ECMO Physiological Factors Influence Pulse Oximetry and Arterial Oxygen Saturation Discrepancies

Abstract

Background:

Cannulation strategy, vasopressors, and hemolysis are important physiological factors that influence hemodynamics in extracorporeal membrane oxygenation (ECMO). We hypothesized these factors influence the discrepancy between oxygen saturation measured by pulse oximetry (SpO₂) and arterial blood gas (SaO₂) in patients on ECMO.

Methods:

We retrospectively analyzed adults (≥18 years) on venoarterial- or venovenous-ECMO at a tertiary academic ECMO center. SpO₂-SaO₂ pairs with oxygen saturation ≥70% and measured within 10 minutes were included. Occult hypoxemia was defined as SaO₂≤88% with a time-matched SpO₂≥92%. Adjusted linear mixed-effects modeling was used to assess the SpO₂-SaO₂ discrepancy with pre-selected demographics and time-matched laboratory variables. Vasopressor usage was quantified by Vasopressor Dose Equivalences.

Results:

Of 139 venoarterial-ECMO and 88 venovenous-ECMO patients, we examined 20,053 SpO₂-SaO₂ pairs. The SpO₂–SaO₂ discrepancy was greater in venovenous-ECMO (1.15%) vs. venoarterial-ECMO (−0.35%, p<0.001). Overall, 81 patients (35%) experienced occult hypoxemia during ECMO. Occult hypoxemia was more common in venovenous-ECMO (65%) vs. venoarterial-ECMO (17%, p<0.001).

In linear mixed-effects modeling, SpO₂ underestimated SaO₂ by 9.48% in central vs. peripheral venoarterial-ECMO (95%CI=−17.1% to −1.79%, p=0.02). Higher Vasopressor Dose Equivalences significantly worsened the SpO₂-SaO₂ discrepancy (p<0.001).

In linear mixed-effects modeling, SpO₂ overestimated SaO₂ by 25.43% in single-lumen-cannulated vs. double-lumen-cannulated venovenous-ECMO (95%CI=5.27%-45.6%, p=0.03). Higher Vasopressor Dose Equivalences and lactate dehydrogenase levels significantly worsened the SpO₂-SaO₂ discrepancy (p<0.001).

Conclusions:

Venovenous-ECMO patients are at higher risk for occult hypoxemia compared to venoarterial-ECMO. A higher vasopressor requirement and different cannulation strategies (central venoarterial-ECMO; single-lumen venovenous-ECMO) were significant factors for clinically significant SpO₂-SaO₂ discrepancy in both ECMO modes.

Introduction

Peripheral oxygen saturation (SpO₂) has been shown to be inaccurate in estimating arterial gas oxygen saturation measurements, compared to the gold-standard for estimating a patient’s true oxygen saturation (SaO₂). While these discrepancies have been attributed to race and ethnicity, [1-3] and biological factors such as carboxyhemoglobin [4, 5] and sepsis, [6] there are limited granular data showing how physiological factors such as hemolysis [4] are associated with significant SpO₂-SaO₂ discrepancy in extracorporeal membrane oxygenation (ECMO) patients. To date, no studies have investigated the association between vasopressor usage and SpO₂-SaO₂ discrepancy in ECMO patients. Furthermore, no prior study has analyzed this discrepancy within VA- (central vs. peripheral) and VV-ECMO (single- vs. double-lumen) cannulation strategies separately.

Higher vasopressor usage has been associated with poor clinical outcomes in infant cardiac surgery patients. [7] Prior non-ECMO studies have found that vasopressor and inotrope usage leads to pulse oximetry inaccuracy through vasoconstriction with corresponding capillary constriction (vasopressor), and vasodilation with corresponding capillary vasodilation (inotrope), respectively. [8] This may be problematic in ECMO patients as they often require heavy doses of vasopressor and inotrope. Additionally, in ECMO, hemolysis commonly occurs, which may play a role in the SpO₂-SaO₂ discrepancy through carboxyhemoglobin generation. [9]. Moreover, inherent differences with blood flow resistance and cross-sectional area [10] between single- and double-lumen cannula in VV-ECMO may also lead to different frequency of hemolysis.

Herein, we aimed to use granular data to demonstrate that the SpO₂-SaO₂ discrepancy in ECMO, in conjunction with race/ethnicity, is exacerbated by 1) high vasopressor usage in VA-ECMO patients, and 2) hemolysis, in VV-ECMO patients.

Materials and Methods

Study Design and Participants

This study was approved by the Johns Hopkins Hospital Institutional Review Board (IRB00216321) on 10/22/2019. A retrospective analysis of a database containing patients undergoing ECMO at a tertiary care center between June 2016 and April 2022 was conducted. We included all adult patients (age ≥18 years) who received VA-ECMO and VV-ECMO.

Data Collection

We collected SaO₂ measured by arterial blood gas (ABG) and SpO₂ measured by pulse oximetry during ECMO cannulation by electronic medical records. Pre-cannulation characteristics included demographics, medical history, and on-ECMO physiological and laboratory variables were also collected by electronic medical records. As a clinical protocol, ABGs were collected every 2–4 hours during ECMO support, and SpO₂ was recorded multiple times a day, usually at least once every 15 minutes. ABGs were acquired pre- and post-ECMO cannulation. If clinically indicated, more frequent ABGs were obtained. A right radial arterial line was placed for 1) accurate and recurrent ABG measurements and 2) as a differential hypoxia marker that was utilized in all VA-ECMO patients. The SaO₂ originated from the partial pressure of oxygen. Hemoglobin, temperature, bilirubin, and lactate dehydrogenase (LDH) were collected at least once daily during ECMO. Pump speeds were recorded every 4 hours during ECMO. Vasopressor dosage was recorded multiple times daily, usually at least once every hour during ECMO. The pulse oximeter probe was placed on the right finger or earlobe in VA-ECMO patients. The pulse oximeter probe was placed on the right or left hand in VV-ECMO patients.

Definitions

SpO₂ represents oxygen saturation measured by pulse oximetry (peripheral). SaO₂ represents arterial oxygen saturation measured by ABG. We excluded SpO₂ and SaO₂ measurements less than 70% as we considered these values to be erroneous, in line with previous landmark studies as such values are likely due to venous blood gases mislabeled as arterial blood gases or data entry error.[2, 3] Time-matching was implemented to control for variation for all SpO₂-SaO₂ pairs; only pairs ≤10 minutes apart were analyzed. VA-ECMO patients were defined as centrally- or peripherally-cannulated. VV-ECMO patients were defined as single-lumen- or double-lumen-cannulated. Vasopressor Dose Equivalence (VDE) scores were used to quantify vasopressor usage and were recalculated for each dose change, similar to the Vasoactive-Inotropic Score (VIS) which has been used in cardiac surgery patients . [11] Derived from a scoping review, the VDE score is an evidence-based formula using equipotent ratios of vasopressors specific to intensive care unit patients. [12] A VDE score of 14 was considered clinically significant based on previous literature in mechanically-ventilated patients as scores ≥14 were associated with enteral nutrition intolerance and corresponding higher mortality rates. [13] We considered the following previously established thresholds to be significant for hemolysis: LDH >1000 units/L, [14] revolutions per minute (RPM) >4500, [15] and hemoglobin <12 g/dL. [16] Occult hypoxemia was defined as SaO₂≤88% with SpO₂≥92%.

Outcomes

The primary outcome was the presence of the SpO₂-SaO₂ discrepancy and occult hypoxemia. The secondary outcome was assessing the association between the SpO₂-SaO₂ discrepancy and mortality.

Statistical Analysis

Data were presented as median with interquartile range for continuous variables and absolute numbers with percentages for categorical variables. When comparing data with continuous and categorical variables, the Wilcoxon rank-sum and Pearson’s χ2 tests were used, respectively. Fisher Z transformation test was used to compare Pearson correlation coefficients between two groups. Differences between SpO₂-SaO₂ (defined as SpO₂ minus SaO₂) pairs across different individuals were compared using the Wilcoxon rank-sum and Kruskal-Wallis tests. A p-value ≤0.05 was considered statistically significant. Bland-Altman was used to assess agreement and visualize differences using the same scale between SpO₂ and SaO₂. Scatterplots were used to assess correlation. Boxplots were used as a descriptive technique to visualize distribution within data. Mean difference (estimated bias) was calculated by subtracting SaO₂ from SpO₂, and then taking the mean (average) of this difference for each SpO₂-SaO₂ pair. Precision was calculated by taking the standard deviation of the mean difference. The limits of agreement were calculated by mean difference +/− 1.96*precision, and root mean square error = square-root(((mean difference - precision)^2)).

The association between cannulation strategy and SpO₂-SaO₂ was studied first using unadjusted linear mixed-effects modeling (LMM), with each patient as a random effect. LMM was conducted to allow for random-effects and account for variability within-subjects and within-groups as there is non-independence in our dataset. Coefficient plots were used to visualize corresponding adjusted estimates. In the adjusted LMM, specific demographics and time-dependent laboratory variables theorized to be correlated with pulse oximetry inaccuracy were included [9]. The covariates were age, sex, body mass index, and race/ethnicity; time-dependent laboratory variables included bilirubin, temperature, LDH, hemoglobin, pump speeds, and VDE scores. Time-independent and time-dependent variables were included as fixed and random effects, respectively, in the LMM. All time-dependent variables were time-matched to be as close as possible to the time of sampling of SpO₂-SaO₂ pairs. Time-dependent variables were considered valid for as many time-matched SpO₂-SaO₂ pairs as appropriate if that was the closest data point to the SpO₂-SaO₂ pair.

To quantify the dose of vasopressors each patient ingested while on ECMO, we calculated the VDE score [12]: Norepinephrine dose (μg/kg/min) x 100 + Epinephrine dose (μg/kg/min) x 100 + Phenylephrine dose (μg/kg/min) x 10 + Dopamine dose (μg/kg/min) x 1 + Vasopressin dose (U/min) x 250 + Angiotensin II dose (μg/kg/min) x 1000 + Metaraminol dose (μg/kg/min) x 12.5. All statistical analyses were performed using R Studio (R 4.1.2, 2022).

Results

56,463 SaO₂ and 514,114 SpO₂ datapoints (n=296) were extracted, yielding 50,572 time-matched SpO₂-SaO₂ pairs (n=295). 27,522 SpO₂-SaO₂ pairs were further excluded, for a total of 23,050 SpO₂-SaO₂ pairs (139 VA-ECMO and 88 VV-ECMO, Figure 1).

Figure 1.

Creation of study cohort.

Demographics and clinical characteristics are presented in Supplemental Tables 1 and 2. Of 139 VA-ECMO patients (median age=59; 65% male), 57 were centrally-cannulated and 82 were peripherally-cannulated. Of 88 VV-ECMO patients (median age=48; 68% male), 41 underwent single-lumen cannulation while 47 underwent double-lumen cannulation. The overall estimated bias was higher for VV-ECMO patients (1.15%) than VA-ECMO patients (−0.36%, p<0.0001, Supplemental Figure 1A).

VDE Score and SpO₂-SaO₂ discrepancy in VA-ECMO

The median VDE score was 4.00 (interquartile range=2.00-7.00). Boxplot analyses revealed SpO₂ underestimated SaO₂ more (mean value) at VDE scores ≥14 (Figure 2A, p<0.001). Bland-Altman analyses displayed that SpO₂ underestimated SaO₂ by 1.5% at VDE scores ≥14, while it underestimated SaO₂ by only 0.22% at VDE scores <14 (p<0.001, Table 1, Supplemental Figure 2A-B). Additionally, scatterplots conveyed a weaker SpO₂-SaO₂ correlation in VA-ECMO patients with VDE scores ≥14 (R=0.15, p<0.001) compared to <14 (R=0.36, p<0.001, Table 1, Supplemental Figure 2C-D). Comparing these two correlation coefficients yielded statistical significance (p=0.05). Subgroup analysis using hemolysis markers was also performed in VA-ECMO patients (Supplemental Table 3).

Figure 2.

Boxplots of (A) venoarterial-ECMO by Vasopressor Dose Equivalence scores and (B) venovenous-ECMO by lactate dehydrogenase where each dot represents a SpO₂-SaO₂ pair. **** indicates p≤0.0001. *** indicates p≤0.0001 ** Indicates p≤0.01

Table 1.

Estimated bias, precision, upper and lower limits of agreement, root mean square error, and Pearson correlation coefficient.

	VA-ECMO			VV-ECMO
	VDE Score ≥14 (778 SpO2-SaO2 pairs, 86 patients)	VDE Score <14 (6,981 SpO2-SaO2 pairs, 135 patients)	VDE Score ≥14 (900 SpO2-SaO2 pairs, 61 patients)	VDE Score <14 (14,235 SpO2-SaO2 pairs, 87 patients)	LDH 0-500 (7,273 SpO2-SaO2 pairs, 64 patients)	LDH 500-1000 (5,748 SpO2-SaO2 pairs, 72 patients)	LDH > 1000 (1,642 SpO2-SaO2 pairs, 38 patients)
Estimated Bias, % (Mean Difference, SpO2-SaO2 )	−1.5	−0.2	1.1	1.2	0.9	1.2	1.9
Precision, % (Standard Deviation)	5.9	3.9	4.9	3.8	3.7	3.7	4.8
Upper Limit of Agreement, % (95% C.I. Limit)	10.1	7.4	10.7	8.6	8.2	8.4	11.3
Lower Limit of Agreement, % (95% C.I. Limit)	−13.1	−7.9	−8.4	−6.2	−6.3	−6.1	−7.4
Root Mean Square Error, %	7.4	4.1	3.8	2.6	2.8	2.5	3.6
Pearson Correlation Coefficient (p-value)	0.15 (P<.001)	0.36 (P<.001)	0.65 (P<.001)	0.70 (P<.001)	0.72 (P<.001)	0.72 (P<.001)	0.60 (P<.001)

In LMM, adjusting for cannulation strategy, age, sex, race/ethnicity, bilirubin, temperature, hemoglobin, pump speeds, and treating VDE scores ≥14 as a binary variable, SpO₂ underestimated SaO₂ by 4.00% in centrally-cannulated patients (95%CI=−7.8% to −0.271%, p=0.036, Supplemental Table 4) vs. peripherally-cannulated.

In sensitivity analysis, with the same adjustment except treating VDE scores as a continuous variable, SpO₂ underestimated SaO₂ by 9.48% in centrally-cannulated patients (95%CI=−17.1% to −1.79%, p=0.022, Figure 3A, Supplemental Table 5) vs. peripherally-cannulated. Higher VDE scores significantly worsened the discrepancy (p<0.001, likelihood ratio test).

Figure 3.

Coefficient plots of adjusted linear mixed-effects models for (A) venoarterial-ECMO and (B) venovenous-ECMO patients. “Regression Estimates” represent the response variable (SpO₂–SaO₂).

Hemolysis and SpO₂-SaO₂ discrepancy in VV-ECMO

Figure 2A and Table 1 depict SpO₂-SaO₂ for three different LDH thresholds in VV-ECMO patients, with an LDH threshold >1000 units/L observing the greatest mean difference (1.94%), compared to LDH=0-500 units/L (0.94%, p<0.001) and LDH=500-1000 units/L (1.17% mean difference, p<0.01). Additionally, scatterplots revealed that SpO₂ was the least correlated to SaO₂ at LDH levels >1000 (R=0.60, p<0.001), though comparing R values to LDH=500-100 and LDH <500 did not yield statistical significance (p=0.15 and p=0.16, respectively). As total bilirubin increased in VV-ECMO patients, there was greater SpO₂-SaO₂ discrepancy (Supplemental Figure 2E-F, R=0.13, p<0.001).

In LMM, adjusting for cannulation strategy, age, sex, race/ethnicity, bilirubin, temperature, hemoglobin, pump speeds, VDE scores, and treating LDH ≥1000 as a binary variable, SpO₂ overestimated SaO₂ in single-lumen-cannulated patients by 18.0% (95%CI=0.35%-32.5%, p=0.001, Supplemental Table 6) vs. double-lumen-cannulated.

In sensitivity analysis, with the same adjustment except treating LDH as a continuous variable, SpO₂ overestimated SaO₂ in single-lumen-cannulated patients by 25.43% (95%CI=5.27%−45.6%, p=0.025, Figure 3B, Supplemental Table 7) vs. double-lumen-cannulated. Higher VDE scores and LDH values were significant factors for SpO₂-SaO₂ discrepancy (p<0.001, likelihood ratio test).

The exploratory analysis of the impact of RPM and hemoglobin on the SpO₂-SaO₂ discrepancy is presented in Supplemental Figures 3 and 4 and Supplemental Table 8.

Occult Hypoxemia

Overall, 81 patients (35%) experienced occult hypoxemia during ECMO support. Occult hypoxemia was more common in venovenous-ECMO (65%) vs. venoarterial-ECMO (17%, p<0.001).

SpO₂-SaO₂ discrepancy and Mortality

A total of 123 ECMO patients (54%) and 718 SpO₂-SaO₂ pairs showed the discrepancy >10% (absolute value of SpO₂–SaO₂). Of 60 VA-ECMO patients with >10% discrepancy (290 pairs), 34 patients died vs. 81 VA-ECMO patients with ≤10% discrepancy, 47 died (mortality 57% vs. 58%, odds ratio=0.95, 95%CI=0.48-1.87, p=0.87). Of 63 VV-ECMO patients with >10% discrepancy (428 pairs), 49 patients died vs. 25 VV-ECMO patients with ≤10% discrepancy, 8 patients died (mortality 78% vs. 32%, odds ratio=7.16, 95%CI=2.62-21.3, p<0.001, Supplemental Table 9).

In multivariable logistic regression, adjusting for pre-specified factors for mortality including age, single-lumen-cannulation, VDE score, on-ECMO APACHE score, and pre-ECMO pCO₂, the SpO₂-SaO₂ discrepancy was still an independent risk factor for mortality in VV-ECMO patients (adjusted odds ratio= 1.14, 95%CI=1.10-1.19, p<0.001, Supplemental Table 10).

ECMO Duration

When analyzing the SpO₂-SaO₂ discrepancy over the duration of ECMO support, the SpO₂-SaO₂ discrepancy equalized as ECMO duration increased in both VA- and VV-ECMO populations (Supplemental Figure 5).

Comment

We demonstrated there are physiological reasons explaining why pulse oximetry inaccurately predicts ABG oxygen saturation in both VA- and VV-ECMO populations. In VA-ECMO patients, higher usage of vasopressors was associated with SpO₂ underestimating SaO_2, likely due to higher doses of vasopressors causing lower perfusion in patient’s extremities, while increased hemolysis was associated with SpO₂ overestimating SaO₂ in VV-ECMO patients. Furthermore, after adjusting for vasopressor usage, hemolysis, and ECMO-specific covariates, we still observed that SpO₂ underestimated SaO₂ more in centrally- vs. peripherally-cannulated VA-ECMO patients and SpO₂ overestimated SaO₂ more in single- vs. double-lumen-cannulated VV-ECMO patients. Additionally, we demonstrated that occult hypoxemia and an SpO₂-SaO₂ discrepancy >10% occurs frequently in VV-ECMO patients.

With an estimated bias of almost 3%, Black VV-ECMO patients in our previous study [17] may have surpassed any unadjusted SpO₂-SaO₂ discrepancy in other studies with different populations that investigated skin pigmentation. These results suggest there is a physiological basis that explains the extreme differences in oxygen saturation measurements observed in our ECMO patient cohort. Moreover, based on our adjustment of key variables in our LMM coupled with our findings that the SpO₂-SaO₂ discrepancy is greatest at the beginning of ECMO cannulation (Supplemental Figure 5), it would be prudent to monitor SaO₂ closely even if SpO₂ measurements are normal in patients with risk factors like high vasopressor usage, central- and single-lumen-cannulation, hemolysis, older age, male sex, and Black race/ethnicity, especially during early ECMO support.

Vasopressor Usage (VA-ECMO)

Although other studies have investigated vasopressor usage and pulse oximetry inaccuracy in non-ECMO populations [6, 18, 19] we are the first group to study this relationship meticulously in ECMO patients with VDE scores and adjustment of pre-specified laboratory variables and clinical characteristics that may influence this discrepancy, suggesting this discrepancy originates from more than just race/ethnicity. Although Wilson [20] found that vasopressors did not affect pulse oximetry in septic shock patients in a smaller cohort (n=88), they did find that vasopressors decreased the precision of the device. This finding is in line with our study, which found a precision of 5.9% in VA-ECMO patients with VDE scores ≥14, compared to only a 3.9% precision in those with VDE scores <14. We also found that VDE scores ≥14 had a much greater range of upper and lower limits of agreement in Bland-Altman analyses (−13.1%−10.1%), compared to VDE scores <14 (−7.9%−7.4%), suggesting this threshold is likely clinically meaningful. Additionally, based on our LMM findings, providers may be cautious to interpret pulse oximetry readings in centrally-cannulated VA-ECMO patient due to SpO₂ grossly inaccurately predicting SaO₂ in this high-risk population.

Hemolysis (VV-ECMO)

Although some studies have examined the SpO₂-SaO₂ discrepancy and proposed that this difference may serve as an indicator of hemolysis [21-23], our study is the first to investigate hemolysis and the SpO₂-SaO₂ discrepancy with granular LDH data and adjustment of this discrepancy with clinically-relevant covariates. Notably, the upper limit of agreement for LDH >1000 was 11.3%. This value is ~3% higher than that of the two lower LDH thresholds (LDH 0-500 and 500-100), further highlighting how hemolysis causes inaccurate oxygen saturation measurements and are thus clinically-important. Moreover, a discrepancy that is greater than 10% (11.3%) with LDH >1000 is particularly important as our SpO₂-SaO₂ >10% analysis showed an ~7 higher odds of mortality in these patients. Our bilirubin and hemoglobin analyses also support our hemolysis hypothesis. Our study was consistent with Nisar’s pertaining to SpO₂ overestimating SaO₂ in VV-ECMO patients. Other studies have also attributed the SpO₂-SaO₂ discrepancy to carboxyhemoglobin production [4, 24, 25]as hemolysis occurs during VV-ECMO due to mechanical stress of red blood cells flowing through the ECMO circuit. [26] Hemolysis causes increased carboxyhemoglobin levels and a left shift in the oxygen dissociation curve, resulting in less oxygen delivery and unloading to peripheral tissue. [27] As Nisar’s study had a small fraction of non-synchronized SpO₂-SaO₂ pairs and lower number of patients, our study’s findings are reassuring and solidify the hemolysis hypothesis. Additionally, we characterized and adjusted for other important ECMO-physiological factors such as vasopressor dose, temperature, pump speeds, and bilirubin, which are different from prior studies.

Furthermore, our adjusted LMM findings of single-lumen cannulated patients being at greater risk of SpO₂ overestimating SaO₂ disagree with Nisar’s findings that double-lumen-cannulated patients underwent greater hemolysis, though they did not compare the SpO₂-SaO₂ discrepancy directly between cannulation strategies. Our hypothesis that single-lumen-cannulated VV-ECMO patients experience greater hemolysis than double-lumen is based on the higher resistance of blood flow and greater overall flow needed for the single-lumen catheter to maintain adequate blood flow. Moreover, recirculation and hemolysis, common in single-lumen VV-ECMO, [28] have been shown to be accentuated in higher ECMO flow [23, 29, 30], contributing to greater SpO₂-SaO₂ discrepancy in single- vs. double-lumen cannula. This theory falls in line with our results, as we saw a greater discrepancy in single-lumen, particularly at higher pump speeds, and are clinically important as recirculation during VV-ECMO can reduce ECMO’s efficacy and its overall ability to oxygenate blood. [31]

Occult Hypoxemia and SpO₂-SaO₂ >10% vs. Mortality

Although there were approximately equal numbers of VA-and VV-ECMO patients with >10% SpO₂-SaO₂ discrepancy, VV-ECMO patients observed far greater associated mortality with this discrepancy, which merits further investigation. Additionally, VV-ECMO observed higher frequencies of occult hypoxemia, which is also associated with a greater mortality risk. [32, 33] A greater SpO₂-SaO₂ discrepancy may predispose patients to poor outcomes as undetected hypoxemia for a prolonged duration can lead to worse organ dysfunction. [33] Acute respiratory distress syndrome constituted most of our VV-ECMO patients (n=76, 86%); their higher mortality rate may also be explained by acute inflammatory processes and corresponding lung injury [34] as hypoxemia can exacerbate inflammation. [35]

Limitations

Although we adjusted for clinically-relevant covariates, our study is still retrospective and residual confounding variables cannot be accounted for. Though we suspect that carboxyhemoglobin may explain the SpO₂-SaO₂ discrepancy in VV-ECMO, our center does not measure this value. Additionally, ELSO guidelines recommend using plasma-free hemoglobin as a hemolysis marker, which was not part of our center’s standard clinical protocol. Instead, we used LDH, bilirubin, and hemoglobin.

Conclusions

Venovenous-ECMO patients are at higher risk for occult hypoxemia compared to venoarterial-ECMO. A higher vasopressor requirement and different cannulation strategies (central venoarterial-ECMO; single-lumen venovenous-ECMO) were significant factors for clinically significant SpO₂-SaO₂ discrepancy in both ECMO modalities. Hemolysis also significantly worsened the SpO₂-SaO₂ discrepancy in venovenous-ECMO. In patient with risk factors such as high vasopressor usage, central- or single-lumen-cannulation, hemolysis, older age, male sex, and Black race/ethnicity, it would be prudent to monitor SaO₂ closely even if SpO₂ measurements are within normal limits during early ECMO support.

Abbreviations

ABG

arterial blood gas

CI

confidence interval

ECMO

extracorporeal membrane oxygenation

LDH

lactate dehydrogenase

LMM

linear mixed-effects modeling

RPM

revolutions per minute

SaO₂

oxygen saturation measured by arterial blood gas

SpO₂

oxygen saturation measured by pulse oximetry

VDE

Vasopressor Dose Equivalence

VA-ECMO

venoarterial ECMO

VV-ECMO

venovenous ECMO