Impact of ethnicity on the accuracy of measurements of oxygen saturations: A retrospective observational cohort study

BACKGROUND: Pulse oximeters are routinely used in community and hospital settings worldwide as a rapid, non-invasive, and readily available bedside tool to approximate blood oxygenation. Potential racial biases in peripheral oxygen saturation (SpO2) measurements may influence the accuracy of pulse oximetry readings and impact clinical decision making. We aimed to assess whether the accuracy of oxygen saturation measured by SpO2, relative to arterial blood gas (SaO2), varies by ethnicity. METHODS: In this large retrospective observational cohort study covering four NHS Hospitals serving a large urban population in Birmingham, United Kingdom, consecutive pairs of SpO2 and SaO2 measurements taken on the same patient within an interval of less than 20 min were identified from electronic patient records. Where multiple pairs of measurements were recorded in a spell, only the first was included in the analysis. The differences between SpO2 and SaO2 measurements were compared across groups of self-identified ethnicity. These differences were subsequently adjusted for age, sex, bilirubin, systolic blood pressure, carboxyhaemaglobin saturations and the time interval between SpO2 and SaO2 measurements. FINDINGS: Paired O2 saturation measurements from 16,818 inpatient spells between 1st January 2017 and 18th February 2021 were analysed. The cohort self-identified as being of White (81.2%), Asian (11.7%), Black (4.0%), or Other (3.2%) ethnicities. Across the cohort, SpO2 was statistically significantly higher than SaO2 (p < 0.0001), with medians of 98% (interquartile range [IQR]: 95–100%) vs. 97% (IQR: 96–99%), and a median difference of 0.5% points (pps; 95% confidence interval [CI]: 0.5–0.6). However, the size of this difference varied considerably with the magnitude of SaO2, with SpO2 overestimating by a median by 3.8pp (IQR: 0.4, 8.8) for SaO2 values <90% but underestimating by a median of 0.4pp (IQR: -2.0, 1.4) for an SaO2 of 95%. The differences between SpO2 and SaO2 were also found to vary by ethnicity, with this difference being 0.8pp (95% CI: 0.6–1.0, p < 0.0001) greater in those of Black vs. White ethnicity. These differences resulted in 8.7% vs. 6.1% of Black vs. White patients who were classified as normoxic on SpO2 actually being hypoxic on the gold standard SaO2 (odds ratio: 1.47, 95% CI: 1.09–1.98, p = 0.012). INTERPRETATION: Pulse oximetry may overestimate O2 saturation, and this is possibly more pronounced in patients of Black ethnicity. Prospective studies are urgently warranted to assess the impact of ethnicity on the accuracy of pulse oximetry, to ensure care is optimised for all. FUNDING: PIONEER, the Health Data Research UK (HDR-UK) Health Data Research Hub in acute care.