A retrospective review of five years of intensive care data at Johns Hopkins found that the same make of pulse oximeter changed its bias over time: early in the period the devices tended to overestimate arterial oxygen saturation, and later they shifted to underestimating saturation at higher oxygen levels. The pattern affected all patients but was worse in patients of color. Investigators compared oximeter readings with arterial blood gas (ABG) measurements, the clinical gold standard, and were unable to identify a clear cause for the year-to-year change.
Pulse oximeters estimate blood oxygenation by transmitting light at two wavelengths through the skin. One wavelength is preferentially absorbed by oxygenated hemoglobin and the other by deoxygenated hemoglobin; the device compares the two absorptions to calculate a percent oxygen saturation. This noninvasive method became ubiquitous during the Covid-19 pandemic but has well-documented limitations, including previously reported overestimation of oxygen saturation in patients with darker skin tones.
The Hopkins study used ABG measurements as the reference standard to evaluate device accuracy. By directly comparing oximeter readings to ABG values over multiple years, the researchers documented a temporal trend in which the devices’ direction of error reversed from overestimation to underestimation at higher oxygen saturations. The reporting emphasizes that these findings came from retrospective ICU data and that the observed shift was not attributable within the study to a single clear cause.
Masimo, the manufacturer of the oximeters used in the Hopkins analysis, stated it was unaware of any sensor, software, or manufacturing changes during the study period that would explain the reported results. The company also noted that a number of factors can influence oximeter readings, including the sensor model and placement, signal quality, and patient perfusion. Masimo pointed to additional company-supported studies that did not find differences in efficacy by skin tone. The Hopkins authors and the reporting, however, emphasize that the available data do not establish the cause of the observed shift from overestimation to underestimation.
The Hopkins work was in part an effort to address questions raised by the FDA-funded EquiOx prospective study, which aimed to investigate why pulse oximeters overestimate oxygenation in people with darker skin. Unexpectedly, EquiOx found underestimation instead of the anticipated overestimation, a result whose explanation remained unclear. Earlier, a landmark 2020 New England Journal of Medicine retrospective analysis highlighted racial bias in oximeter readings, catalyzing follow-up investigations.
Another prospective study from Emory found evidence of racial disparities in oximeter accuracy but concluded that differences were not fully explained by skin tone alone. Investigators there and at Hopkins pointed to other clinical factors that may influence readings — for example, conditions such as anemia, cardiovascular disease, or liver disease — and warned that these comorbidities, which are unequally distributed across populations, may contribute to observed disparities.
Unpredictable or shifting bias in pulse oximeter readings complicates clinical decision making. When oximeters overestimate arterial oxygenation, clinicians risk failing to provide necessary respiratory support. When devices underestimate oxygenation, patients may receive unnecessary interventions or remain hospitalized longer than required. Several authors and commentators quoted in the reporting stressed that if device bias were consistent and predictable clinicians could adjust; but the observed inconsistency — a device that may run "fast" or "slow" depending on timing and context — undermines that approach.
The reporting makes clear that neither the Hopkins retrospective data nor manufacturer statements established a definitive cause for the observed shift from overestimation to underestimation. Prospective studies to date have produced mixed and sometimes surprising results, suggesting the problem is multifactorial and not solely explained by skin pigmentation. Investigators called for further research to identify device- and patient-level factors that alter oximeter performance over time and across populations, and emphasized the need for transparency about device behavior and calibration. Until causes are clarified, clinicians and health systems face ongoing uncertainty about how reliably pulse oximeter readings reflect true arterial oxygenation across diverse patients.