Treatment decisions based on inaccurate readings may perpetuate racial disparities in care
Pulse oximeters—small, noninvasive devices that estimate oxygen levels in the blood—are ubiquitous in hospitals. Clinicians rely on the information that these devices provide to make health care decisions, such as administering oxygen treatment or moving patients to a specialized area, such as the intensive care unit (ICU). However, a team of researchers have reported that the performance of pulse oximeters among minority patient groups may lead to the reduced delivery of supplemental oxygen, which could contribute to race-based disparities in health care.
“Racial disparities and inequities in the U.S. health care system have been well documented, with a surge of such reports emerging from the COVID-19 pandemic,” said Qi Duan, Ph.D., program director in the Division of Health Informatics Technologies at NIBIB. “This study offers one potential explanation for suboptimal care among minority patients, and provides further evidence that medical devices designed for healthy, white individuals may not perform adequately across all patient populations.”
It has been previously demonstrated that pulse oximeters have reduced accuracy among Black patients, where the device gives a falsely high estimate of oxygen saturation levels. Pulse oximeters work by measuring changes in the body’s absorption of red and infrared light, and can be affected by skin color. The clinical significance of this observation has been largely ignored; however, the U.S. Food and Drug Administration (FDA) issued a safety communication earlier this year about factors that can affect the accuracy of pulse oximeters, including skin pigmentation. The current study evaluated how the performance of pulse oximeters affected treatment decisions among a cohort of diverse patients in the ICU – specifically, the administration of supplemental oxygen.
The study, reported in JAMA Internal Medicine, analyzed data from 3,069 ICU patients (207 were Black, 112 were Hispanic, 83 were Asian, and the remaining patients were white) from a single hospital in Boston. In order to compare readings from the pulse oximeter with true arterial blood oxygenation levels, the researchers averaged oxygenation levels that were measured in the patients’ blood. They found that for a given level of oxygen measured in the blood, the pulse oximeter readings were significantly higher in non-white patients, confirming previous findings that these devices have reduced accuracy among minority patient groups.
The researchers also found that, compared with white patients, non-white patients received less supplemental oxygen, and that race was a significant factor. To better understand this association, they added a new variable into their model: the discrepancy in blood oxygenation levels, as estimated by pulse oximetry or measured by a blood draw. They found that higher discrepancies between estimated and true blood oxygenation levels were associated with lower supplemental oxygen delivery, and that race was not a significant factor when including this additional variable. In other words, the data show that minority patients did receive less supplemental oxygen, but this treatment decision was dictated by the readings from their pulse oximeter, and not inherently because of their race.
“We need to understand the drivers of health care disparities, and it turns out that one potential driver is the technology that clinicians rely on to triage patients and guide care,” explained senior study author Leo Anthony Celi, M.D., M.S., M.P.H., who is affiliated with Massachusetts Institute of Technology (MIT), Beth Israel Deaconess Medical Center, and Harvard. “It is important to reflect on the design and development of medical products—not just medical devices, but also drugs and interventions,” he added. “We need to be mindful that medical products developed using non-inclusive populations can contribute to the existing disparities that we are observing in the medical setting.”
“Our study indicates that we are giving less oxygen to minority patients in critical care settings, and that finding alone is somewhat shocking,” said first study author Eric Gottlieb, M.D., M.S., a lecturer at MIT and physician at Mount Auburn Hospital in Cambridge, MA. (Gottlieb was a nephrology research fellow at Brigham and Women's Hospital in Boston when this research was conducted.) “The disparities in pulse oximetry accuracy affect very real, concrete medical decisions, and this is something that the field needs to pay attention to.”
Celi also warned that it is important to understand the bias in real-world data when building machine learning algorithms for health care decisions. “We need to understand what factors are shaping clinicians’ treatment decisions, and what biases are seeping into the crevices of electronic health records, before we can confidently build these algorithms,” he said. “As artificial intelligence becomes a mainstay of health care, we need to be cautious and have a full understanding of what the data represents.”
This study was funded by a grant from NIBIB (R01EB017205) and a grant from NIDDK (for National Institute of Diabetes and Digestive and Kidney Diseases; T32DK007527).
Study reference: Eric R. Gottlieb, Jennifer Ziegler, Katharine Morley, Barret Rush, Leo A. Celi. Assessment of Racial and Ethnic Differences in Oxygen Supplementation Among Patients in the Intensive Care Unit. JAMA Internal Medicine, 2022; published online July 11, 2022.