If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Division of Neurosciences Critical Care, Department of Neurology, Neurosurgery, Anesthesiology and Critical Care Medicine, Johns Hopkins Hospital, Baltimore, Md
Division of Neurosciences Critical Care, Department of Neurology, Neurosurgery, Anesthesiology and Critical Care Medicine, Johns Hopkins Hospital, Baltimore, Md
Address for reprints: Sung-Min Cho, DO, MHS, Division of Neurosciences Critical Care, The Johns Hopkins Hospital, 600 N. Wolfe St, Phipps 455, Baltimore, MD 21287.
Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MdDivision of Neurosciences Critical Care, Department of Neurology, Neurosurgery, Anesthesiology and Critical Care Medicine, Johns Hopkins Hospital, Baltimore, Md
To determine whether there is racial/ethnical discrepancy between pulse oximetry (SpO2) and oxygen saturation (SaO2) in patients receiving extracorporeal membrane oxygenation (ECMO).
Methods
This was a retrospective observational study at a tertiary academic ECMO center with adults (>18 years) on venoarterial (VA) or venovenous (VV) ECMO. Datapoints were excluded if oxygen saturation ≤70% or SpO2–SaO2 pairs were not measured within 10 minutes. The primary outcome was the presence of a SpO2–SaO2 discrepancy between different races/ethnicities. Bland–Altman analyses and linear mixed-effects modeling, adjusting for prespecified covariates, were used to assess the SpO2–SaO2 discrepancy between races/ethnicities. Occult hypoxemia was defined as SaO2 <88% with a time-matched SpO2 ≥92%.
Results
Of 139 patients receiving VA-ECMO and 57 patients receiving VV-ECMO, we examined 16,252 SpO2–SaO2 pairs. The SpO2–SaO2 discrepancy was greater in VV-ECMO (1.4%) versus VA-ECMO (0.15%). In VA-ECMO, SpO2 overestimated SaO2 in Asian (0.2%), Black (0.94%), and Hispanic (0.03%) patients and underestimated SaO2 in White (−0.06%) and nonspecified race (−0.80%) patients. The proportion of SpO2–SaO2 measurements considered occult hypoxemia was 70% from Black compared to 27% from White patients (P < .0001). In VV-ECMO, SpO2 overestimated SaO2 in Asian (1.0%), Black (2.9%), Hispanic (1.1%), and White (0.50%) patients and underestimated SaO2 in nonspecified race patients (−0.53%). In linear mixed-effects modeling, SpO2 overestimated SaO2 by 0.19% in Black patients (95% confidence interval, 0.045%-0.33%, P = .023). The proportion of SpO2–SaO2 measurements considered occult hypoxemia was 66% from Black compared with 16% from White patients (P < .0001).
Conclusions
SpO2 overestimates SaO2 in Asian, Black, and Hispanic versus White patients, and this discrepancy was greater in VV-ECMO versus VA-ECMO, suggesting the need for physiological studies.
Race/ethnicity biases pulse oximetry measurements in ECMO patients, leading to occult hypoxemia in Black ECMO patients. There may be further physiological explanations for this SpO2–SaO2 discrepancy.
Considering the severity of illness of ECMO patients and ECMO's increasing popularity, accurate and precise oxygen saturation measurements for ECMO patients are crucial, particularly at hypoxemic levels. Black ECMO patients seem to be at greatest risk for occult hypoxemia, and clinicians should note for this SpO2–SaO2 discrepancy when monitoring pulse oximetry and treating these patients.
See Commentary on page XXX.
Oxygen saturation measured by pulse oximetry (SpO2) is a noninvasive method to continuously monitor oxygenation in place of arterial gas oxygen saturation (SaO2). SpO2 has been known to inaccurately predict SaO2 readings, especially in the intensive care unit (ICU),
Racial bias and reproducibility in pulse oximetry among medical and surgical inpatients in general care in the Veterans Health Administration 2013-19: multicenter, retrospective cohort study.
Analysis of discrepancies between pulse oximetry and arterial oxygen saturation measurements by race and ethnicity and association with organ dysfunction and mortality.
As pulse oximetry works by spectrophotometry, such inaccuracy in SpO2 measurements in predicting SaO2 has been attributed to skin color among other physiological reasons such as dyshemoglobinemia interference, low perfusion, and sickle cell anemia.
Currently, sparse data exist on the SpO2 and SaO2 discrepancy between racial/ethnical groups in VA- and VV-ECMO populations. We hypothesized that this discrepancy would be heightened in patients receiving ECMO, as the result of their critical illness, and complex physiology such as differential hypoxia. We also hypothesized that different cannulation strategies in patients receiving VA- and VV-ECMO would affect this discrepancy in addition to race/ethnicity.
Methods
Study Design
This study was approved on October 22, 2019, by the Johns Hopkins Hospital Institutional Review Board with a waiver of informed consent, as this was a retrospective observational study (IRB00216321) entitled “Retrospective Analysis of Outcomes of Patients on Extracorporeal Membrane Oxygenation,” in accordance with the ethical standards of the responsible committee on human experimentation (institutional or regional) and with the Helsinki Declaration of 1975. A retrospective analysis of a database containing patients undergoing ECMO at a tertiary care center between June 2016 and April 2021 was performed. All patients were managed in the Cardiovascular Surgery Intensive Care Unit, Cardiac Critical Care Unit and obtained neurocritical care consultations based on our standardized neuromonitoring protocol.
The on-call ECMO attending physician rounded for all ECMO patients on both ICUs.
Participants
We included all adult patients (age ≥18 years) who received VA-ECMO and VV-ECMO. Patients without race/ethnicity or SpO2 and SaO2 information were excluded.
Data Collection
For all study patients, we collected SaO2 measured by arterial blood gas (ABG), SpO2 measured by pulse oximetry during ECMO support, and ECMO cannulation strategy extracted from electronic medical records. Precannulation characteristics included demographics, medical history, and on-ECMO physiological and laboratory values. Postcannulation characteristics such as discharge location, ECMO duration, mortality, neurological outcome, and the number of SpO2 measurements per individual were also acquired. ABGs were collected every 2 to 4 hours during ECMO support, and SpO2 was recorded every 15 minutes, according to the standard clinical protocol at Johns Hopkins Hospital with more recurrent collections if clinically indicated. All patients with VA-ECMO had a right radial arterial line for accurate and recurrent ABG measurements and as a sensitive marker of differential hypoxia. Baseline ABGs before ECMO cannulation and serial ABGs after ECMO cannulation were collected. SaO2 from ABG was calculated based on the partial pressure of oxygen. Vital signs were also collected at least every 15 minutes pre- and postcannulation. SpO2 and SaO2 measurements were recorded as a single reading at a particular time and date. For patients receiving VA-ECMO, the pulse oximeter probe was placed on the right finger or right earlobe. For patients receiving VV-ECMO, the pulse oximeter probe was placed on the right or left hand. All SpO2 and SaO2 measurements that were recorded outside of ECMO duration were excluded.
Definitions
SpO2 was defined as peripheral oxygen saturation measured by pulse oximetry, whereas SaO2 was defined as arterial oxygen saturation measured by ABG. SpO2 and SaO2 values of less than 70% were excluded from analysis, because these were determined to likely be from erroneous measurements. SpO2–SaO2 pairs were matched by time-only values that were measured ≤10 minutes apart and were used for the analysis to control for fluctuations over time. Our data contained one entry for each race/ethnicity: Asian, Black, Hispanic, Others, and White. “Others” denoted races/ethnicities that were not specified in the previously aforementioned entries (ie, “nonspecified races”). Occult hypoxemia was defined as SaO2 <88% with a time-matched SpO2 ≥92%.
Outcomes
The primary outcome was the presence of a SpO2–SaO2 discrepancy between patients of different races and ethnicities. White race/ethnicity was used as the reference comparator. Our secondary outcome was the presence of a SpO2–SaO2 discrepancy between different cannulation strategies in patients with VA- and VV-ECMO. In addition, we assessed the accuracy and precision of SpO2 in predicting SaO2 in patients with VA-ECMO and VV-ECMO support.
Statistical Analysis
Median data were presented (interquartile range [IQR]) for continuous variables and absolute numbers with percentages for binary/categorical variables. Wilcoxon rank-sum test was used for comparing data with continuous variables and Pearson χ2 test for binary/categorical variables. Differences between SpO2–SaO2 pairs across different individuals were compared using the Wilcoxon rank-sum and Kruskal–Wallis tests. Bland–Altman analyses were conducted through the following: mean difference (estimated bias) = the average of SpO2 and SaO2 and then subtracting SaO2 from SpO2, precision = the standard deviation of the mean difference, limits of agreement = mean difference +/− 1.96∗precision, and root mean square error = sqrt(((mean difference – precision)2)), as described in previous studies.
The relationship between race/ethnicity and the difference between SpO2 and SaO2 measurements were analyzed first using unadjusted linear mixed-effects modeling (LMM), with the individual patient as a random effect. This LMM was then adjusted for preselected covariates posited to be associated with pulse oximetry accuracy, including demographics and time-dependent clinical and laboratory variables.
Covariates were age, sex, vasopressor or inotrope requirement during ECMO, and cannulation strategy. Time-dependent clinical and laboratory variables included pH, temperature, lactate dehydrogenase (LDH), and hemoglobin. Time-independent and -dependent covariates were included as fixed and random effects, respectively, in the LMM. Patients receiving VA- and VV-ECMO were analyzed separately.
Three different thresholds of SaO2 were selected as 88%, 92%, and 95%, based on previous literature
Tidal volume reduction for prevention of ventilator-induced lung injury in acute respiratory distress syndrome. The Multicenter Trail Group on Tidal Volume reduction in ARDS.
to determine the sensitivity and specificity of SpO2 to predict SaO2 accurately. For each SaO2 threshold, we tested all SpO2 values acquired in the study and calculated sensitivities and specificities for the cut-off point of SpO2 to detect SaO2 at the threshold or below. We also determined an “optimal” SpO2 using the receiver-operating characteristic (ROC) curve and area under the ROC curve analyses, ultimately based on Youden's
index. Moreover, we determined the median sensitivity, specificity, positive predictive value, and negative predictive value for each SaO2 threshold and corresponding SpO2 cut-off value.
All statistical analyses were performed using R Studio (R 4.1.2, 2022). LMM was fitted using the lme4 package, and ROC analyses were conducted using the pROC package.
Results
Of 196 patients (139 patients receiving VA-ECMO; 57 patients receiving VV-ECMO), we collected 37,514 SaO2 and 164,212 SpO2 data points. A total of 16,252 SpO2–SaO2 pairs were used in our final analysis, as they were measured 10 minutes or less between each other and had 70% or greater oxygen saturation (Figure 1).
Figure 1Flow diagram of creation of study cohort with time-matched SpO2-SaO2 data points that are at least 70% saturation. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry; ECMO, extracorporeal membrane oxygenation; VA, venoarterial; VV, venovenous.
Our demographics and clinical characteristics information, stratified by race and ethnicity within each ECMO type (VA- and VV-ECMO), are presented in Tables 1 and 2. Of 139 VA-ECMO (median age, 60 years, 63% male) and 57 VV-ECMO (median age, 47 years, 56% male) patients, 5 underwent both VA-ECMO and VV-ECMO support and were accounted for in both analyses. Overall, patients receiving VV-ECMO were cannulated over 4 times longer (median, 348.98 hours; IQR, 151.42-605.58 hours) than patients receiving VA-ECMO (median, 95.88 hours; IQR, 58.32-191.62 hours). Black patients receiving VA-ECMO had the greatest number of SpO2 measurements recorded per patient (median, 61; IQR, 21.5-91.75), and a correspondingly longer ECMO duration time (median, 93.9 hours; IQR, 58.3-150.9 hours) compared with other races/ethnicities. In addition, Hispanic patients receiving VA-ECMO had a 100% mortality and a correspondingly high median BMI (median, 36.50 kg/m2; IQR, 36.45-36.55 kg/m2) and significantly shorter ECMO duration time (median, 27.52 hours; IQR, 24.98-40.77 hours). Asian VA-ECMO patients also had a 100% mortality rate and the greatest median age (70 years; IQR, 58-78 years). Table E1 summarizes VA- and VV-ECMO patient analyses by race and ethnicity. The overall estimated bias (mean difference) was greater for patients receiving VV-ECMO (1.4%) than patients receiving VA-ECMO (0.15%).
Table 1Baseline characteristics and clinical variables of patients receiving venoarterial extracorporeal oxygenation membrane (VA-ECMO)
Variables were collected within the first 12 h of ECMO initiation; creatinine, platelet, lactate, AST, ALT, LDH, and hemoglobin measurements, and SOFA score represent the worst value collected in the first 12 h of ECMO.
Creatinine, mg/dL
1 (1-2)
1 (1-2)
2 (1-2)
2 (1-2)
2 (1.5-6.5)
1 (1-2)
Platelet, units in thousands/μL
83.0 (52.0-116.2)
93.0 (77-120)
67 (53-138)
79 (71-99)
49 (48.5-55.5)
50 (35-57)
Lactate, mmol/L
6 (3-10)
5 (2-9)
5 (3-11)
6 (4-9)
10 (8.5-11)
4 (3-6)
AST, units/L
168 (71-757.5)
160 (60-697)
131 (65.5-625)
138 (126-185)
56 (48.5-366)
314 (95-851)
ALT, IU/L
74 (24-396.5)
45 (26-288.8)
47 (20.5-285)
31 (25-103)
15 (14.5-146.5)
114 (24-386)
SOFA score
11 (10-13)
11 (9-13)
11 (9.5-14)
12 (10-13)
9 (8-10.5)
13 (9-15)
LDH, units/L
876 (562-1929.5)
1114 (532.5-1729.5)
1794 (655.5-5091)
626 (415.5-731)
NA
1698 (1287-2109)
Hemoglobin, g/dL
8.6 (7.6-10.2)
8.75 (7.65-10.18)
7.9 (6.95-9.7)
9.25 (8.275-10.375)
8.3 (7.75-8.85)
9.6 (8.175-11.125)
VA-ECMO indications
Cardiogenic shock
48 (35%)
31 (34%)
11 (41%)
2 (22%)
0 (0%)
4 (44%)
ECPR
25 (18%)
16 (18%)
6 (22%)
0 (0%)
0 (0%)
3 (33%)
Postcardiotomy shock
25 (18%)
14 (15%)
5 (19%)
3 (33%)
1 (33%)
2 (22%)
Cannulation strategy
Central
69 (50%)
46 (51%)
10 (37%)
7 (78%)
1 (33%)
5 (56%)
Peripheral
70 (50%)
45 (49%)
17 (63%)
2 (22%)
2 (66%)
4 (44%)
Discharge location
Home
19 (14%)
14 (15%)
5 (19%)
0 (0%)
0 (0%)
0 (0%)
Acute rehabilitation
12 (9%)
11 (0%)
0 (0%)
0 (0%)
0 (0%)
1 (11%)
Long-term facility
2 (1%)
1 (0%)
1 (4%)
0 (0%)
0 (0%)
0 (0%)
Skilled nursing facility
6 (4%)
6 (7%)
0 (0%)
0 (0%)
0 (0%)
0 (0%)
ECMO duration, h
95.9 (58.3-191.6)
110.8 (59.2-209.7)
93.9 (58.3-150.9)
95.88 (58.32-191.62)
27.52 (24.98-40.77)
83.78 (59.17-143.83)
Mortality
100 (72%)
59 (65%)
21 (78%)
9 (100%)
3 (100%)
8 (89%)
Good neurological outcome, mRS ≤3
23 (17%)
15 (16%)
6 (22%)
0 (0%)
0 (0%)
2 (22%)
Number of SpO2 measurements per patient
36 (18.5-74)
36 (19-72.5)
61(21.5-91.75)
36 (13-50)
5 (3.50-28)
28 (23-45)
PaCO2, Partial pressure of carbon dioxide; PaO2, partial pressure of oxygen; HCO3-, bicarbonate ion; SaO2, arterial gas oxygen saturation; ECMO, extracorporeal membrane oxygenation; AST, aspartate transaminase; ALT, alanine transaminase; SOFA, sequential organ failure assessment score; LDH, lactate dehydrogenase; NA, not available; VA, venoarterial; ECPR, extracorporeal cardiopulmonary resuscitation; mRS, modified Rankin Scale; SpO2, peripheral oxygen saturation.
∗ Variables were collected within the first 12 h of ECMO initiation; creatinine, platelet, lactate, AST, ALT, LDH, and hemoglobin measurements, and SOFA score represent the worst value collected in the first 12 h of ECMO.
Variables were collected within the first 12 h of ECMO initiation; creatinine, platelet, lactate, aspartate transaminase, alanine transaminase, lactate dehydrogenase, and hemoglobin measurements, and SOFA score represent the worst value collected in the first 12 h of ECMO.
∗ Variables were collected within the first 12 h of ECMO initiation; creatinine, platelet, lactate, aspartate transaminase, alanine transaminase, lactate dehydrogenase, and hemoglobin measurements, and SOFA score represent the worst value collected in the first 12 h of ECMO.
Figure 2, A, depicts SpO2–SaO2 for each race/ethnicity, with Black patients receiving VA-ECMO having the greatest discrepancy. White patients had minimal bias (mean difference) at –0.06%, whereas Black, Asian, Hispanic, and nonspecified race patients had estimated biases of 0.94%, 0.2%, 0.03%, and –0.80%, respectively (P < .001 for all, Table E1, Figure E1). Overall, SpO2–SaO2 correlation coefficients were weak for all races/ethnicities (all R < 0.50), with Hispanic patients having the worst overall SpO2–SaO2 correlation (R = 0.39, Table E1, Figure E2). Notably, patients with peripherally cannulated VA-ECMO had a positive estimated bias (0.3%) and stronger SpO2–SaO2 correlation (R = 0.51), whereas centrally cannulated patients had a bias close to zero (0.004%) and worse correlation (R = 0.44, Figure E3, Table E2).
Figure 2Boxplots of patients receiving (A) VA-ECMO and (B) VV-ECMO stratified by race/ethnicity. The horizontal lines represent the median, and the top and bottom ends of each box represent the 75% and 25% limits of the interquartile range, respectively. The lower and upper whiskers represent the minimum and maximum values of nonoutliers, whereas the extra dots represent outliers. Only data points within 2 standard deviations of the mean are shown. Each small dot represents an individual data point (SpO2–SaO2 pair), colored by race. Red, dark yellow, light green, blue, and magenta dots represent White, Black, Asian, Hispanic, and nonspecified race/ethnicity patients, respectively. Wilcoxon rank-sum test was used for comparisons between 2 races. Kruskal–Wallis test was used for a global comparison (significance shown in the top left corner). ∗∗∗∗Indicates a P ≤ .0001. ∗∗∗Indicates P ≤ .001. ∗Indicates P ≤ .05. “Others” refer to nonspecified races (patients who did not identify as Asian, Black, Hispanic, or White). VA, Venoarterial; ECMO, extracorporeal membrane oxygenation; SpO2, oxygen saturation measured by pulse oximetry; SaO2, oxygen saturation measured by arterial gas; VV, venovenous.
There were a total of 422 SpO2–SaO2 pairs where SpO2 overestimated SaO2 by ≥4%. Of these pairs, 40% occurred in White patients, 55% in Black patients (P = .002), 0% in Hispanic patients (P < .001), 2% in Asian patients (P < .001), and 2% in “Others” patients (P < .001). The proportion of matched SpO2–SaO2 measurement pairs in patients receiving VA-ECMO with occult hypoxemia (88 total pairs) from White patients was 27%, whereas this rate was 70% from Black patients, 0% from Hispanics patients, and 1% from both Asian and nonspecified race patients (P < .001 for all).
In unadjusted LMM, compared with White patients, SpO2 overestimated SaO2 by 1.02% for Black patients (95% confidence interval [CI], 0.30%-1.74%, P = .007, Table E3). In our adjusted VA-ECMO LMM, adjusting for age, sex, vasopressor/inotrope usage, central cannulation strategy, pH, temperature, and hemoglobin, SpO2 underestimated SaO2 by −11.38% in nonspecified race patients (95% CI, −22.4% to −0.37%, P = .043), and SpO2 underestimated SaO2 by −3.8% in centrally-cannulated non-specified race patients (95% CI, −7.4% to −0.21%, P = .038, Figure 3, A, Table E4). Other race/ethnicity comparisons were not statistically significant, although the method for evaluating statistical significance in LMM is not entirely clear and thus should be interpreted cautiously.
Figure 3Coefficient plots of both adjusted linear mixed-effects models for patients receiving (A) VA-ECMO and (B) VV-ECMO. A, Age, sex, usage of vasopressor/inotrope during ECMO, and cannulation strategy (central vs peripheral) were included as fixed-effects, whereas pH, temperature, and hemoglobin were incorporated as random-effects. B, Age, sex, and cannulation strategy (single vs double lumen) were included as fixed-effects, whereas lactate dehydrogenase, pH, and temperature were incorporated as random-effects. “Regression estimates” represent the response variable (SpO2–SaO2). “Others” represent nonspecified races (patients who did not identify as Asian, Black, Hispanic, or White). Statistically significant (P < .05) regression estimates are labeled on the coefficient plots. VA, Venoarterial; ECMO, extracorporeal membrane oxygenation; SpO2, oxygen saturation measured by pulse oximetry; SaO2, oxygen saturation measured by arterial gas; VV, venovenous.
Comparing the SpO2–SaO2 difference between different cannulation strategies within VA-ECMO patients through boxplot analyses (Figure E4, A), we found that peripherally cannulated patients had a greater SpO2–SaO2 discrepancy compared with centrally cannulated patients (P < .0001).
In patients receiving VA-ECMO, the 88% SaO2 threshold had the greatest sensitivity (98%), and the 95% SaO2 threshold had the greatest specificity (100%) for SpO2 reliably predicting SaO2 (Figure 4, B, Table E5). The optimal SpO2 value (97%) was greatest for the 88% SaO2 threshold, which was the lowest threshold we assessed (Table E6).
Figure 4Receiver-operating characteristic (ROC) curves and area under the ROC curves (AUCs) with 95% confidence interval limits for (A) all patients receiving extracorporeal membrane oxygenation (ECMO), (B) patients receiving venoarterial (VA) ECMO, and (C) patients receiving venovenous (VV) ECMO. ROC curves are shown for each SaO2 threshold. SaO2 = 88% (red), 92% (green), and 95% (blue), representing the sensitivity (y-axis) and 1—specificity (x-axis) of all cut off points of SpO2 to accurately detect a SaO2 at all three of these predefined SaO2 thresholds. SaO2, Oxygen saturation measured by arterial gas.
Figure 2, B, depicts SpO2–SaO2 for each race/ethnicity, with Black patients receiving VV-ECMO having the greatest discrepancy. Similar to VA-ECMO, White, Black, Asian, Hispanic, and nonspecified race patients had estimated bias (mean difference) values of 0.50%, 2.9%, 1.0%, 1.1%, and −0.53%, respectively (Figure E5, Table E1). Comparing different cannulation strategies, patients with single-lumen cannula had a greater estimated bias (1.8%) and worse SpO2–SaO2 correlation (R = 0.62) than those with double-lumen (1.1% and R = 0.73, Figure E6, Table E2). Overall, correlation coefficients varied by race/ethnicity, with Asians having the worst overall SpO2–SaO2 correlation (R = 0.46, Figure E7).
There were a total of 1706 SpO2–SaO2 pairs where SpO2 overestimated SaO2 by ≥4%. Of these pairs, 19% occurred in White patients, 55% in Black patients (P < .001), 16% in Hispanic patients (P = .06), 9% in Asian patients (P < .001), and 0.4% in “Others” patients (P < .001). The proportion of matched SpO2–SaO2 measurement pairs in patients receiving VV-ECMO with occult hypoxemia (385 total pairs) from White patients was 16%, whereas this rate was 66% from Black (P < .001), 11% from Hispanic (P = .09), 6% from Asian (P < .001), and 1% from nonspecified race patients (P < .001).
In unadjusted LMM, compared with White patients, SpO2 overestimated SaO2 by 2.77% in Black patients (95% CI, 1.57%-3.96%, P < .001, Table E7), and this discrepancy persisted after we adjusted for age, sex, cannulation strategy, LDH, pH, and temperature, as SpO2 still overestimated SaO2 by 0.19% in Black patients (95% CI, 0.045%-0.33%, P = .023, Figure 3, B, Table E8).
Comparing the SpO2–SpO2 difference between different cannulation strategies within patients receiving VV-ECMO through boxplot analyses (Figure E4, B), single-lumen cannulated patients had a greater SpO2–SaO2 discrepancy, compared with double-lumen cannulated patients (P < .0001).
In VV-ECMO, the 88% SaO2 threshold had the greatest sensitivity (76%), and the 95% SaO2 threshold had the greatest specificity (100%) for SpO2 reliably predicting SaO2 (Table E5 and Figure 4, C). The optimal SpO2 value (85%) was greatest for the 88% SaO2 threshold, which was the lowest threshold we assessed (Table E6).
Exploratory Analysis
When analyzing the SpO2–SpO2 discrepancy in both VA and VV-ECMO populations, we found that lower arterial gas oxygen saturation values were correlated with a greater difference between SpO2 and SaO2 (Figure E8). Notably, SpO2 tended to overestimate SaO2 at lower SaO2 values in the VV-ECMO population more frequently than in VA-ECMO; conversely, SpO2 underestimated SaO2 at greater SaO2 values in the VA-ECMO population more frequently than in VV-ECMO.
Discussion
Race/Ethnicity Discrepancy
Herein, we demonstrated that pulse oximetry consistently overestimated SaO2 in both VA- and VV-ECMO populations with a greater SpO2–SaO2 discrepancy in patients receiving VV-ECMO versus VA-ECMO (Figure 5). Furthermore, the discrepancy was increased in Asian, Black, and Hispanic patients receiving ECMO compared with White patients, with Black patients having the greatest overestimation, and therefore, inaccuracy, of their true oxygen levels. Our analysis is clinically important and novel, as we present a detailed racial/ethnical discrepancy in oxygen levels in patients receiving ECMO with rich SpO2 and SaO2 data from a single tertiary academic ECMO center. In addition, our results are particularly clinically significant, as we show a much greater degree of occult hypoxemia occurring in Black patients receiving ECMO as compared with other races/ethnicities, potentially suggesting specific medical management changes unique to these patients, as greater rates of undetected hypoxemia in severely ill patients have been shown to lead to poorer rates of survival.
Analysis of discrepancies between pulse oximetry and arterial oxygen saturation measurements by race and ethnicity and association with organ dysfunction and mortality.
Figure 5Summary of key findings. Pulse oximetry consistently overestimates arterial blood gas oxygen saturation in Black patients undergoing extracorporeal membrane oxygenation (ECMO), compared with White patients undergoing ECMO. This overestimation of their true oxygen saturation places Black patients undergoing ECMO at greater risk of occult hypoxemia (oxygen saturation measured by arterial gas; [SaO2] <88% with a time-matched oxygen saturation measured by pulse oximetry [SpO2] ≥92%), and should be noted when monitoring and treating them in the intensive care unit setting. The image of the ECMO circuit was retrieved from BioRender (www.Biorender.com). VA, Venoarterial; VV, venovenous
showed that SpO2 overestimated SaO2 in Black compared with White patients in adults with respiratory failure, placing Black patients at risk for occult hypoxemia. However, this study was conducted using only a single timepoint of oxygen saturation data inconsistently measured approximately 6 hours before ECMO cannulation. They found a comparable risk of occult hypoxemia in Asian and Hispanic patients compared with White patients, similar to what Wong and colleagues
Analysis of discrepancies between pulse oximetry and arterial oxygen saturation measurements by race and ethnicity and association with organ dysfunction and mortality.
reported in patients in the ICU. Both studies are in line with our study's findings concerning occult hypoxemia in Asian and Hispanic patients receiving ECMO. However, our study reported greater overall SpO2–SaO2 mean differences in Asian and Hispanic patients receiving ECMO, which may be partly explained due to the severity of illness of these patients, coupled with vasopressor/inotrope usage, different blood flow due to ECMO cannulation, and complex physiology, that may exacerbate the inaccuracy of pulse oximetry measurements compared with non-ECMO patients. Accordingly, the SpO2–SaO2 discrepancy in patients receiving ECMO by race/ethnicity is greater because of these factors that underlie ECMO. Furthermore, this exacerbation of SpO2 overestimating SaO2 is apparent in Black and Hispanic patients receiving VV-ECMO who had a greater bias (2.9% and 1.1%) than what was found in Valbuena and colleagues' pre-ECMO cohort study
(1.7% and 0.8%). Because of this significant SpO2–SaO2 discrepancy, ECMO presents a unique challenge in predicting SpO2 based on SaO2, which prompted us to analyze the VA- and VV-ECMO populations separately. In addition, as we surmise the SpO2–SaO2 discrepancy between different races/ethnicities primarily arises due to measuring specific wavelengths of light and calibration to a White person, other measures that have similar methodologies, such as pulse-wave contour analysis and bioimpedance, may have similar racial discrepancies and thus warrant further investigation.
VA-ECMO
To our knowledge, no study has examined the SpO2–SaO2 discrepancy by cannulation strategy in patients receiving VA-ECMO. In addition to discrepancies based on race/ethnicity, we found that patients receiving peripherally cannulated VA-ECMO have a greater SpO2–SaO2 discrepancy compared with patients receiving centrally cannulated VA-ECMO. One explanation for this discrepancy is due to differential hypoxia, which mainly occurs in peripherally cannulated patients.
In contrast, in centrally cannulated VA-ECMO, blood is fed immediately into the ascending aorta, and thus no lowly oxygenated blood should be measured by pulse oximetry or arterial blood gas since they are distal to the mixing zone.
Another potential confounding factor is the usage of vasopressors or inotropes during ECMO. If patients receiving VA-ECMO are on vasopressors, vasoconstriction occurs, leading to capillary constriction, and ultimately elevated pulse oximetry levels.
If patients are on inotropes, vasodilation may occur, leading to decreased SpO2 levels. In our LMM, we accounted for usage of vasopressor/inotrope, including norepinephrine, epinephrine, phenylephrine, dopamine, and vasopressin, and this adjustment resolved the SpO2–SaO2 difference in patients receiving centrally cannulated VA-ECMO, as expected.
Several studies have also shown SpO2 to underestimate SaO2 in critically ill patients,
in line with our results of the nonspecified race patients receiving VA- and VV-ECMO, which may be related to sepsis, peripheral vasodilation, severe inflammation, and venous pulsatility.
VV-ECMO
We demonstrated that patients receiving VV-ECMO who underwent single-lumen cannulation had a greater SpO2 overestimation of SaO2 (1.8%) than double-lumen cannulated patients (1.1%). Although further investigation is required, SpO2 may have overestimated SaO2 in patients receiving VV-ECMO due to greater carboxyhemoglobin levels.
Still, no study has analyzed this discrepancy between cannulation strategies within VV-ECMO. The SpO2–SaO2 discrepancy may potentially result from varying degrees of hemolysis,
Evaluation of recirculation during venovenous extracorporeal membrane oxygenation using computational fluid dynamics incorporating fluid–structure interaction.
We reported the presence of more frequent SpO2 overestimation of SaO2 values at lower oxygen saturation levels, especially in the VV-ECMO population, as these patients are at a greater risk for blood gas derangements, given their primary respiratory failure. Accordingly, we hypothesized that the VV-ECMO population had a greater SpO2–SaO2 discrepancy than the VA-ECMO population, which was confirmed in our data and consistent with previous data.
Accuracy, Precision, Sensitivity and Specificity, and Threshold
Overall, the area under the ROC curves for SpO2 predicting SaO2 values were lower for patients receiving VA-ECMO than for patients receiving VV-ECMO, although patients receiving VV-ECMO had a greater SpO2–SaO2 mean difference in our Bland–Altman analyses than patients receiving VA-ECMO. For all ECMO patients together, SpO2 was less accurate at greater SaO2 values. These results surprisingly oppose those found in evaluating pulse oximetry in critical care patients in other studies,
thus raising the need for additional studies regarding pulse oximetry accuracy and thresholds in this unique patient population.
Limitations
Our study is limited by the lack of consistency in the placement (location) of the pulse oximeter probe in both patients undergoing VA-ECMO and VV-ECMO, thus leading to variation in accuracy and bias of the device. Furthermore, exact locations of the probes correlating with each SpO2 measurement were also not recorded. In addition, other potential confounders such as patient's skin temperature and ECMO flow and sweep should be noted. The majority of patients receiving VA-ECMO were White, and the numbers of non-White and non-Black patients are relatively small and, thus, an external validation is necessary with a larger sample size; however, there was a more equal distribution of each race/ethnicity in the VV-ECMO population, suggesting a more robust generalizability of our results in this cohort. Another limitation in this study and those preceding ours is equalizing race and ethnicity with an individual's skin color. Furthermore, because of the small numbers of patients in this analysis, dividing the Asian racial group between northern and southern Asia, which likely have different skin color and genetics, was not possible. In addition, as a single-institution and retrospective observational study, prospective multicenter studies are required to validate our findings.
Conclusions
Pulse oximetry is a widely used, noninvasive method of obtaining a patient's oxygen saturation. We demonstrated that the pulse oximetry and arterial blood gas oxygen saturation discrepancy, between different races, is clinically relevant in patients supported with ECMO. Furthermore, our results imply prospective physiological explanations for this discrepancy; specifically, further analysis of granular data such as vasoactive-inotropic score and LDH should be further examined.
Conflict of Interest Statement
The authors reported no conflicts of interest.
The Journal policy requires editors and reviewers to disclose conflicts of interest and to decline handling or reviewing manuscripts for which they may have a conflict of interest. The editors and reviewers of this article have no conflicts of interest.
We are indebted to the patients supported with ECMO participating in our study.
Appendix E1
Figure E1Bland–Altman plots of (SpO2 + SaO2)/2 (x-axis) vs SpO2 – SaO2 (y-axis), highlighting the discrepancy between SpO2 and SaO2 values in patients receiving venoarterial extracorporeal membrane oxygenation (VA-ECMO) separated by race and ethnicity. A, Red, yellow–green, light green, blue, and purple dots represent White, Black, Asian, Hispanic, and others (nonspecified races) patients receiving VA-ECMO, respectively. The upper red dashed line represents the upper 95% confidence interval limit of agreement (7.0%), and the lower red dashed line represents the lower 95% confidence interval limit of agreement (–6.7%), within which 95% of the differences between SpO2 and SaO2 fall. The solid black line represents the mean difference between SpO2 and SaO2 (0.15%), calculated by first determining the averages of SpO2 and SaO2, individually, and then subtracting the average SaO2 from the average SpO2 value. The further away SpO2 and SaO2 data points are vertically from the solid black line, the more discrepant SpO2 and SaO2 are. Data points above the solid black line indicate SpO2 overestimated SaO2 whereas data points below indicate SpO2 underestimated SaO2. Panels B, C, D, E, and F are separate Bland–Altman plots for each race: White, Black, Asian, Hispanic, and others, respectively. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry.
Figure E2Scatterplots of pulse oximetry oxygen saturation levels (SpO2, x-axis) vs arterial oxygen saturation levels (SaO2, y-axis) in patients receiving venoarterial extracorporeal membrane oxygenation (VA-ECMO) separated by race and ethnicity. Red, yellow–green, light green, blue, and purple dots represent White, Black, Asian, Hispanic, and others (nonspecified races) patients receiving VA-ECMO, respectively. A 45-degree line is shown to model the ideal 1:1 ratio of SpO2 to SaO2, further highlighting the discrepancy between SpO2 and SaO2 values for patients receiving VA-ECMO between different races and ethnicities, as most SpO2 and SaO2 points do not fit on this 45-degree line. Pearson correlation coefficients were generated to measure the linear correlation between SpO2 and SaO2. Panels B, C, D, E, and F are separate scatterplots for each race: White, Black, Asian, Hispanic, and others, respectively. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry.
Figure E3Scatterplots of pulse oximetry oxygen saturation levels (SpO2, x-axis) vs arterial oxygen saturation levels (SaO2, y-axis) and Bland–Altman plots of (SpO2 + SaO2)/2 (x-axis) vs SpO2 – SaO2 (y-axis) in patients receiving venoarterial extracorporeal membrane oxygenation (VA-ECMO) stratified by cannulation strategy. A, Brown and pink dots represent centrally and peripherally cannulated patients receiving VA-ECMO, respectively. A 45-degree line is shown to model the ideal 1:1 ratio of SpO2 to SaO2, further highlighting the discrepancy between SpO2 and SaO2 values for patients with VA-ECMO that differ by cannulation strategy as most SpO2 and SaO2 points do not fit on this 45-degree line. Pearson correlation coefficients were generated to measure the linear correlation between SpO2 and SaO2. Panels C and E are separate scatterplots for patients undergoing centrally- and peripherally-cannulated VA-ECMO, respectively. B, The upper red dashed line represents the upper 95% confidence interval limit of agreement (7.0%), and the lower red dashed line represents the lower 95% confidence interval limit of agreement (–6.7%), within which 95% of the differences between SpO2 and SaO2 fall. The solid black line represents the mean difference between SpO2 and SaO2 (0.15%), calculated by first determining the averages of SpO2 and SaO2, individually, and then subtracting the average SaO2 from the average SpO2 value. Panels D and F are separate Bland–Altman plots for patients receiving centrally and peripherally cannulated VA-ECMO, respectively. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry.
Figure E4Boxplots of SpO2–SaO2 discrepancy by cannulation strategies in patients receiving venoarterial extracorporeal membrane oxygenation (VA-ECMO) and venovenous (VV)-ECMO. In these boxplots of VA-ECMO (A) and VV-ECMO (B) patients stratified by cannulation strategy, the horizontal lines represent the median, and the top and bottom ends of each box represent the 75% and 25% limits of the interquartile range, respectively. Only data points within 2 standard deviations of the mean are shown. Each small dot represents an individual data point (SpO2–SaO2 pair), colored by cannulation strategy. In (A), brown and pink dots represent patients receiving centrally and peripherally cannulated VA-ECMO, respectively. In B, green and violet dots represent single-lumen (two sites) and double-lumen (one site) cannulation strategy patients receiving VV-ECMO. Wilcoxon rank-sum test was used for comparisons between 2 different cannulation strategies. ∗∗∗∗ indicates a P ≤ .0001. SpO2, Oxygen saturation measured by pulse oximetry; SaO2, oxygen saturation measured by arterial gas.
Figure E5Bland–Altman plot of (SpO2 + SaO2)/2 (x-axis) vs SpO2 – SaO2 (y-axis), highlighting the discrepancy between SpO2 and SaO2 values in patients receiving venovenous extracorporeal membrane oxygenation (VV-ECMO) separated by race and ethnicity. A, Red, yellow–green, light green, blue, and purple dots represent White, Black, Asian, Hispanic, and others (nonspecified races) patients receiving VV-ECMO, respectively. The upper red dashed line represents the upper 95% confidence interval limit of agreement (9.2%), and the lower red dashed line represents the lower 95% confidence interval limit of agreement (–6.4%), within which 95% of the differences between SpO2 and SaO2 fall. The solid black line represents the mean difference between SpO2 and SaO2 (1.4%), calculated by first determining the averages of SpO2 and SaO2, individually, and then subtracting the average SaO2 from the average SpO2 value. The further away SpO2 and SaO2 data points are vertically from the solid black line, the more discrepant SpO2 and SaO2 are. Data points above the solid black line indicate SpO2 overestimated SaO2, whereas data points below indicate SpO2 underestimated SaO2. Panels B, C, D, E, and F are separate Bland–Altman plots for each race: White, Black, Asian, Hispanic, and others, respectively. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry.
Figure E6Scatterplot of pulse oximetry oxygen saturation levels (SpO2, x-axis) vs arterial oxygen saturation levels (SaO2, y-axis) and Bland–Altman plot of (SpO2 + SaO2)/2 (x-axis) vs SpO2 – SaO2 (y-axis), in patients receiving venovenous extracorporeal membrane oxygenation (VV-ECMO) stratified by cannulation strategy. A, Green and purple dots represent single- and double-lumen cannulated patients receiving VV-ECMO, respectively. A 45-degree line is shown to model the ideal 1:1 ratio of SpO2 to SaO2, further highlighting the discrepancy between SpO2 and SaO2 values for patients receiving VV-ECMO that differ by cannulation strategy, as most SpO2 and SaO2 points do not fit on this 45-degree line. Pearson correlation coefficients were generated to measure the linear correlation between SpO2 and SaO2. C and E are separate scatterplots for single- and double-lumen cannulated VV-ECMO patients, respectively. B, The upper red dashed line represents the upper 95% confidence interval limit of agreement (9.2%), and the lower red dashed line represents the lower 95% confidence interval limit of agreement (–6.4%), within which 95% of the differences between SpO2 and SaO2 fall. The solid black line represents the mean difference between SpO2 and SaO2 (1.4%), calculated by first determining the averages of SpO2 and SaO2, individually, and then subtracting the average SaO2 from the average SpO2 value. D and F are separate Bland–Altman plots for patients receiving single- and double-lumen cannulated VV-ECMO, respectively. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry.
Figure E7Scatterplot of pulse oximetry oxygen saturation levels (SpO2, x-axis) vs arterial oxygen saturation levels (SaO2, y-axis) in patients receiving venovenous extracorporeal membrane oxygenation (VV-ECMO) separated by race and ethnicity. A, Red, yellow–green, light green, blue, and purple dots represent White, Black, Asian, Hispanic, and others (nonspecified races). A 45-degree line is shown to model the ideal 1:1 ratio of SpO2 to SaO2, further highlighting the discrepancy between SpO2 and SaO2 values for VV-ECMO patients between different races and ethnicities as most SpO2 and SaO2 points do not fit on this 45-degree line. Pearson correlation coefficients were generated to measure the linear correlation between SpO2 and SaO2. B, C, D, E, and F are separate scatterplots for each race: White, Black, Asian, Hispanic, and others, respectively. SaO2, Oxygen saturation measured by arterial gas; SpO2, oxygen saturation measured by pulse oximetry.
Figure E8Difference between SpO2 and SaO2 (y-axis) plotted as a function of SaO2 (x-axis). As oxygen saturation decreases, SpO2–SaO2 increases. There are more SpO2–SaO2 pairs at lower saturations with a greater SpO2–SaO2 discrepancy (SpO2overestimating SaO2) in patients receiving venovenous extracorporeal membrane oxygenation (VV-ECMO) patients (red dots) versus patients receiving venoarterial extracorporeal membrane oxygenation (VA-ECMO) (turquoise dots). Conversely, there were more SpO2–SaO2 pairs at greater oxygen saturations where SpO2underestimated SaO2 in the VA-ECMO population. SpO2, Oxygen saturation measured by pulse oximetry; SaO2, oxygen saturation measured by arterial gas.
Table E1Estimated bias, precision, upper and lower limits of agreement, root mean square error, and Pearson correlation coefficient, stratified by race and ethnicity in patients receiving ECMO in all SpO2–SaO2 pairs measured 10 minutes or less between each other
Table E2Estimated bias, precision, upper and lower limits of agreement, root mean square error, and Pearson correlation coefficient, stratified by cannulation strategy in patients receiving ECMO in all SpO2–SaO2 pairs measured 10 minutes or less between each other
Venoarterial (VA) ECMO
Venovenous (VV) ECMO
All with cannulation strategy (139 patients, 7295 pairs)
Central (69 patients, 3924 pairs)
Peripheral (70 patients, 3371 pairs)
All with cannulation strategy (56 patients, 8805 pairs)
Table E3Unadjusted venoarterial ECMO linear mixed-effects model assessing the relationship between race and SpO2–SaO2 for all time-matched and ≥70% oxygen saturation data points
Table E4Adjusted venoarterial ECMO linear mixed-effects model assessing the relationship between race and each covariate as a fixed effect for all SpO2–SaO2 time-matched and ≥70% oxygen saturation data points
Table E5ROC and AUC analyses for SpO2 to accurately detect 88%, 92%, and 95% SaO2 thresholds
Patient type
SaO2 threshold (%)
AUC
Specificity (%)
Sensitivity (%)
All (VA-ECMO + VV-ECMO)
88
0.888 (0.879-0.897)
76
88
92
0.868 (0.862-0.875)
95
48
95
0.850 (0.844-0.856)
100
0
VA-ECMO only
88
0.802 (0.766-0.837)
23
98
92
0.822 (0.804-0.840)
69
83
95
0.802 (0.789-0.815)
100
0
VV-ECMO only
88
0.880 (0.870-0.890)
84
76
92
0.855 (0.847-0.863)
98
26
95
0.846 (0.838-0.854)
100
0
SaO2, Oxygen saturation measured by arterial gas; AUC, area under the receiver-operating characteristic curve; VA, venoarterial; ECMO, extracorporeal membrane oxygenation; VV, venovenous.
Table E7Unadjusted venovenous ECMO linear mixed-effects model assessing the relationship between race and SpO2–SaO2 for all time-matched and ≥70% oxygen saturation data points
Table E8Adjusted venovenous ECMO linear mixed-effects model assessing the relationship between race and each covariate as a fixed effect for all SpO2–SaO2 time-matched and ≥70% oxygen saturation data points
Racial bias and reproducibility in pulse oximetry among medical and surgical inpatients in general care in the Veterans Health Administration 2013-19: multicenter, retrospective cohort study.
Analysis of discrepancies between pulse oximetry and arterial oxygen saturation measurements by race and ethnicity and association with organ dysfunction and mortality.
Tidal volume reduction for prevention of ventilator-induced lung injury in acute respiratory distress syndrome. The Multicenter Trail Group on Tidal Volume reduction in ARDS.
Evaluation of recirculation during venovenous extracorporeal membrane oxygenation using computational fluid dynamics incorporating fluid–structure interaction.
S.P.K. is supported by the National Heart, Lung, and Blood Institute (NHLBI; 5K08HL14332). S.M.C. is supported by NHLBI (1K23HL157610).
Drs Kim and Cho contributed equally to this article.
HERALD (Hopkins Education, Research, and Advancement in Life support Devices) Investigators: Kate Calligy, RN, Patricia Brown, RD-AP, LDN, CNSC, Diane Alejo, BS, Scott Anderson, BA, Matthew Acton, MD, Hannah Rando, MD, Henry Chang, MD, PhD.
00038-4&id=fx1.jpg){kind=link}
00038-4&id=fx2.jpg){kind=link}
00038-4&id=gr1.jpg){kind=link}
00038-4&id=gr2.jpg){kind=link}
00038-4&id=gr3.jpg){kind=link}
00038-4&id=gr4.jpg){kind=link}
00038-4&id=gr5.jpg){kind=link}
00038-4&id=fx3.jpg){kind=link}
00038-4&id=fx4.jpg){kind=link}
00038-4&id=fx5.jpg){kind=link}
00038-4&id=fx6.jpg){kind=link}
00038-4&id=fx7.jpg){kind=link}
00038-4&id=fx8.jpg){kind=link}
00038-4&id=fx9.jpg){kind=link}
00038-4&id=fx10.jpg){kind=link}