Abstract
Background:
Exposure to hyperoxia, a high arterial partial pressure of oxygen (PaO2), may be associated with worse outcomes in patients receiving extracorporeal membrane oxygenator (ECMO) support. We examined hyperoxia in the Extracorporeal Life Support Organization (ELSO) Registry among patients receiving venoarterial (VA) ECMO for cardiogenic shock (CS).
Methods:
We included ELSO Registry patients from 2010 to 2020 who received VA ECMO for CS, excluding extracorporeal CPR. Patients were grouped based on PaO2 after 24 hours of ECMO: normoxia (PaO2 60–150 mmHg), mild hyperoxia (PaO2 151–300 mmHg), and severe hyperoxia (PaO2 >300 mmHg). In-hospital mortality was evaluated using multivariable logistic regression.
Results:
Among 9959 patients, 3005 (30.2%) patients had mild hyperoxia and 1972 (19.8%) had severe hyperoxia. In-hospital mortality increased across groups: normoxia, 47.8%; mild hyperoxia, 55.6% (adjusted OR 1.37, 95% CI 1.23–1.53, p <0.001); severe hyperoxia, 65.4% (adjusted OR 2.20, 95% CI 1.92–2.52, p <0.001). A higher PaO2 was incrementally associated with increased in-hospital mortality (adjusted OR 1.14 per 50 mmHg higher, 95% CI 1.12–1.16, p <0.001). Patients with a higher PaO2 had increased in-hospital mortality in each subgroup and when stratified by ventilator settings, airway pressures, acid-base status, and other clinical variables. Higher PaO2 was the second strongest predictor of in-hospital mortality, after older age.
Conclusion:
Exposure to hyperoxia during VA ECMO support for CS is strongly associated with increased in-hospital mortality, independent from hemodynamic and ventilatory status. Until clinical trial data are available, we suggest targeting a normal PaO2 and avoiding hyperoxia in CS patients receiving VA ECMO.
Keywords: cardiogenic shock, extracorporeal membrane oxygenation, hyperoxia, oxygen, shock, oxygen therapy
Introduction
Venoarterial (VA) extracorporeal membrane oxygenation (ECMO) provides biventricular cardiac and pulmonary support for patients with refractory cardiogenic shock (CS).(1–3) By oxygenating venous blood and returning it to the arterial circulation, VA ECMO can restore systemic blood flow and tissue perfusion during critical CS.(1) The arterial partial pressure of oxygen (PaO2) during VA ECMO support is determined by both the blood returned to the left ventricle from the pulmonary system and the blood circulated to the arterial system by ECMO.(1) A variety of patient-specific physiological variables and support-specific iatrogenic factors, including mechanical ventilator and ECMO circuit settings, can influence the PaO2 for an individual during ECMO support.
Hyperoxia, defined as an increased (supraphysiologic) PaO2, can produce potentially harmful physiologic effects and has been associated with adverse outcomes across critical illness settings and acute cardiovascular conditions.(4–8) In a prior analysis from the Extracorporeal Life Support Organization (ELSO) Registry including patients who received different forms of ECMO for a variety of indications, Munshi and colleagues reported worse outcomes with hyperoxia in some ECMO patient subgroups, but there was not a significant increase in mortality associated with hyperoxia for patients who received VA ECMO.(9) By contrast, a recent study by Tonna, et al. found a strong association between a higher PaO2 and worse outcomes in cardiac arrest patients from the ELSO Registry receiving extracorporeal cardiopulmonary resuscitation (ECPR).(10) Published studies examining the association between hyperoxia and outcomes in patients receiving VA ECMO for CS are substantially limited by small sample size, single-center design, or inclusion of patients who received ECPR (a distinct population).(9–18) These studies have provided conflicting evidence, leading to uncertainty regarding whether it is necessary to meticulously adjust ventilator and ECMO circuit settings to avoid hyperoxia in patients receiving ECMO for CS.
Accordingly, we sought to examine a large contemporary cohort of patients with CS who were supported with VA ECMO in the ELSO Registry to determine the association between hyperoxia during VA ECMO support and in-hospital mortality. Additionally, we sought to determine whether shock severity, ventilator settings and lung mechanics influenced the relationship between PaO2 and outcomes.
Methods
The authors declare that all supporting data are available within the manuscript and its online supplementary files.
Population
This was a retrospective analysis of the prospective, multicenter, multinational ELSO Registry cohort using deidentified data performed after IRB approval. We analyzed de-identified ECMO patient-run data from the ELSO Registry to identify adult patient-runs from 2010 to 2020 who received VA ECMO for cardiac support. We included patients with a diagnosis of CS based on International Classification of Diseases (ICD)-9/10 codes (785.51 and R57.1, respectively) and excluded ECPR patient-runs. We only included patient-runs with available data on PaO2 after 24 hours on ECMO and pre-ECMO hemodynamic support.(3) We excluded patients with hypoxemia (defined as a PaO2 <60 mmHg after 24 hours on ECMO), who are known to have worse outcomes that could confound the comparison of patients with and without hyperoxia.(9)
Exposures of interest
The primary exposure of interest was the PaO2 after 24 hours on VA ECMO support, which was used to classify patients as normoxia (PaO2 60–150 millimeters of mercury [mmHg]), mild hyperoxia (PaO2 151–300 mmHg) or severe hyperoxia (PaO2 >300 mmHg).(9) Additional exposures of interest included the ventilator fraction of inspired oxygen (FiO2), peak and mean airway pressures, positive end-expiratory pressure, and arterial blood gas measurements after 24 hours of ECMO support. The PaO2/FiO2 (PF) ratio was calculated based on the ventilator FiO2 (treated as a fraction by convention), and the oxygenation index was calculated as mean airway pressure * FiO2 (treated as a percent by convention) / PaO2. Post-cardiotomy CS was defined as the presence of any Clinical Procedural Terminology (CPT) code for cardiac surgery or use of cardiopulmonary bypass (CPB) prior to ECMO. The Society for Cardiovascular Angiography and Interventions (SCAI) Shock Classification was assigned based on the number of vasopressors and temporary mechanical circulatory support (MCS) devices required at the time of ECMO initiation.(3)
Outcomes of interest
The primary outcome of interest was all-cause in-hospital mortality. The key secondary outcomes were 30-day in-hospital mortality (censored at hospital discharge), and native heart survival, defined as survival to hospital discharge free from ongoing ECMO support or heart/lung transplantation.(3)
Statistical analysis
Analyses were performed using BlueSky version 7.40 (BlueSky LLC, Chicago, IL). Continuous variables were summarized using the median and interquartile range (IQR), with groups compared using the Kruskal-Wallis rank-sum test. Categorical variables were summarized as number and percent, with groups compared using the Pearson chi square test. Correlations between continuous variables were assessed using Spearman’s rho coefficient.
Locally estimated scatterplot smoother (LOESS) curves were generated to demonstrate the association between continuous exposures with outcomes of interest. In-hospital mortality was then stratified by the PaO2 quartile and quartiles of other exposures of interest (including ventilator parameters and relevant physiological variables), to screen for potential interactions that might explain the association between PaO2 and outcomes.
Odds ratio (OR) and 95% confidence interval (CI) values for in-hospital mortality and native heart survival were estimated using logistic regression, before and after multivariable adjustment. In-hospital mortality up to 30 days (censored at hospital discharge) was evaluated using Kaplan-Meier analysis, with group compared using the log-rank test. Hazard ratio (HR) values for 30-day in-hospital mortality were estimated using Cox proportional-hazards analysis. Multivariable models were adjusted for age, sex, race, year, post-cardiotomy, pre-ECMO cardiac arrest, renal replacement therapy, number of vasopressors and MCS devices at the time of ECMO initiation, weight, heart or lung transplant (except for analysis of native heart survival), primary diagnosis of CS, and the 24-hour values of ECMO circuit flow, FiO2, pH, PaCO2 and bicarbonate.
Random forest models were built using these same variables (number of trees = 1000, number of variables per split = 4) for prediction of in-hospital mortality and native heart survival, and the relative importance of each variable for outcome prediction was evaluated using the mean decrease in Gini index (Gini impurity) values, with higher values denoting a more important variable for accurate prediction.
Results
Population
The dataset included 14037 adults who received VA ECMO for CS from 2010 to 2020, not including those who received ECPR (Figure S1). We excluded 635 patients who did not have VA ECMO as their primary ECMO modality and 1296 who did not have available data on pre-ECMO hemodynamic support. Data on PaO2 after 24 hours on ECMO were available for 10394 patients, and 435 were excluded due to hypoxemia. The remaining 9959 patients comprised the final study population. Included patients had a median age of 57.8 (IQR 46.6, 65.9) years and 3167 (31.9%) were females; 2901 (29.1%) had post-cardiotomy CS.
Prevalence of hyperoxia
The median PaO2 after 24 hours on ECMO was 150 (IQR 99, 265) mmHg. The distribution of PaO2 values is shown in Figure 1A. Using our study definitions, 4982 (50.0%) patients were classified as having normoxia, 3005 (30.2%) as having mild hyperoxia and 1972 (19.8%) as having severe hyperoxia. Patients with mild and severe hypoxemia differed from patients with normoxia in terms of demographics (including a greater proportion of female sex and non-White race), hemodynamic support, ventilator settings and physiological variables (Table 1). We observed inverse correlations between the PaO2 after 24 hours of ECMO support and both the ventilator FiO2 (rho = −0.045, p <0.001) and year of ECMO initiation (rho = −0.083, p <0.001; Figure S2). We observed a decrease in the prevalence of hyperoxia over time (from 75.0% in 2010 to 45.2% in 2020; Figure 1B).
Figure 1AB:
Histogram of PaO2 after 24 hours on ECMO (A) in the population (p <0.001 for comparison of hospital survivors versus in-hospital deaths) and prevalence of hyperoxia by year of ECMO initiation (B).
Abbreviations: ECMO, extracorporeal membrane oxygenation; PaO2, partial pressure of arterial oxygen (arterial oxygen tension)
Table 1:
Clinical variables according to hyperoxia group based on the PaO2 after 24 hours on ECMO. Data are displayed as median (interquartile range) and number (percent). P value represents comparison of the normoxia (PaO2 60–150 mmHg), mild hyperoxia (PaO2 150–300 mmHg), and severe hyperoxia (PaO2 >300 mmHg) groups using the Kruskal-Wallis rank-rum and Pearson ch-squared tests.
| Normoxia (N=4982) | Mild hyperoxia (N=3005) | Severe hyperoxia (N=1972) | Total (N=9959) | p value | |
|---|---|---|---|---|---|
| Demographics and baseline characteristics | |||||
| Age (years) | 57.8 (46.9, 65.8) | 57.6 (45.7, 65.9) | 58.1 (46.9, 66.3) | 57.8 (46.6, 65.9) | 0.41 |
| Female | 1439 (29.0%) | 1040 (34.7%) | 688 (34.9%) | 3167 (31.9%) | < 0.001 |
| Race/Ethnicity | < 0.001 | ||||
| White | 3174 (66.2%) | 1744 (60.6%) | 1131 (60.9%) | 6049 (63.4%) | |
| Black | 508 (10.6%) | 398 (13.8%) | 259 (13.9%) | 1165 (12.2%) | |
| Hispanic | 311 (6.5%) | 227 (7.9%) | 134 (7.2%) | 672 (7.0%) | |
| Asian | 418 (8.7%) | 263 (9.1%) | 158 (8.5%) | 839 (8.8%) | |
| Multiple/Other | 386 (8.0%) | 247 (8.6%) | 176 (9.5%) | 809 (8.5%) | |
| Year | < 0.001 | ||||
| 2010–2015 | 839 (16.8%) | 671 (22.3%) | 439 (22.3%) | 1949 (19.6%) | |
| 2016–2018 | 1906 (38.3%) | 1137 (37.8%) | 784 (39.8%) | 3827 (38.4%) | |
| 2019–2020 | 2237 (44.9%) | 1197 (39.8%) | 749 (38.0%) | 4183 (42.0%) | |
| Weight (kg) | 86 (73, 102) | 84 (70, 99) | 84 (70, 99) | 85 (71, 100) | < 0.001 |
| Primary diagnosis of CS | 3043 (61.1%) | 1905 (63.4%) | 1286 (65.2%) | 6234 (62.6%) | 0.003 |
| Postcardiotomy | 1196 (24.0%) | 1001 (33.3%) | 704 (35.7%) | 2901 (29.1%) | < 0.001 |
| Cardiac arrest prior to ECMO | 1721 (35.0%) | 994 (33.6%) | 692 (35.5%) | 3407 (34.7%) | 0.31 |
| Ventilator prior to ECMO | 3820 (92.7%) | 2204 (93.0%) | 1378 (92.6%) | 7402 (92.8%) | 0.89 |
| Intubation To ECMO (hours) | 9 (2, 23) | 8 (2, 21) | 8 (3, 20) | 8 (2, 22) | 0.36 |
| RRT during hospitalization | 517 (10.4%) | 276 (9.2%) | 167 (8.5%) | 960 (9.6%) | 0.03 |
| Hemodynamic support prior to ECMO initiation | |||||
| Vasoactives | 4276 (85.8%) | 2490 (82.9%) | 1580 (80.1%) | 8346 (83.8%) | < 0.001 |
| Vasopressors | 4016 (80.6%) | 2345 (78.0%) | 1490 (75.6%) | 7851 (78.8%) | < 0.001 |
| Inodilators | 2226 (44.7%) | 1293 (43.0%) | 745 (37.8%) | 4264 (42.8%) | < 0.001 |
| MCS | 2848 (57.2%) | 1874 (62.4%) | 1217 (61.7%) | 5939 (59.6%) | < 0.001 |
| SCAI Shock Stage | < 0.001 | ||||
| B | 338 (6.8%) | 178 (5.9%) | 145 (7.4%) | 661 (6.6%) | |
| C | 3190 (64.0%) | 1873 (62.3%) | 1178 (59.7%) | 6241 (62.7%) | |
| D | 1164 (23.4%) | 730 (24.3%) | 490 (24.8%) | 2384 (23.9%) | |
| E | 290 (5.8%) | 224 (7.5%) | 159 (8.1%) | 673 (6.8%) | |
| Laboratory studies after 24 hours on ECMO | |||||
| PaO2 (mmHg) | 99 (81, 121) | 206 (173, 250) | 400 (347, 463) | 150 (99, 265) | < 0.001 |
| SaO2 (%) | 97 (95, 99) | 99 (98, 100) | 100 (99, 100) | 99 (97, 100) | < 0.001 |
| PF ratio | 203.4 (141.1, 276.0) | 437.5 (316.1, 571.3) | 855.0 (564.2, 1102.6) | 315.0 (196.0, 530.0) | < 0.001 |
| pH | 7.42 (7.38, 7.47) | 7.42 (7.38, 7.47) | 7.41 (7.36, 7.46) | 7.42 (7.38, 7.47) | < 0.001 |
| PCO2 (mmHg) | 37.7 (33.5, 42.0) | 37.6 (33.3, 42.0) | 39.0 (35.0, 44.0) | 38.0 (34.0, 42.0) | < 0.001 |
| Bicarbonate (mEq/L) | 24.0 (22.0, 27.0) | 24.3 (21.8, 27.0) | 24.8 (22.0, 28.0) | 24.3 (22.0, 27.0) | 0.06 |
| Lactate (mmol/L) | 2.0 (1.3, 3.4) | 2.3 (1.4, 4.3) | 2.8 (1.7, 6.0) | 2.2 (1.4, 4.0) | < 0.001 |
| Ventilator parameters after 24 hours on ECMO | |||||
| Rate (per minute) | 12 (10, 16) | 12 (10, 15) | 12 (10, 16) | 12 (10, 16) | < 0.001 |
| FiO2 (%) | 50 (40, 60) | 40 (40, 60) | 40 (40, 70) | 45 (40, 60) | 0.006 |
| Peak airway pressure (cmH2O) | 23 (20, 26) | 23 (20, 27) | 24 (20, 29) | 23 (20, 27) | < 0.001 |
| PEEP (cmH2O) | 8 (5, 10) | 8 (5, 10) | 8 (5, 10) | 8 (5, 10) | < 0.001 |
| Mean airway pressure (cmH2O) | 12 (10, 14) | 12 (10, 14) | 12.0 (10, 15) | 12 (10, 14) | 0.006 |
| Oxygenation index | 5.8 (3.8, 9.1) | 2.6 (1.9, 3.9) | 1.4 (1.0, 2.2) | 3.6 (2.0, 6.2) | < 0.001 |
| ECMO Pump Flow (L/min) | 4.1 (3.4, 4.7) | 4.1 (3.5, 4.7) | 4.1 (3.5, 4.6) | 4.1 (3.5, 4.7) | 0.68 |
| Blood pressure after 24 hours on ECMO | |||||
| Systolic BP (mmHg) | 96 (85, 109) | 94 (82, 106) | 91 (78, 104) | 95 (82, 107) | < 0.001 |
| Diastolic BP (mmHg) | 63 (56, 71) | 65 (56, 73) | 65 (56, 74) | 64 (56, 72) | < 0.001 |
| Mean BP (mmHg) | 73 (67, 81) | 73 (67, 81) | 73 (66, 82) | 73 (67, 81) | 0.90 |
| Clinical outcomes | |||||
| LOS (days) | 20 (10, 37) | 18 (8, 35) | 15 (6, 32) | 19 (9, 35) | < 0.001 |
| Hours on ECMO | 135 (81, 214) | 125 (73, 205) | 125 (65, 216) | 130 (74, 212) | < 0.001 |
| Hours from ECMO to discharge or death | 244 (1, 602) | 179 (0, 536) | 5 (0, 440) | 191 (0, 555) | < 0.001 |
| Discharge Alive | 2490 (50.0%) | 1279 (42.6%) | 639 (32.4%) | 4408 (44.3%) | < 0.001 |
| Discharge home | 990 (34.5%) | 540 (35.3%) | 247 (30.6%) | 1777 (34.2%) | 0.06 |
| Transplant | 281 (5.6%) | 202 (6.7%) | 109 (5.5%) | 592 (5.9%) | 0.10 |
| In-hospital mortality | 2379 (47.8%) | 1671 (55.6%) | 1289 (65.4%) | 5339 (53.6%) | < 0.001 |
| Native heart survival | 2326 (46.7%) | 1168 (38.9%) | 586 (29.7%) | 4080 (41.0%) | < 0.001 |
Abbreviations: BP, blood pressure; CS, cardiogenic shock; ECMO, extracorporeal membrane oxygenation; FiO2, fraction of inspired oxygen; LOS, length of stay; MCS, mechanical circulatory support; PaCO2, partial pressure of arterial carbon dioxide (arterial carbon dioxide tension); PaO2, partial pressure of arterial oxygen (arterial oxygen tension); PEEP, positive end-expiratory pressure; RRT, renal replacement therapy; SaO2, arterial oxygen saturation; SCAI, Society for Cardiovascular Angiography and Interventions.
In-hospital mortality and PaO2
During the index hospitalization, 5339 (53.6%) patients died. A progressive and nearly linear increase in in-hospital mortality was observed with incrementally higher PaO2 (adjusted OR 1.14 per 50 mmHg higher, 95% CI 1.12–1.16, p <0.001; Figure 2). Compared to patients with normoxia (47.8%), in-hospital mortality was higher for those with mild hyperoxia (55.6%, adjusted OR 1.37, 95% CI 1.23–1.53, p <0.001) or severe hyperoxia (65.4%, adjusted OR 2.20, 95% CI 1.92–2.52, p <0.001). A higher ventilator FiO2 was weakly associated with increased adjusted in-hospital mortality (Table 2). A stepwise increase in in-hospital mortality was seen as a function of rising PaO2 quartile (adjusted OR 1.30 per each higher quartile, 95% CI 1.24–1.36, p <0.001; Figure S3); the risk of mortality was particularly high in the fourth PaO2 quartile (adjusted OR 2.27 versus first quartile, 95% CI 1.97–2.60, p <0.001).
Figure 2:
Locally estimated scatterplot smoother (LOESS) curve showing in-hospital mortality as a function of the PaO2 at 24 hours after ECMO initiation.
Abbreviations: ECMO, extracorporeal membrane oxygenation; PaO2, partial pressure of arterial oxygen (arterial oxygen tension)
Table 2:
Results of multivariable logistic regression for prediction of in-hospital mortality (left) and native heart survival (right), defined as discharge from the hospital alive without heart/lung transplant or ongoing ECMO support. Data are presented as odds ratio (OR) and 95% confidence interval (CI) values. All physiologic variables were recorded at 24 hours after ECMO initiation. Note that transplant was not included in the model for prediction of native heart survival because it was used to define this endpoint.
| In-hospital mortality | Native heart survival | |||
|---|---|---|---|---|
| Adjusted OR (95% CI) | P value | Adjusted OR (95% CI) | P value | |
| Age (per year) | 1.03 (1.03–1.03) | < 0.001 | 0.98 (0.97–0.98) | <0.001 |
| Female sex | 1.03 (0.93–1.15) | 0.56 | 1.03 (0.92–1.14) | 0.65 |
| White race | 0.90 (0.81–0.99) | 0.04 | 1.12 (1.01–1.24) | 0.03 |
| Weight (per kg) | 1.00 (1.00–1.01) | 0.001 | 1.00 (1.00–1.00) | 0.03 |
| Year | 0.95 (0.93–0.97) | < 0.001 | 1.02 (1.00–1.04) | 0.08 |
| Primary diagnosis of CS | 0.96 (0.87–1.06) | 0.40 | 1.07 (0.97–1.18) | 0.20 |
| Postcardiotomy | 1.18 (1.05–1.33) | 0.006 | 1.00 (0.89–1.13) | 0.99 |
| Cardiac arrest | 1.09 (0.98–1.20) | 0.10 | 0.98 (0.89–1.09) | 0.74 |
| Renal replacement therapy | 1.81 (1.53–2.16) | < 0.001 | 0.56 (0.46–0.66) | <0.001 |
| # of vasopressors prior to ECMO | 1.09 (1.04–1.14) | < 0.001 | 0.96 (0.92–1.00) | 0.08 |
| # of MCS devices prior to ECMO | 1.06 (0.98–1.15) | 0.13 | 0.89 (0.82–0.96) | 0.003 |
| PaO2 (per 50 mmHg) | 1.14 (1.12–1.16) | < 0.001 | 0.87 (0.85–0.89) | <0.001 |
| pH (per 0.1 unit) | 0.74 (0.69–0.80) | < 0.001 | 1.32 (1.22–1.43) | <0.001 |
| PaCO2 (per 1 mmHg) | 0.99 (0.99–1.00) | < 0.001 | 1.01 (1.01–1.01) | <0.001 |
| Bicarbonate (per 1 meq/L) | 0.98 (0.96–0.99) | < 0.001 | 1.02 (1.01–1.03) | 0.005 |
| Ventilator FiO2 (per 10%) | 1.02 (1.00–1.05) | 0.04 | 0.99 (0.96–1.01) | 0.20 |
| ECMO pump flow (per 1 L/min) | 1.11 (1.05–1.18) | < 0.001 | 0.90 (0.85–0.95) | <0.001 |
| Transplant | 0.69 (0.56–0.85) | < 0.001 | --- | --- |
Abbreviations: CS, cardiogenic shock; ECMO, extracorporeal membrane oxygenation; FiO2, fraction of inspired oxygen; MCS, mechanical circulatory support; PaCO2, partial pressure of arterial carbon dioxide (arterial carbon dioxide tension); PaO2, partial pressure of arterial oxygen (arterial oxygen tension).
The association between higher PaO2 and higher in-hospital mortality was consistent across all examined subgroups divided by age, sex, race, year, post-cardiotomy status, pre-ECMO cardiac arrest, and SCAI Shock Stage (Figures 3 and 4). None of these subgroups had a significant statistical interaction with PaO2 for prediction of in-hospital mortality (all p >0.05). A rising PaO2 was incrementally associated with higher in-hospital mortality in the normoxia (unadjusted OR 1.15, 95% CI 1.02–1.28, p = 0.02) and mild hyperoxia (unadjusted OR 1.19, 95% CI 1.10–1.30, p <0.001) groups, but not in the severe hyperoxia group (adjusted OR 0.98, 95% CI 0.93–1.03, p = 0.42).
Figure 3:
In-hospital mortality in patients with normoxia, mild hyperoxia, and severe hyperoxia in selected subgroups. All between groups differences were significant (p <0.05) in each subgroup.
Figure 4:
In-hospital mortality in patients with normoxia, mild hyperoxia, and severe hyperoxia according to SCAI Shock Stage at the time of ECMO initiation.
Abbreviations: PaO2, partial pressure of arterial oxygen (arterial oxygen tension); SCAI, Society for Cardiovascular Angiography and Interventions
In-hospital 30-day mortality and PaO2
A higher PaO2 was incrementally associated with increased 30-day in-hospital mortality (adjusted HR 1.10 per 50 mmHg higher, 95% CI 1.08–1.11; p <0.001). An incremental increase in 30-day mortality (Figure 5) was observed for the mild hyperoxia (adjusted HR 1.32, 95% CI 1.22–1.42; p <0.001) and severe hyperoxia (adjusted HR 1.78; 95% CI 1.63–1.94; p <0.001) groups, with a decrease in median in-hospital survival time across groups (41 versus 27 versus 19 days). A stepwise increase in 30-day mortality was seen as a function of rising PaO2 quartile (adjusted HR 1.22 per each higher quartile, 95% CI 1.18–1.25, p <0.001; Figure S4); the risk of mortality was particularly high in the fourth PaO2 quartile (adjusted HR 1.75 versus first quartile, 95% CI 1.59–1.93, p <0.001).
Figure 5:
Kaplan-Meier curves demonstrating 30-day in-hospital mortality between hyperoxia groups; p value is for log-rank test.
In-hospital mortality and other exposures
Higher in-hospital mortality was observed with higher PF ratio (Figure S5) and lower oxygenation index (Figure S6). In-hospital mortality increased with rising PaO2 quartile when patients were stratified by ventilator FiO2 (Figure 6A), ECMO flow (Figure 6B), peak inspiratory pressure (Figure S7), or mean airway pressure (Figure S8). The highest mortality was observed in patients with a high PaO2 coupled with high values of each of these variables. Higher in-hospital mortality was also observed in patients with an increasing PaO2 when stratified by pH, lactate, PaCO2, bicarbonate, or PEEP (data not shown).
Figure 6AB:
In-hospital mortality in patients as a function of PaO2 quartile versus quartile of ventilator FiO2 (A) and ECMO flow (B) after 24 hours on ECMO.
Abbreviations: ECMO, extracorporeal membrane oxygenation; FiO2, fraction of inspired oxygen; PaO2, partial pressure of arterial oxygen (arterial oxygen tension)
Native heart survival and PaO2
A total of 4080 patients (88.3% of hospital survivors and 41.0% of all patients) were discharged alive with native heart survival. A progressive decrease in native heart survival was observed with an increase in PaO2 (adjusted OR 0.87 per 50 mmHg higher, 95% CI 0.85–0.89, p <0.001; Figure S9). Compared to patients with normoxia (46.7%), native heart survival was lower for those with mild hyperoxia (38.9%, adjusted OR 0.71, 95% CI 0.64–0.80, p <0.001) or severe hyperoxia (29.7%, adjusted OR 0.44, 95% CI 0.39–0.51, p <0.001). A stepwise decrease in native heart survival was seen as a function of rising PaO2 quartile (adjusted OR 0.76 per each higher quartile, 95% CI 0.73–0.80, p <0.001; Figure S10); native heart survival was particularly low in the fourth PaO2 quartile (adjusted OR versus first quartile 0.42, 95% CI 0.37–0.48, p <0.001).
Random forest models
Based on the highest mean decrease in Gini index values from the random forest model (Table S2 and Figure S11), the most important variables for prediction of in-hospital mortality were (in descending order of importance) age, PaO2, weight, ECMO pump flow, bicarbonate, PaCO2 and pH; the most important variables for prediction of native heart survival were (in descending order of importance) age, PaO2, ECMO pump flow, weight, bicarbonate, PaCO2 and pH.
Discussion
Summary of findings
In this analysis of nearly 10,000 adults from the ELSO Registry with CS from diverse etiologies (including both medical and post-cardiotomy CS) who received VA ECMO for cardiac support, we observed a strong and continuous dose-response relationship between exposure to a higher PaO2 after 24 hours of ECMO and higher in-hospital mortality. This association persisted after adjustment for ECMO flows, the FiO2 delivered by the ventilator, lung mechanics, shock severity, clinical variables, physiologic markers of ventilation, and acid-base status. The hazard associated with hyperoxia appeared to be independent from that associated with ventilator settings, airway pressures, or ECMO circuit flow, suggesting that this is a physiologic effect not directly related to pulmonary function. Indeed, PaO2 was the second most important variable for prediction of adverse outcomes after age. These data suggest that the FiO2 on the ventilator and especially the ECMO circuit should be adjusted to avoid hyperoxia in this population. Our data did not show a clear inflection point in the relationship between higher PaO2 and adverse outcomes, arguing in favor of maintaining a normal PaO2 for most CS patients during VA ECMO support.
Comparison with prior ELSO registry studies
Our analysis must be contrasted with the prior ELSO registry report by Munshi, et al. that included patients from 2010 to 2015 (n = 775 VA ECMO patients), who accounted for only a minority of patients included in our analysis due to the rapid expansion of ECMO use over time.(9) Notably, the adjusted odds ratio for severe hyperoxia (PaO2 >300 mmHg) in the prior analysis was 1.49, but the confidence interval was wide and crossed 1 suggesting an underpowered analysis.(9) The number of included patients is substantially higher in our study, allowing us greater statistical power to examine the relationship between hyperoxia and outcomes. We observed a stronger association between severe hyperoxia and mortality than this prior analysis, even after adjustment. Although overall mortality and the prevalence of hyperoxia were higher in the 2010–2015 period, we observed the same incremental association between hyperoxia and adverse outcomes across time epochs without a statistical interaction between PaO2 and year. We can speculate that increasing awareness of the potential harms associated with exposure to hyperoxia could have resulted in changes in practice that led to a lower prevalence of hyperoxia over time, in a manner that could correlate with other improvements in ECMO management and overall care resulting in better outcomes during recent years. Our analysis did not include patients who received venovenous (VV) ECMO or ECPR, two groups in which significant harm associated with hyperoxia was reported by this prior analysis.(9) Inclusion of a larger, more contemporary, and less heterogeneous cohort allowed us to effectively demonstrate the association between hyperoxia and adverse outcomes in VA ECMO recipients. Our greater sample size allowed us to perform a variety of exploratory analyses to better understand the modifying factors associated with hyperoxia, including ventilator settings and lung mechanics.
Comparison with other studies
Few studies have examined the association between hyperoxia and outcomes in VA ECMO recipients who did not receive ECPR. A single-center analysis by Al-Kawaz, et al. found an incremental dose-response relationship between the duration and severity of hyperoxia with poor neurologic outcomes in VA ECMO recipients, although one-third had received ECPR.(11) A small single-center study by Ross, et al. did not observe an association between severe hyperoxia (>=300 mmHg) and mortality in patients who received VA ECMO for CS due to acute myocardial infarction.(12) Studies of infants and children treated with VA ECMO have demonstrated an association between hyperoxia and poor outcomes.(13,14) Studies examining hyperoxia in patients receiving ECPR may not apply to the overall CS population, as the potential for anoxic brain injury in cardiac arrest patients could modify the association between hyperoxia and outcomes (particularly if the post-anoxic brain is more susceptible to hyperoxia, as has been proposed).(6,9,10,15,17–20) Nonetheless, our results closely mirror those of Tonna, et al. that showed a strong and incremental association between rising PaO2 and higher in-hospital mortality for patients with cardiac arrest receiving ECPR.(10) Notably, we observed similar associations between hyperoxia and adverse outcomes in our CS patients with and without a prior cardiac arrest.
Pathophysiologic rationale
By pumping fully oxygenated blood directly into the arterial system, VA ECMO is highly prone to causing hyperoxia which is often severe, as was observed in 20% of our patients.(12) Hyperoxia is a modifiable iatrogenic exposure that might be both a marker of sicker patients and a direct contributor to harm. The patients with hyperoxia in our study had some markers suggesting greater illness severity (including higher lactate levels) despite less vasoactive drug use, which could imply that a higher PaO2 was maintained to improve systemic oxygen delivery (DO2) in the setting of poor hemodynamics or impaired perfusion (i.e., inverse causation). However, hyperoxia is known to have numerous potentially deleterious physiological consequences, including adverse effects on the cardiovascular system and organ perfusion that could explain worse outcomes.(6–8,21,22) This suggests that using a higher PaO2 to increase DO2 may be a poor therapeutic option, particularly as dissolved free oxygen typically contributes only minimally to the oxygen content of arterial blood which is determined primarily by oxygen bound to hemoglobin.(6,8) More importantly, exposure to high levels of free oxygen during hyperoxia may increase oxygen free radical production and compromise beneficial hypoxia-dependent molecular mechanisms, resulting in a multitude of damaging cellular effects on the end organs.(8,22) Our observation that hyperoxia is associated with worse outcomes regardless of the FiO2 suggests that there is likely to be a harmful systemic effect of hyperoxia, as opposed to an effect mediated through oxygen-induced lung injury.(23) The hyperoxia in our patients was probably caused by high FiO2 on the ECMO circuit, as only a minority had a high ventilator FiO2 and the ventilator FiO2 was inversely associated with PaO2. This raises questions about the relevance of using the ventilator FiO2 for calculating the PF ratio and oxygenation index in patients receiving VA ECMO, despite evidence of associations with outcomes in our study. Notably, we observed possible evidence of sex- and race-based disparities, as females and individuals of non-White race were more often exposed to hyperoxia; while we did not identify statistical interactions between sex or race and the harmful effects of hyperoxia, White race was associated with a lower risk of adverse outcomes after adjustment.
Implications for clinical practice
Evidence has accumulated supporting the use of conservative oxygenation targets for many populations of critically ill patients and patients with acute cardiovascular disease.(6,20,24–26) Our analysis potentially expands the concept of conservative oxygen therapy to patients receiving VA ECMO support for CS. While our findings will need to be confirmed by an adequately powered randomized trial, we believe that titrating the FiO2 on both the ventilator and ECMO circuit to maintain the PaO2 below 150 mmHg is reasonable for patients with CS who are adequately supported on ECMO. Given the incrementally higher risk of adverse outcomes associated with higher PaO2 even among patients in the normoxemic group, it may be reasonable to target a lower PaO2 goal (e.g., 80 to 100 mmHg) in the absence of hypoperfusion for these patients. Our data do not provide insights into whether down-titrating the FiO2 on the ventilator versus ECMO circuit is more important for preventing hyperoxia, recognizing that high levels of FiO2 on the ventilator have been implicated in ventilator-associated lung injury and were associated with worse outcomes in our analysis additive to hyperoxia.(23) However, insofar as we believe that the hyperoxia in our patients was primarily caused by high FiO2 on the ECMO circuit, interventions to limit the FiO2 on the ECMO circuit may be more effective for avoiding hyperoxia. The safety of alternative approaches to increasing DO2 to reverse hypoperfusion must be evaluated in comparison to increasing the PaO2, which our data suggest may be harmful.
Limitations
As a retrospective cohort study, our analysis cannot demonstrate causal relationships and we cannot prove that hyperoxia was directly linked with harm. Unmeasured confounders could have influenced our findings, including the possibility that hyperoxia was a marker for other differences in overall care at the institution or provider level which we lacked data to fully evaluate. We did not have data on medical history, concomitant medical therapy, candidacy for advanced heart failure therapies, or ECMO complications. Detailed data regarding ECMO circuit settings were not available, precluding us from evaluating the association between ECMO circuit FiO2 and sweep flow on hyperoxia and outcomes. Hemoglobin and venous oxygen saturation values were not available to calculate the DO2 or oxygen extraction, and anemia could be a reason why a higher PaO2 was allowed by providers. However, we did not observe a difference in ECMO pump flow according to PaO2 group, and hyperoxia was associated with worse outcomes regardless of ECMO pump flow. Data were limited to a single assessment after 24 hours on ECMO, which prevented us from evaluating pre-ECMO exposure to hyperoxia, oxygenation trends, or overall exposure to hyperoxia over time; furthermore, we cannot comment on outcomes among patients who did not survive long enough to have PaO2 recorded after 24 hours on ECMO.(11) We could not determine the specific site where blood was obtained for arterial blood gas analysis, preventing us from identifying the presence of differential hypoxemia (“harlequin syndrome”); excluding patients with hypoxemia likely reduced the prevalence of this condition in our population.(27) Because all patients in this cohort received VA ECMO, we cannot determine whether our findings apply to CS patients receiving medical therapy or other forms of temporary MCS, nor can we necessarily expand our results to patients receiving venovenous (VV) ECMO or ECPR. Although hyperoxia was associated with adverse outcomes in our patients with a pre-ECMO cardiac arrest, we excluded ECPR patients and cannot determine whether our findings can be extrapolated to this population.(9,16–18,20) Our survival analyses should be considered exploratory, as we only included in-hospital events and would have missed deaths occurring after hospital discharge. Finally, although we included patients from 2020, we could not determine which of the included patients definitely had COVID, limiting our ability to extrapolate to this population.
Conclusion
Hyperoxia is incrementally associated with higher rates of in-hospital mortality and lower rates of native heart survival in patients with CS who received VA ECMO. Until randomized clinical trial data are available, we believe that it is reasonable to follow the emerging trend in critical care toward more conservative oxygen therapy targets for patients with CS who are adequately supported hemodynamically with VA ECMO.
Supplementary Material
Summary:
What is new?
This analysis shows a strong and incremental association between higher levels of arterial oxygen (PaO2) and worse in-hospital outcomes, including mortality or native heart survival, in 10,000 patients with cardiogenic shock supported with venoarterial extracorporeal membrane oxygenation. This association was consistent across subgroups and persisted after adjustment for known predictors of outcomes in this population, including ventilator settings.
What are the clinical implications?
Providers caring for patients with cardiogenic shock supported with venoarterial extracorporeal membrane oxygenation should adjust the ventilator and circuit settings to avoid excessive levels of arterial oxygen (defined as a PaO2 >150 mmHg, especially >300 mmHg). Maintaining normal levels of arterial oxygen should be considered in this population.
Funding:
No extramural funding was involved in the conduct of this research.
Abbreviations:
- CI
confidence interval
- CS
cardiogenic shock
- ECMO
extracorporeal membrane oxygenation
- ECPR
extracorporeal cardiopulmonary resuscitation
- ELSO
Extracorporeal Life Support Organization
- FiO2
fraction of inspired oxygen
- HR
hazard ratio
- LOESS
locally estimated scatterplot smoother
- MCS
mechanical circulatory support
- OR
odds ratio
- PaCO2
partial pressure of arterial carbon dioxide (arterial carbon dioxide tension)
- PaO2
partial pressure of arterial oxygen (arterial oxygen tension)
- PF ratio
ratio of PaO2 to FiO2
- SCAI
Society for Cardiovascular Angiography and Interventions
- VA
venoarterial
Footnotes
Disclosures: The authors have no relevant financial disclosures or conflicts of interest.
References
- 1.Guglin M, Zucker MJ, Bazan VM et al. Venoarterial ECMO for Adults: JACC Scientific Expert Panel. J Am Coll Cardiol 2019;73:698–716. [DOI] [PubMed] [Google Scholar]
- 2.Henry TD, Tomey MI, Tamis-Holland JE et al. Invasive Management of Acute Myocardial Infarction Complicated by Cardiogenic Shock: A Scientific Statement From the American Heart Association. Circulation 2021;143:e815–e829. [DOI] [PubMed] [Google Scholar]
- 3.Jentzer JC, Baran DA, Bohman JK et al. Cardiogenic shock severity and mortality in patients receiving venoarterial extracorporeal membrane oxygenator support. Eur Heart J Acute Cardiovasc Care 2022;11:891–903. [DOI] [PubMed] [Google Scholar]
- 4.Helmerhorst HJ, Roos-Blom MJ, van Westerloo DJ, de Jonge E. Association Between Arterial Hyperoxia and Outcome in Subsets of Critical Illness: A Systematic Review, Meta-Analysis, and Meta-Regression of Cohort Studies. Crit Care Med 2015;43:1508–19. [DOI] [PubMed] [Google Scholar]
- 5.Ni YN, Wang YM, Liang BM, Liang ZA. The effect of hyperoxia on mortality in critically ill patients: a systematic review and meta analysis. BMC Pulm Med 2019;19:53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Thomas A, van Diepen S, Beekman R et al. Oxygen Supplementation and Hyperoxia in Critically Ill Cardiac Patients: From Pathophysiology to Clinical Practice. JACC Adv 2022;1:100065. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Smit B, Smulders YM, van der Wouden JC, Oudemans-van Straaten HM, Spoelstra-de Man AME. Hemodynamic effects of acute hyperoxia: systematic review and meta-analysis. Crit Care 2018;22:45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Helmerhorst HJ, Schultz MJ, van der Voort PH, de Jonge E, van Westerloo DJ. Bench-to-bedside review: the effects of hyperoxia during critical illness. Crit Care 2015;19:284. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Munshi L, Kiss A, Cypel M, Keshavjee S, Ferguson ND, Fan E. Oxygen Thresholds and Mortality During Extracorporeal Life Support in Adult Patients. Crit Care Med 2017;45:1997–2005. [DOI] [PubMed] [Google Scholar]
- 10.Tonna JE, Selzman CH, Bartos JA et al. The association of modifiable mechanical ventilation settings, blood gas changes and survival on extracorporeal membrane oxygenation for cardiac arrest. Resuscitation 2022;174:53–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Al-Kawaz MN, Canner J, Caturegli G et al. Duration of Hyperoxia and Neurologic Outcomes in Patients Undergoing Extracorporeal Membrane Oxygenation. Crit Care Med 2021;49:e968–e977. [DOI] [PubMed] [Google Scholar]
- 12.Ross P, Miller C, Sheldrake J, McGuiness W, Udy A, Burrell A. Hyperoxia in patients with cardiogenic shock after myocardial infarction supported with venoarterial extracorporeal membrane oxygenation. Aust Crit Care 2021;34:55–59. [DOI] [PubMed] [Google Scholar]
- 13.Sznycer-Taub NR, Lowery R, Yu S, Owens ST, Hirsch-Romano JC, Owens GE. Hyperoxia Is Associated With Poor Outcomes in Pediatric Cardiac Patients Supported on Venoarterial Extracorporeal Membrane Oxygenation. Pediatr Crit Care Med 2016;17:350–8. [DOI] [PubMed] [Google Scholar]
- 14.Cashen K, Reeder R, Dalton HJ et al. Hyperoxia and Hypocapnia During Pediatric Extracorporeal Membrane Oxygenation: Associations With Complications, Mortality, and Functional Status Among Survivors. Pediatr Crit Care Med 2018;19:245–253. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Chang WT, Wang CH, Lai CH et al. Optimal Arterial Blood Oxygen Tension in the Early Postresuscitation Phase of Extracorporeal Cardiopulmonary Resuscitation: A 15-Year Retrospective Observational Study. Crit Care Med 2019;47:1549–1556. [DOI] [PubMed] [Google Scholar]
- 16.Kashiura M, Yasuda H, Kishihara Y et al. Association between short-term neurological outcomes and extreme hyperoxia in patients with out-of-hospital cardiac arrest who underwent extracorporeal cardiopulmonary resuscitation: a retrospective observational study from a multicenter registry. BMC Cardiovasc Disord 2022;22:163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Kobayashi M, Kashiura M, Yasuda H, Sugiyama K, Hamabe Y, Moriya T. Hyperoxia Is Not Associated With 30-day Survival in Out-of-Hospital Cardiac Arrest Patients Who Undergo Extracorporeal Cardiopulmonary Resuscitation. Front Med (Lausanne) 2022;9:867602. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Bonnemain J, Rusca M, Ltaief Z et al. Hyperoxia during extracorporeal cardiopulmonary resuscitation for refractory cardiac arrest is associated with severe circulatory failure and increased mortality. BMC Cardiovasc Disord 2021;21:542. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Patel JK, Kataya A, Parikh PB. Association between intra- and post-arrest hyperoxia on mortality in adults with cardiac arrest: A systematic review and meta-analysis. Resuscitation 2018;127:83–88. [DOI] [PubMed] [Google Scholar]
- 20.Young PJ, Bailey M, Bellomo R et al. Conservative or liberal oxygen therapy in adults after cardiac arrest: An individual-level patient data meta-analysis of randomised controlled trials. Resuscitation 2020;157:15–22. [DOI] [PubMed] [Google Scholar]
- 21.Grecu L Extracorporeal Membrane Oxygenation and Hyperoxia: To Blend or Not to Blend? Crit Care Med 2019;47:1660–1662. [DOI] [PubMed] [Google Scholar]
- 22.Cornet AD, Kooter AJ, Peters MJ, Smulders YM. The potential harm of oxygen therapy in medical emergencies. Crit Care 2013;17:313. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Kallet RH, Matthay MA. Hyperoxic acute lung injury. Respir Care 2013;58:123–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Barbateskovic M, Schjorring OL, Krauss SR et al. Higher vs Lower Oxygenation Strategies in Acutely Ill Adults: A Systematic Review With Meta-Analysis and Trial Sequential Analysis. Chest 2021;159:154–173. [DOI] [PubMed] [Google Scholar]
- 25.Chu DK, Kim LH, Young PJ et al. Mortality and morbidity in acutely ill adults treated with liberal versus conservative oxygen therapy (IOTA): a systematic review and meta-analysis. Lancet 2018;391:1693–1705. [DOI] [PubMed] [Google Scholar]
- 26.Liu L, Tian Y. Liberal or conservative oxygen therapy for ventilated patients in the ICU: a meta-analysis of randomized controlled trials. J Cardiothorac Surg 2021;16:261. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Falk L, Sallisalmi M, Lindholm JA et al. Differential hypoxemia during venoarterial extracorporeal membrane oxygenation. Perfusion 2019;34:22–29. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.