Analysis of Discrepancies Between Pulse Oximetry and Arterial Oxygen Saturation Measurements by Race and Ethnicity and Association With Organ Dysfunction and Mortality

Key Points Question Do pulse oximetry discrepancies, hidden hypoxemia, and clinical outcomes differ among racial and ethnic subgroups? Findings In this cross-sectional study of 5 databases with 87 971 patients, significant disparities in pulse oximetry accuracy across racial and ethnic subgroups (ie, Asian, Black, Hispanic, and White individuals) were found, with higher rates of hidden hypoxemia associated with mortality, future organ dysfunction, and abnormal laboratory test results. Meaning In this study, discrepancies in pulse oximetry accuracy among racial and ethnic subgroups were associated with higher rates of hidden hypoxemia, mortality, and organ dysfunction.


SpO2 variability over five minutes
To further characterize SpO2 variability before an ABG, all SpO2 assessments were evaluated in the five minutes before the ABG. SpO2 variability was assessed by the difference between the maximum and the minimum values recorded during this time period.

Co-oximetry
98.2% of all ABGs in Emory Healthcare had co-oximetry, identifying MetHb and COHb.

rSOFA and hidden hypoxemia
Potential association between hidden hypoxemia and rSOFA, was closely examined. First, the rate of hidden hypoxemia was stratified by rSOFA score to determine whether risk for hidden hypoxemia increased with higher rSOFA. Second, in-hospital mortality was stratified by both rSOFA score and hidden hypoxemia to determine whether hidden hypoxemia (SpO2≥88% despite SaO2<88%) independently increased in-hospital mortality after adjustment from rSOFA.

Multivariate modeling
Multivariate models were examined both without and with laboratory values.

eFigure 6. Directed Acyclic Graph for Race and Ethnicity, Hidden Hypoxemia, Organ Dysfunction, and Mortality
This directed acyclic graph (DAG) is a proposed graph suggesting causality and confounders for race-ethnicity, hypoxemia, organ dysfunction, and mortality.
Race-ethnicity, through skin color, appears to influence hidden hypoxemia. Race-ethnicity may be directly associated with organ dysfunction, but this may also be influenced by the presence of hidden hypoxemia. In addition to hypoxemia and organ dysfunction, race-ethnicity is a factor that could influence mortality.
More work must be done to elucidate the causal mechanisms.

Organ dysfunction
In-hospital mortality eTable 1. Literature Review The table presents an extensive review of the literature on pulse oximetry, skin pigmentation, and race-ethnicity.

ICU STUDIES IN CRITICALLY ILL ADULTS -SKIN COLOR IS INCLUDED AS A VARIABLE
Skin pigmentation was objectively quantified.
Number of SaO2-SpO2 pairs that were analyzed was not listed.
A prospective observational study, convenience sample. Patients were recruited based on the appearance of darkly pigmented skin. Skin pigmentation was verified with a portable reflectance spectrophotometer.
Hypoxemia Note Per authors, the pulse oximeter readings at SaO2<92% tended to be lower than co-oximeter ABG values. A few measurements fell below 2 SD. Insufficient number of values at low SaO2 range to make a conclusive statement as only five SaO2 readings were <92% -the total number of analyses was not provided.
Study participants were selected for darkly pigmented skin.
Skin pigment color addressed. Adler et al. A prospective observational study, convenience sample, over four months.
Overall, pulse oximetry (SpO2) overestimated SaO2 (ABG) but no significant difference was reported for skin color.
Participants' skin pigment was classified into three groups using the Munsell Color tile system.
"Bias values were not significantly different across the 3 groups (p=0.79). Although precision was of borderline significance (p=0.051) no dose-response relationship was seen." Pulse oximetry signal failure was <1%.
It was not reported how many participants were in each of the three groups, although percentages were reported as: Skin pigment color addressed.
Ebmeier et al. -Participants were breathing a mixture of air-nitrogen-carbon dioxide via a mouthpiece and partial re-breathing circuitobtained end-tidal gases also.
-"A computer used end-tidal oxygen and carbon dioxide concentrations determined by mass spectrometry to estimate breath-by-breath SaO2, from which an operator adjusted inspired gas to rapidly achieve 2to 3-min stable plateaus of desaturation" Descriptive: Data on bias was presented for a range of SaO2 values (60-70%; 70-80%; 80-90% and 90-100%) compared by skin pigment light or dark (See Table 1 in paper) for 1,067 data points. Skin pigment color addressed.

Definitions
Descriptive: SaO2 and SpO2 percentages: range, mean, standard deviation (SD) as reported by each study.

Bias:
The mean differences between two values. Bias is a measure of trueness. SaO2 is the gold-standard. Some studies used SaO2 minus SpO2. However, other studies used SpO2 minus SaO2 to report bias. Some studies did not state how bias was calculated and it was not clear from the paper whether SaO2 -SpO2 or the reverse (i.e., SpO2 -SaO2) was used; in the table, we qualify this uncertainty as follows: "bias measurement method unclear". Precision: The standard deviation (SD) of the bias.

. Characterization of Variable Missingness Across the 5 EHR Data Sets
Missingness overall and per dataset. A value of 0.0% indicates that no data was missing. Note that SOFA scores in eICU-CRD, MIMIC-III, and MIMIC-IV datasets were calculated by carrying forward the value of its six components. Thus, there are situations where there may not be a cvSOFA value mapping to the hour preceding the ABG, in which case a SOFA score is calculated with the carried-forward cvSOFA value (instead of a newly updated cvSOFA value).

eTable 5. SpO2 Variability for All SpO2 Values Within the 5 Minutes Preceding the ABG Measurement
The table presents the variability in SpO2 measurements contained within a five-minute interval preceding the ABG, as measured by the difference between the two most extreme pulse oximetry values. Results are aggregated across the five databases and stratified by race-ethnicity.

Racial-ethnic subgroup
SpO2 variability (mean difference between minimum and maximum value ± standard error)

eTable 6. Percentage of Encounters With Arterial Blood Gases, by Race and Ethnicity
This table describes the percentage of encounters in all datasets (eICU-CRD, MIMIC-III, MIMIC-IV, Emory, and Grady) that have at least 1 ABG by race-ethnicity compared to all encounters examined. Chi square test of this distribution is p < 0.001.
Because of the composition of the Emory and Grady datasets, this table includes both ICU and non-ICU patients.