity. The answer is evidence triangulation: device reading, waveform or signal quality, symptoms, examination, history, trend and confirmatory tests.

When those sources conflict, the conflict is data. It should trigger investigation rather than an automatic victory for the screen.

6. Home use changes the responsibility to explain

Consumer pulse oximeters became familiar during respiratory outbreaks, but home users may not know how to assess signal quality or interpret variation. A person can become falsely reassured by one reading or frightened by harmless fluctuation.

Instructions should state that the number is an estimate, explain common sources of error, advise warming and resting the hand, and emphasise symptoms and trends. People should know the action plan supplied by their clinician and when severe symptoms require emergency care regardless of the display.

Warnings about pigmentation must be usable. Saying only that “skin colour may affect accuracy” transfers uncertainty to the patient without guidance. The practical message is that an unexpectedly reassuring result should not override concerning symptoms and that confirmation may be necessary in the critical range.

Healthcare systems should also avoid making patients responsible for correcting a device-design problem. Better validation standards, procurement, training and access to confirmatory testing are institutional duties.

7. A practical framework for unequal measurement error

When a device result influences care, ask:

1. What biological quantity is measured directly, and what is estimated?
2. What uncertainty is hidden by the displayed precision?
3. Were relevant skin pigmentations and clinical conditions represented in validation?
4. Does error differ in size or direction between groups?
5. Is performance tested in the range where treatment changes?
6. Which symptoms, trends and measurements can corroborate or contradict the result?
7. When is confirmatory testing required?
8. Are adverse events and subgroup errors fed back into device standards and practice?

These questions move beyond asking whether the instrument works “on average”. A medical device is fit for purpose only if its uncertainty is acceptable for the people and decisions in which it is used.

Conclusion: a precise display is not equally precise evidence

Pulse oximetry is fast, non-invasive and clinically indispensable. Its limitations are not reasons to abandon it. They are reasons to calibrate its authority.

My judgement is that a reading such as 92 per cent should be treated as one piece of evidence whose uncertainty depends on the device, signal, clinical setting and patient. Known systematic differences associated with skin pigmentation require distributional validation, explicit guidance and a lower threshold for corroboration when symptoms and display do not agree.

The average patient used to validate an instrument is not present at the bedside. A particular person is. Measurement becomes fair only when the system can see the difference.