SHM Converge Scientific Abstract ePoster Gallery-Detecting Racial Bias in Pulse Oximetry and Modelling Its Impact on Clinical Outcomes in Hospitalized Covid-19 Patients

Pdf Summary

This poster studies racial bias in pulse oximetry among hospitalized COVID-19 patients and how that bias affects outcome prediction. Pulse oximeters can overestimate oxygen saturation (SpO2), especially in darker-skinned patients, but SpO2 distributions alone may be misleading because clinicians actively adjust supplemental oxygen to keep patients within a target range.<br /><br />To better assess bias, the authors introduced the estimated PaO2/FiO2 ratio (ePFR), which incorporates oxygen settings and may better reflect true hypoxemia severity. They analyzed 1,100 COVID-19 hospital encounters from March 2020 to February 2021 and compared Black patients with other patients using empirical cumulative distribution functions, Kolmogorov-Smirnov distance, and logistic regression for a primary outcome of death or ICU transfer within 24 hours.<br /><br />The key finding was that racial differences were much more visible with ePFR than with SpO2. The separation between Black and other patients’ SpO2 distributions was small, suggesting clinicians were maintaining similar measured SpO2 levels across groups. But the ePFR distributions showed a much wider gap, implying Black patients were receiving lower FiO2 for the same apparent oxygen saturation and may have been more severely hypoxemic than SpO2 suggested.<br /><br />Outcome prediction also differed by marker. Models based on SpO2 showed little racial disparity and largely failed to detect outcome differences. In contrast, ePFR-based models revealed stronger racial effects and better captured the relationship between hypoxemia and deterioration. For the same ePFR values, Black patients had worse outcomes than others.<br /><br />Overall, the study concludes that ePFR is a more effective tool than SpO2 for detecting racially biased pulse oximetry and its clinical consequences. The authors argue that ePFR could be used in electronic medical records and predictive models to support more equitable care and to help address structural inequities in pulse oximeter performance.

Asset Subtitle

Shrirang M. Gadrey

Meta Tag

Author List Beck Jacobsen, Brendan McNamara, Christopher Moore, Jessica J. Dreicer, Katherine Webb, Kira Dubester, Matthew Clark, Randall Moorman, Rebecca Kowalski, Robert Andris, Sarah Ratcliffe, Sean Haughey, Shrirang M. Gadrey, Swet Patel

Category Research

Concept Pulse Oximetry

Concept Estimated P/F Ratio

Concept Oxygen Saturation

Concept Racial Bias

Concept Hypoxemia

Distinguished Non-Finalist

Presenter Organization University of Virginia

Presenting Author Shrirang M. Gadrey

Track Outcomes Research

Keywords

racial bias

pulse oximetry

COVID-19

SpO2

ePFR

hypoxemia

Black patients

oxygen saturation

outcome prediction

ICU transfer

Pulse Oximetry

Estimated P/F Ratio

Oxygen Saturation

Racial Bias

Hypoxemia