Johns Hopkins study finds pulse-oximeter errors changed direction


Five years of intensive-care records showed readings shifting from too high to too low at higher oxygen levels, with worse inaccuracies in patients of color.
A five-year review of Johns Hopkins Hospital’s intensive-care records found that pulse oximeters—devices that use light to estimate blood oxygen—gradually shifted from overestimating to underestimating oxygen levels at higher levels. STAT reported on October 6 that the inaccuracies affected all patients at those higher levels but were worse in patients of color.
Earlier studies had shown that pulse oximeters can overestimate blood oxygen in people with darker skin. That can make patients appear healthier than they are. The Hopkins study, published in the Annals of the American Thoracic Society, adds a problem beyond consistently high readings: the direction of error changed over time.
The study was retrospective, meaning researchers examined existing records rather than collecting readings as patients were treated. They compared readings from the same make of oximeter with arterial blood gas tests, which test blood from an artery. STAT describes these more invasive tests as the gold standard for measuring blood oxygen. The hospital’s trend does not establish that other devices or hospitals experienced the same shift.
Pulse oximeters shine light through the skin at two wavelengths. Blood carrying oxygen and blood without oxygen absorb that light differently. The device compares the absorption to estimate the percentage of blood that is oxygenated.
A predictable error would be easier for clinicians to account for, STAT explains. If a device consistently reads too high or too low, clinicians know the direction of its error. When that direction changes, compensating for the reading becomes harder.
The two kinds of error carry different risks. An overestimate can lead clinicians to miss a patient’s need for care. An underestimate can lead to unnecessary care or a hospital stay that lasts longer than necessary. These are potential consequences of inaccurate readings, not outcomes established by the Hopkins finding.
The Hopkins researchers were investigating a question raised by EquiOx, a Food and Drug Administration-funded study published earlier in 2026. EquiOx was prospective: researchers actively collected pulse-oximeter readings and compared them with arterial blood gas tests. It set out to investigate overestimates in patients with darker skin but instead found underestimates.
A separate prospective study from Emory University found evidence of racial bias, but differences in skin tone did not fully explain the inaccurate readings. Its leader, intensive-care physician Sivasubramanium Bhavani, suggested that conditions such as anemia, cardiovascular disease and liver disease could also affect readings. He said the causes and how they interact are not fully understood.
Masimo, which made the oximeters used in the Hopkins study, told STAT it was unaware of sensor, software or manufacturing changes during the study period that would explain the results. Its spokesperson said readings can be affected by sensor type, placement, signal quality and perfusion—the flow of blood through tissue—and that the study’s data did not establish the cause of the shift.
The company also pointed to two studies it supported that, according to its response, found no differences in pulse-oximeter effectiveness based on skin tone.
Why the readings changed year by year remains unresolved. Ashraf Fawzy, a Hopkins pulmonologist and critical-care physician who co-authored the study, told STAT he did not know what caused the finding.

Her attorneys say she continues receiving medical care after surviving Tennessee’s lethal-injection attempt on 30 September.
6 min

The figures also show 25–34-year-olds becoming the age group most likely to vape, while adult cigarette smoking declined across the UK.
2 min

The programme’s first satellite launch is planned for late 2028 or early 2029, unless exceptional events disrupt the schedule.
2 min