Pulse oximetry, a commonly used method to monitor oxygen levels in patients, has long been trusted as a key diagnostic tool. However, for patients with darker skin tones, this technology may not be accurate, which can lead to hidden hypoxemia, a condition in which pulse oximetry readings overestimate oxygen levels and mask dangerously low blood oxygen levels. Ian Wong, MD, assistant professor of medicine and biostatistics and bioinformatics and team, in collaboration with colleagues from Emory University have been awarded a 5 year $3.4 million R01 from NIH to develop methods and technology to improve the accuracy of pulse oximetry.
Hidden hypoxemia disproportionately affects patients of color and can cause delayed treatment, which can have devastating consequences, including organ dysfunction and death.
Wong and team plan to use machine learning to reduce the risks associated with hidden hypoxemia. “By leveraging existing data from electronic health records (EHRs), we hope to create a machine learning-based model that identifies patients at high risk for hidden hypoxemia,” Wong said. “This system would use abnormal lab values to flag at-risk individuals, prompting clinicians to take additional steps such as ordering arterial blood gas tests or increasing oxygen therapy.”
This work may also lead to improving the current pulse oximetry equipment by integrating skin tone measurements, which would be a more cost-effective and time efficient measure than replacing all of the devices. At Duke alone, replacing these devices would cost between $14-20 million due to their widespread use across various medical settings, including operating rooms, ICU monitors, and even home devices.
Ultimately, the long-term goal of this research is to mitigate hidden hypoxemia and address healthcare inequities for minority patients. The models developed could become the foundation for future clinical trials aimed at improving pulse oximetry accuracy and reducing algorithmic bias.