Evaluating the Accuracy of Pulse Oximetry in Children According to Race

Abstract

This cross-sectional study investigates the association between race and pulse oximeter accuracy in children in the US.

Several studies using large retrospective electronic health record (EHR) data sets have found that pulse oximetry may overestimate actual blood oxygen levels in Black patients with hypoxemia more often than in White patients.^1,2,3 If true, a commonly used medical device could systematically mislead clinicians to undertreat hypoxemia in Black patients, a flaw with substantial implications for health care equity and outcomes.⁴ However, prior EHR-based studies are limited by imprecise linkages between pulse oximetry values and blood gas laboratory results obtained up to 10 minutes apart.^1,2,3 There are also few studies of race and pulse oximetry in children.³

Here we aimed to address these limitations by comparing arterial blood oxygen saturation (SaO₂) and pulse oximetry (SpO₂) values in Black and White children using data from a pediatric cardiac catheterization laboratory where SpO₂ is captured at 1-minute intervals in the EHR and is able to be linked to arterial blood gases processed immediately via co-oximeter.

Methods

We conducted this retrospective, EHR-based study at a free-standing children’s hospital. The Children’s Hospital of Philadelphia institutional review board deemed the study exempt and waived informed consent because this was secondary research. We included patients aged 1 to 17 years with self-reported Black or African American or White race who underwent a cardiac catheterization between 2016 and 2021. We included only cases where Sao₂ was measured from a location reflective of systemic arterial saturation (descending aorta, femoral artery, systemic ventricle). This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

We included the first Sao₂ measurement per patient matched with an SpO₂ value within 1 minute of the Sao₂ measurement. We calculated diagnostic test characteristics to examine detection of occult hypoxemia by race (SpO₂ <92% as test for Sao₂ <88%; 4% difference is consistent with prior studies^1,2,3 and US Food and Drug Administration guidance⁵). We calculated standard metrics for assessing pulse oximetry accuracy: bias (SpO₂ − Sao₂) and accuracy root mean square (ARMS).^5,6 We used multivariable linear regression to examine the association between race and bias while adjusting for clinical covariates (eMethods in Supplement 1). Statistical analyses were performed using R, version 4.1.2 (R Core Team).

Results

The study cohort included 774 patients (201 Black or African American [26.0%]; 51% female; 573 White [74.0%]; 43% female). Black or African American patients had a mean (SD) age of 6.6 (5.3) years. White patients had a mean (SD) age of 7.3 (5.3) years. The most common diagnoses in both groups included hypoplastic left heart syndrome/single ventricle, tetralogy of Fallot, and patent ductus arteriosus. The median (IQR) SpO₂ value used in the analysis was 96% (90%-98%). Diagnostic test characteristics by race are shown in the Figure. Among patients with true hypoxemia (Sao₂ <88%), 12% (7 of 57) of Black or African American patients vs 4% (4 of 110) of White patients had peripheral oxygen saturation levels suggesting normoxemia (SpO₂ ≥92%) (false negatives). Conversely, among patients with peripheral oxygen saturation levels suggesting normoxemia (SpO₂ ≥92%), 5% (7 of 139) of Black or African American patients vs 1% (4 of 407) of White patients had true hypoxemia (Sao₂ <88%).

Figure. Scatterplots of Paired Pulse Oximetry (SpO₂) and Arterial Blood Oxygen Saturation (SaO₂) Observations for Black or African American Patients and White Patients.

Black solid lines represent the line of equality and the trend line for the given set of paired data points. Blue solid lines denote the following quadrants: true positive includes patients whose paired SaO₂ and SpO₂ measurements both indicate hypoxemia (SpO₂ <92% and SaO₂ <88%). True negative includes patients whose SaO₂ and SpO₂ both indicate normoxemia. False positive includes patients whose SaO₂ is normoxemic but whose SpO₂ may suggest hypoxemia. False negative includes patients whose Sao₂ is hypoxemic but whose SpO₂ wrongly suggests normoxemia, placing the patient at risk of undertreatment (withholding supplemental O₂).

The unadjusted bias for Black or African American patients was 2.58 (95% CI, 2.15-3.00) compared with 0.89 (95% CI, 0.64-1.15) for White patients (P < .001). Results adjusted for clinical covariates by categorized SpO₂ range can be found in the Table.

Table. Bias and Accuracy Root Mean Square (ARMS) by Categorized Pulse Oximetry (SpO₂) Range^a.

Variable	Overall			SpO2
				≥92%			84%-91%			<84%
	Black/African American	White	Between-group difference	Black/African American	White	Between-group difference	Black/African American	White	Between-group difference	Black/African American	White	Between-group difference
No.	201	573	NA	139	407	NA	44	117	NA	18	49	NA
Adjusted bias (95% CI)b	2.61 (2.19-3.04)	0.88 (0.63-1.13)	1.73 (1.24-2.22)	2.46 (2.37-2.54)	0.79 (0.74-0.85)	1.66 (1.56-1.76)	2.84 (2.65-3.04)	1.07 (0.96-1.19)	1.77 (1.54-2.00)	2.88 (2.53-3.23)	1.29 (1.08-1.50)	1.59 (1.18-1.99)
ARMSc	4.36	3.01	NA	3.60	2.55	NA	4.50	3.40	NA	7.93	5.00	NA

Abbreviations: NA, not applicable; SaO₂, arterial blood oxygen saturation.

^a A multivariable linear regression model was used to model the association between race and bias, adjusting for patient characteristics (age, sex, blood pressure, temperature, hemoglobin, and medication use). Note: SpO2 was not included as a covariate in the regression model for bias. Predicted bias from the multivariable model was obtained for SpO₂ categories.

^b Bias = SpO₂ − SaO₂.

Discussion

Results of this cross-sectional study using data from a highly controlled clinical environment with tightly coupled Sao₂ and SpO₂ values suggest that pulse oximetry overestimated arterial oxygen saturation in children of Black or African American race.³ The discrepancy has been attributed to light absorption properties of melanin.⁵ Race is an imperfect proxy for skin pigmentation with the inherent assumption that Black or African American patients had darker skin and more melanin than White patients. Future studies in children should prospectively evaluate the association between SpO₂ and SaO₂ with reliable, direct measurement of skin pigmentation.

Supplement 1.

eMethods.

Supplement 2.

Data Sharing Statement