Five NHS pulse oximeters read high on darker skin, and the misses cluster there

During the pandemic, NHS England sent fingertip pulse oximeters to people isolating at home so they could watch their own oxygen levels and call for help if the number fell. A study published in The BMJ on 14 January has now tested five of the devices used in that scheme against arterial blood measurements in critically ill patients, and reports that all five read systematically higher in people with darker skin tones [s1].

The design matters, because most of the argument about pulse oximetry and skin colour has been conducted using race recorded in medical notes rather than skin measured on the patient. The EXAKT study did the latter. Skin tone was measured objectively with a handheld spectrophotometer and expressed as the individual typology angle, a continuous scale rather than a category [s1].

What was measured

Researchers enrolled 903 critically ill adults across 24 intensive care units in England between June 2022 and August 2024, all of them screened for or enrolled in a trial of different oxygen therapy approaches [s1]. Each pulse oximeter reading of peripheral oxygen saturation was paired with an arterial blood sample analysed by co-oximetry, the reference standard [s1]. That produced 11,018 paired measurements [s1].

The measurement bias itself is small. On average, readings were 0.6 to 1.5 percentage points higher for patients at an individual typology angle of −44° — the darker end of the range — than for those at 46° [s1]. All five devices also shared a pattern independent of skin tone: they overestimated at lower true saturations and underestimated at higher ones [s1].

Where a small bias becomes a large one

A bias of roughly one percentage point sounds trivial until it is applied at a decision threshold. The clinically relevant question is not the average error but how often a device says a patient is fine when they are not.

The study assessed that as occult hypoxaemia: a reading above 92% paired with a true arterial saturation of 92% or below. Using a reading threshold above 94%, the proportion of measurements that missed genuine hypoxaemia ranged from 7.6% to 62.2% in patients with darker skin tones, against 1.2% to 26.9% in those with lighter tones — a gap of 5.3 to 35.3 percentage points, and rate ratios between 2.3 and 7.1 [s1]. The wide ranges are across the five different devices, not uncertainty within one.

False positive rates moved the other way, decreasing with darker skin tones [s1]. That is the arithmetic consequence of a device reading high: it raises fewer false alarms and misses more real events. Which error a health system would rather make depends on what the reading is used for, and a home monitoring scheme built around a self-referral threshold is close to the worst case for a device that misses low saturations.

The authors' conclusion is stated carefully: the differences "could translate into potentially clinically important differences" in false negative and false positive rates [s1]. This is a measurement and diagnostic accuracy study, not an outcomes study. It does not show that anyone came to harm, and it was conducted in intensive care rather than in the homes where the scheme actually operated.

The study that found the opposite

Two days later, Cardiology in the Young published a retrospective analysis pointing in a different direction in a different population. Researchers examined 1,645 infants aged seven days or younger in neonatal and cardiac intensive care at a single centre in Atlanta between 2010 and 2022, comparing pulse oximetry with arterial saturation measured within ten minutes [s2].

Occult hypoxaemia occurred in 20.5% of infants overall [s2]. But compared with White infants, there was no difference in the likelihood of occult hypoxaemia for Black infants (adjusted prevalence ratio 1.13, 95% CI 0.92 to 1.4) or Hispanic infants (1.06, 0.77 to 1.45) [s2]. The mean bias was 8.7% and the accuracy root mean square 13.7%, with no significant difference between groups [s2].

The authors' own reading is that they did not find the systemic discrepancy described in older children and adults, while noting that pulse oximetry significantly overestimated arterial oxygen saturation in all groups [s2]. They also name the limitation that separates their study from EXAKT: they classified patients by recorded race and ethnicity rather than measuring skin pigmentation, and call for prospective studies that measure pigmentation objectively [s2].

What the two studies do and do not settle together

They are not in direct contradiction. One measured skin tone in adults; the other used administrative race categories in neonates. Race is a poor proxy for the optical property a pulse oximeter actually interacts with, and a study using it can fail to detect a real pigmentation effect. Newborn skin is also less pigmented than the same child's skin will later be, which gives a mechanistic reason why a pigmentation effect might genuinely be smaller in the first week of life.

What they agree on is the finding least discussed: in the Atlanta cohort, one infant in five had a saturation reading that looked acceptable while the arterial value did not [s2], and in EXAKT, every device tested overestimated at low saturations regardless of skin tone [s1]. A device that reads high when the patient is sickest is a problem for everybody; skin tone determines who absorbs more of it.

What to watch

Whether device regulators require pigmentation-stratified performance data rather than self-reported race, and whether either study prompts a re-examination of the reading thresholds used in home monitoring schemes. Neither paper makes a recommendation about individual devices, and neither is a basis for any reader to change how they use one.

Sources

1. [s1] The impact of skin tone on performance of pulse oximeters used by NHS England COVID Oximetry @home scheme: measurement and diagnostic accuracy study. BMJ, 14 January 2026. https://doi.org/10.1136/bmj-2025-085535
2. [s2] Pulse oximetry and arterial saturation bias in neonates: retrospective analysis by race & ethnicity. Cardiology in the Young, 16 January 2026. https://doi.org/10.1017/S1047951125110718