Mortality in Patients with Obesity and Acute Respiratory Distress Syndrome Receiving Extracorporeal Membrane Oxygenation: The Multicenter ECMObesity Study

Abstract

Rationale

Patients with obesity are at increased risk for developing acute respiratory distress syndrome (ARDS). Some centers consider obesity a relative contraindication to receiving extracorporeal membrane oxygenation (ECMO) support, despite growing implementation of ECMO for ARDS in the general population.

Objectives

To investigate the association between obesity and mortality in patients with ARDS receiving ECMO.

Methods

In this large, international, multicenter, retrospective cohort study, we evaluated the association of obesity, defined as body mass index ⩾ 30 kg/m², with ICU mortality in patients receiving ECMO for ARDS by performing adjusted multivariable logistic regression and propensity score matching.

Measurements and Main Results

Of 790 patients with ARDS receiving ECMO in our study, 320 had obesity. Of those, 24.1% died in the ICU, compared with 35.3% of patients without obesity (P < 0.001). In adjusted models, obesity was associated with lower ICU mortality (odds ratio, 0.63 [95% confidence interval, 0.43–0.93]; P = 0.018). Examined as a continuous variable, higher body mass index was associated with decreased ICU mortality in multivariable regression (odds ratio, 0.97 [95% confidence interval, 0.95–1.00]; P = 0.023). In propensity score matching of 199 patients with obesity to 199 patients without, patients with obesity had a lower probability of ICU death than those without (22.6% vs. 35.2%; P = 0.007).

Conclusions

Among patients receiving ECMO for ARDS, those with obesity had lower ICU mortality than patients without obesity in multivariable and propensity score matching analyses. Our findings support the notion that obesity should not be considered a general contraindication to ECMO.

At a Glance Commentary

Scientific Knowledge on the Subject

Few publications address the association between obesity and mortality in patients with acute respiratory distress syndrome (ARDS) supported with extracorporeal membrane oxygenation (ECMO). All prior studies have been single center, except for one registry-based study, each with inherent limitations. In the past three years, most publications related to ECMO use have focused predominantly on coronavirus disease (COVID-19)–related ARDS, leaving little evidence to support practice for non–COVID-19–related etiologies of ARDS.

What This Study Adds to the Field

This large, international, multicenter, retrospective cohort study examines data preceding the pandemic and is one of the first to present robust evidence that patients with obesity have lower ICU mortality than patients without in multivariable analysis corroborated by propensity score matching. Given the increasing prevalence of obesity and the fact that some centers consider obesity a contraindication to ECMO, our findings aid clinical decision making with respect to appropriate selection and management of patients with obesity who have ARDS and may benefit from ECMO. Our results support the hypothesis that obesity should not be considered a general contraindication to ECMO and advocate for the inclusion of patients with obesity in future clinical trials, where patients with higher degrees of obesity have been previously excluded from study populations.

The obesity epidemic remains a major driver of mortality worldwide, with the prevalence of obesity nearly tripling in the past half century (1, 2). As the proportion of patients with obesity admitted to ICUs increases (3, 4), our understanding of the impact of obesity on ICU outcomes continues to evolve. Whereas prior literature supports the concept of an “obesity paradox,” with better outcomes in patients with obesity compared with those without (5, 6), recent studies of critically ill patients, specifically with coronavirus disease (COVID-19), demonstrate no survival advantage with obesity in ICU patients compared with patients without obesity and suggest an age-dependent association (7–9). Notably, patients with obesity are at an increased risk for developing acute respiratory distress syndrome (ARDS) (10, 11) but are reported to have lower ARDS-specific mortality (12, 13).

The implementation of extracorporeal membrane oxygenation (ECMO) for patients with severe forms of ARDS has increased substantially over the past decade (14, 15), especially in light of growing evidence to support the use of ECMO in this setting (16–19). However, since the original version of the Extracorporeal Life Support Organization (ELSO) guidelines published in 2009, “size of patient” has been traditionally considered a relative contraindication to ECMO because of difficulties with cannulation, transport, and early mobilization or out of concern for poorer outcomes (4, 20–22). In addition, patients with higher body mass index (BMI) were excluded from high-quality trials examining the use of ECMO in ARDS (16), further highlighting the lack of familiarity with ECMO in this patient population and widening the data gap in supporting evidence-based care for such patients. The claim that obesity is a contraindication to ECMO has been challenged in recent years (23–26). Until recently, most of the evidence has been based on single-center, observational studies that included patients receiving extracorporeal life support for indications other than ARDS. One recent publication used the ELSO registry to evaluate the association of obesity on in-hospital mortality and reported lower mortality risk in higher classes of obesity (27). Nevertheless, confirmation of these results derived from more granular data with more precise effect estimates regarding the impact of obesity on outcomes in patients receiving ECMO for ARDS is still lacking, and patients with obesity, representing a substantial share of the population, may yet have ECMO withheld from them unnecessarily. We performed a large, international, multicenter, retrospective cohort study to investigate the association between obesity and mortality in patients with ARDS supported with ECMO. Some of the results of this study were previously reported in the form of an abstract (28).

Methods

Additional details on the methods are provided in the online supplement.

Data Sources

Data were collected from Columbia University Medical Center (New York), Alfred Hospital (Melbourne, Australia), St. Vincent’s Hospital (Sydney, Australia), Sorbonne University Assistance Publique–Hôpitaux de Paris (Paris, France), a previously published study (29), the Duke University School of Medicine (Durham, North Carolina), and the University of Milan (Milan, Italy) between July 1, 2012, and June 30, 2017, and combined with the LIFEGARDS (Ventilation Management of Patients with Extracorporeal Membrane Oxygenation for Acute Respiratory Distress Syndrome) database that incorporates data from 23 ICUs across 10 countries (30), with respective institutional and ethics review board approvals obtained.

Study Participants

All adult patients supported with ECMO for ARDS for hypoxemic respiratory failure were included in the analysis. Peripartum and postpartum patients up to 3 months after delivery were excluded because of the inability to ascertain the contribution of pregnancy to BMI (31).

Outcomes

The primary outcome was ICU mortality in patients with obesity compared with those without, where obesity was defined as a BMI ⩾ 30 kg/m² in accordance with World Health Organization definitions (31, 32). The association of BMI as a continuous variable, per 1 kg/m² increase, with ICU mortality was examined using multivariable logistic regression. Secondary outcomes included ICU length of stay (LOS), hospital LOS, and number of days receiving ECMO and mechanical ventilation. Given the competing risk of death, mechanical ventilator–free, ECMO-free, and hospital-free days at 90 days in patients with obesity compared with those without were also calculated using previously described methods, considering zero free days for deceased patients (33).

Statistical Analysis

We followed the Strengthening the Reporting of Observational Studies in Epidemiology recommendations for reporting this study (see Figure E1 in the online supplement) (34). Continuous variables are expressed as mean (SD) or as median (interquartile range [IQR]) and categorical variables as counts and proportions. Normality of the data distribution was visually assessed using histograms. Proportions were compared using the χ² test or Fisher exact test. Continuous variables were compared using Student’s t test or the Wilcoxon rank sum test.

Using a directed acyclic graph (see Figure E2) (35), a priori–defined pre-ECMO covariates presumed to be associated with the presence of obesity or patient outcomes were included in a multivariable logistic regression analysis to determine association of BMI with ICU mortality. These covariates were used to determine the propensity score of obesity for each patient. In the matched cohort analysis, taking into account the clustered structure of our dataset, patients with obesity were matched with patients without obesity according to the propensity score, using a 1:1 nearest neighbor matching procedure without replacement and a caliper width of 0.2, resulting in a relatively narrow difference between matched variables (see Figure E3). The McNemar test was subsequently performed.

Given the previously demonstrated nonlinear association between BMI and mortality (36), to evaluate for potential nonlinear associations between BMI and risk of ICU mortality, we used generalized additive models (GAMs) with the “gam” function with a smoothing function for BMI in R. We compared the Akaike information criterion of both linear models and GAMs to choose the model with the best fit. StataCorp 15.0 and R version 4.1.0 (The R Foundation for Statistical Computing) were used for these analyses.

Results

Among the participating centers, 790 patients receiving ECMO support for ARDS were included in the study, 350 of whom were recruited from the LIFEGARDS database. Demographics and precannulation clinical characteristics for the overall cohort and stratified by obesity are presented in Table 1. The average age was 44.2 years (SD, 15.5 yr), without a significant difference between the groups. In patients with obesity, the majority (58.8%) were women. APACHE (Acute Physiology and Chronic Health Evaluation) II scores were lower in patients with obesity compared with those without (P = 0.028). The etiology of respiratory failure was available in 534 patients, with bacterial, viral, and aspiration pneumonias being the most common, constituting 33.3%, 26.4%, and 14.6%, respectively, without a significant difference between the groups. Before ECMO cannulation, 63% of patients received neuromuscular blocking agents (NMBAs), 30% received inhaled nitric oxide (iNO), and 19% were placed in the prone position, with no significant differences in the use of these adjunctive therapies between groups except in the use of NMBAs, for which 280 patients with obesity (59.6%) versus 219 patients without (68.4%) received NMBAs (P = 0.014). Precannulation ratios of Pa_O2 to Fi_O2 (Pa_O2:Fi_O2) were 63.9 mm Hg (SD, 28.6 mm Hg) and 71.5 mm Hg (SD, 31.1 mm Hg) for patients with and without obesity, respectively (P = 0.001). Median plateau airway pressures before cannulation were 32 (IQR, 30–36) cm H₂O for patients with obesity and 31 (IQR, 28–35) cm H₂O for those without (P = 0.001; see Figure E4). Precannulation pH did not significantly differ between the groups. Further stratifications of our population by BMI and geographic origin are depicted in Tables E1 and E2, respectively.

Table 1.

Demographic, Clinical, and Mechanical Ventilation Characteristics before Extracorporeal Membrane Oxygenation Cannulation, for the Overall Cohort and Stratified by Obesity

		BMI (kg/cm2)
	All Patients (n = 790)	Patients without Obesity (n = 470)	Patients with Obesity (n = 320)	P Value	SMD
Age, yr	44.2 ± 15.5	44.1 ± 16.8	44.4 ± 13.4	0.753	0.022
Sex				0.018	0.176
Female	286 (36.2)	154 (32.8)	132 (41.2)	—	—
Male	504 (63.8)	316 (67.2)	188 (58.8)	—	—
BMI, kg/cm2	30.3 ± 9.2	24.6 ± 3.3	38.8 ± 8.3	<0.001	2.255
BMI ⩾ 30 kg/cm2	320 (40.5)	0 (0.00)	320 (100)	<0.001	—
BMI ⩾ 35 kg/cm2	193 (24.4)	0 (0.00)	193 (60.3)	<0.001	—
BMI ⩾ 40 kg/cm2	110 (13.9)	0 (0.00)	110 (34.4)	<0.001	—
APACHE II score	23.3 ± 9.7 (n = 660)	23.9 ± 10.1 (n = 425)	22.2 ± 8.9 (n = 235)	0.028	0.175
SOFA score	10.0 ± 4.1 (n = 772)	10.0 ± 4.4 (n = 461)	9.9 ± 3.7 (n = 311)	0.795	0.019
Duration of IMV before ECMO, d	2.00 (1.00–6.00)	2.00 (1.00–6.00)	3.00 (1.00–6.00)	0.165	0.020
Ventilation parameters
FiO2, mm Hg	100 (100–100) (n = 762)	100 (100–100) (n = 447)	100 (100–100) (n = 315)	0.054	0.155
PEEP, cm H2O	14 (10–16) (n = 717)	12 (10–15) (n = 423)	15 (12–18) (n = 294)	<0.001	0.466
Vt, ml/kg PBW	6.33 ± 1.88 (n = 419)	6.13 ± 1.63 (n = 287)	6.79 ± 2.26 (n = 132)	0.003	0.335
Plateau pressure, cm H2O	32 [29–35] (n = 529)	31 [28–35] (n = 331)	32 [30–36] (n = 198)	0.001	0.328
Driving pressure, cm H2O*	18 [14–22] (n = 274)	18 [14–23] (n = 148)	18 [14–22] (n = 126)	0.951	0.006
Respiratory rate, breaths/min	25.0 ± 7.6 (n = 686)	24.5 ± 7.7 (n = 402)	25.8 ± 7.3 (n = 284)	0.023	0.176
Precannulation arterial blood gas values
pH	7.24 ± 0.14 (n = 735)	7.24 ± 0.14 (n = 433)	7.24 ± 0.14 (n = 302)	0.774	0.021
PaCO2, mm Hg	62.7 ± 26.7 (n = 628)	61.0 ± 23.6 (n = 402)	65.6 ± 31.2 (n = 226)	0.055	0.166
Worst PaO2:FiO2 ratio	68.4 ± 30.3 (n = 755)	71.5 ± 31.1 (n = 447)	63.9 ± 28.6 (n = 308)	0.001	0.257
Rescue therapy before ECMO
Neuromuscular blockade	499 (63.2)	280 (59.6)	219 (68.4)	0.014	0.185
Inhaled nitric oxide	234 (29.6)	141 (30.0)	93 (29.1)	0.838	0.021
Prone positioning	152 (19.2)	94 (20.0)	58 (18.1)	0.572	0.048
HFOV	36 (4.6)	21 (4.5)	15 (4.7)	1.000	0.010
RRT before ECMO	226 (28.6)	131 (27.9)	95 (29.7)	0.635	0.040

Definition of abbreviations: APACHE = Acute Physiology and Chronic Health Evaluation; BMI = body mass index; ECMO = extracorporeal membrane oxygenation; HFOV = high-frequency oscillatory ventilation; IMV = invasive mechanical ventilation; PBW = predicted body weight; PEEP = positive end-expiratory pressure; RRT = renal replacement therapy; SMD = standardized mean difference; SOFA = sequential organ failure assessment.

Data are expressed as mean ± SD, n (%), or median (interquartile range).

^*Plateau pressure minus PEEP.

Of 320 patients with obesity, 77 (24.1%) died before discharge from the ICU, compared with 166 of 470 (35.3%) patients without obesity (P < 0.001). In multivariable analysis, obesity was associated with a lower risk of death in the ICU (odds ratio, 0.63 [95% confidence interval, 0.43–0.93]; P = 0.018), adjusted for age, sex, days of invasive mechanical ventilation (IMV) before cannulation, precannulation positive end-expiratory pressure (PEEP), pH, Pa_O2:Fi_O2, use of NMBA, iNO, prone positioning, and renal replacement therapy (RRT) (Figure 1; Table 2). Examined as a continuous variable, per 1 kg/m², higher BMI was similarly associated with decreased ICU mortality in multivariable analysis (odds ratio, 0.97 [95% confidence interval, 0.95–1.00]; P = 0.023). Mortality stratified by BMI is depicted in Figure E5. Younger age, fewer days of IMV before ECMO, higher PEEP, higher pH, and not receiving iNO or RRT before ECMO were also associated with lower ICU mortality (Table 3). As APACHE II scores were not available for all patients, sensitivity analyses adjusting for APACHE II score in the multivariable logistic regression, examining the association of obesity and BMI as a continuous variable with ICU mortality, were performed on a smaller cohort (578 patients) and are shown in Tables E3 and E4, respectively. Sensitivity analyses adjusting for sequential organ failure assessment (SOFA) score are shown in Tables E5 and E6.

Figure 1.

Forest plot based on the results of multivariable analysis of the factors associated with ICU death in patients with obesity compared with those without obesity. Protective associations are shown in blue, harmful associations in red, and nonsignificant associations in black. CI = confidence interval; ECMO = extracorporeal membrane oxygenation; iNO = inhaled nitric oxide; NMBA = neuromuscular blocking agent; OR = odds ratio; PEEP = positive end-expiratory pressure; PF = Pa_O2:Fi_O2.

Table 2.

Variables Associated with ICU Death in Patients Supported by Extracorporeal Membrane Oxygenation in Multivariable Analysis Examining the Association of Obesity with Risk of Death in the ICU

	Odds Ratio for ICU Death (95% CI)	P Value
Obesity	0.63 (0.43–0.93)	0.018
Age, for 10-yr increase	1.37 (1.22–1.54)	<0.001
Sex, male	0.93 (0.65–1.35)	0.710
Days of IMV before ECMO, for 1 d	1.07 (1.03–1.11)	<0.001
PEEP before ECMO, for 1 cm H2O increase	0.95 (0.91–0.99)	0.018
pH before ECMO, for 0.01 increase	0.98 (0.96–0.99)	<0.001
PF ratio before ECMO, for 10 mm Hg increase	1.00 (0.94–1.06)	0.954
Use of inhaled nitric oxide before ECMO	1.70 (1.10–2.61)	0.017
Use of NMBA before ECMO	0.72 (0.48–1.06)	0.010
Use of prone positioning before ECMO	0.90 (0.55–1.43)	0.653
Need for dialysis before ECMO	1.58 (1.07–2.33)	0.020

Definition of abbreviations: CI = confidence interval; ECMO = extracorporeal membrane oxygenation; IMV = invasive mechanical ventilation; NMBA = neuromuscular blocking agents; PEEP = positive end-expiratory pressure; PF = Pa_O2:Fi_O2.

Table 3.

Variables Associated with ICU Death in Patients Supported by Extracorporeal Membrane Oxygenation in Multivariable Analysis Examining the Association of Body Mass Index as a Continuous Variable with Risk of Death in the ICU

	Odds Ratio for ICU Death (95% CI)	P Value
BMI, per 1 kg/m2 increase	0.97 (0.95–1.00)	0.023
Age, for 10-yr increase	1.36 (1.21–1.53)	<0.001
Sex, male	0.93 (0.64–1.34)	0.690
Days of IMV before ECMO, for 1 d	1.07 (1.04–1.12)	<0.001
PEEP before ECMO, for 1 cm H2O increase	0.96 (0.92–1.00)	0.035
pH before ECMO, for 0.01 increase	0.98 (0.97–0.99)	<0.001
PF ratio before ECMO, for 10 mm Hg increase	1.00 (0.94–1.06)	0.951
Use of inhaled nitric oxide before ECMO	1.75 (1.13–2.70)	0.011
Use of NMBA before ECMO	0.73 (0.49–1.08)	0.118
Use of prone positioning before ECMO	0.87 (0.54–1.40)	0.578
Need for dialysis before ECMO	1.62 (1.10–2.39)	0.015

Definition of abbreviations: BMI = body mass index; CI = confidence interval; ECMO = extracorporeal membrane oxygenation; IMV = invasive mechanical ventilation; NMBA = neuromuscular blocking agents; PEEP = positive end-expiratory pressure; PF = Pa_O2:Fi_O2.

In GAMs, greater BMI was associated with a decreased risk of death, though this association may plateau at greater values of BMI (P for linearity = 0.002; Figure 2). In comparing linear models and GAMs, values were similar across all models, indicating that our findings are robust in multiple sensitivity analyses. We chose to display the GAM to demonstrate the lack of certainty in estimates for those with high BMIs and the possibility of a plateau or change in the direction of association between BMI and survival at very high values of BMI.

Figure 2.

(A and B) Odds ratios of mortality by BMI, unadjusted (A) and adjusted for age, sex, days of invasive mechanical ventilation before cannulation, precannulation positive end-expiratory pressure, pH, Pa_O2:Fi_O2, use of neuromuscular blocking agents, inhaled nitric oxide, prone positioning, and renal replacement therapy (B). The continuous association of BMI and odds of ICU death may differ somewhat from the primary model, as it is based on a generalized additive model that smooths the line segments to allow curved transitions over the ranges of BMI values. The solid line represents the estimated odds ratio. The dashed lines represent 95% confidence bands. BMI = body mass index.

In propensity score matching analysis, 199 of 320 patients with obesity could be matched with 199 of the patients without obesity with good performance (see Table E7 and Figure E6) on age, sex, APACHE II score, respiratory rate, Vt, PEEP, plateau airway pressure, pH, Pa_CO2, Pa_O2:Fi_O2, use of high-frequency oscillatory ventilation, iNO, NMBA, prone positioning, and the receipt of RRT before cannulation (see Table E8). Demographic and clinical differences between patients with obesity, as well as the overall cohort, who could be matched versus those who went unmatched are depicted in Tables E9 and E10, respectively. Notably, matched patients in the overall cohort were sicker than patients who went unmatched, as evidenced by higher SOFA score, higher respiratory rate, higher plateau pressure, and lower Pa_O2:Fi_O2, despite higher use of NMBA before cannulation. In this analysis, once again, patients with obesity had a lower probability of death in the ICU (22.6% vs. 35.2%; P = 0.007; Figure 3).

Figure 3.

Kaplan-Meier graph for propensity score–matched populations. BMI = body mass index.

For the overall cohort, the median durations of ECMO and IMV were 10 (IQR, 7–18) days and 19 (IQR, 11–33) days, respectively (Table 4). The median LOSs in the ICU and the hospital were 24 (IQR, 13–39) days and 36 (IQR, 21–55) days, respectively. ICU and hospital LOSs, as well as duration of IMV and ECMO, did not differ significantly between patients with obesity and those without. There was no difference in ECMO-free days at 90 days between the groups. However, patients without obesity had fewer ventilator-free days and hospital-free days compared with patients with obesity, with medians of 50 (IQR, 0–75) days versus 60 (IQR, 0–75) days (P = 0.050) and 22 (IQR, 0–58 days) versus 40 (IQR, 0–58) days (P = 0.012), respectively. Furthermore, examining these variables in survivors, the median ECMO duration was 10 (IQR, 7–15) days compared with 14 (IQR, 6–25.5) days in nonsurvivors. Duration of IMV did not differ significantly between survivors versus nonsurvivors. In contrast, median ICU and hospital LOSs were significantly different, at 24 (IQR, 16–39) days and 41 (IQR, 26–60) days, respectively, in survivors, compared with 22 (IQR, 11–36.5) days and 28 (IQR, 13–44) days in nonsurvivors (Table 5; see Figure E7). Clinical outcomes among ICU survivors stratified by obesity are described in Table E11.

Table 4.

Clinical Outcomes for the Overall Cohort and Stratified by Obesity

	All Patients (n = 790)	Body Mass Index (kg/cm2)		P Value
		Patients without Obesity (n = 470)	Patients with Obesity (n = 320)
ECMO duration, d	10.0 (6.00–18.0)	10.0 (6.00–18.0)	11.0 (7.00–16.0)	0.450
Mechanical ventilation duration, d	19.0 (11.0–34.0)	17.0 (10.0–34.0)	20.0 (12.0–32.0)	0.534
Length of stay in ICU, d	24.0 (14.0–39.0)	24.0 (14.0–41.0)	24.0 (14.5–37.0)	0.850
Length of stay in hospital, d	36.0 (21.0–56.0)	35.0 (20.0–58.0)	37.0 (23.0–53.0)	0.754
ECMO-free days, at 90 d	76.0 (0.0–82.0)	74.0 (0.0–82.0)	77.0 (18.0–82.0)	0.098
Ventilator-free days, at 90 d	56.0 (0.0–75.0)	49.0 (0.0–75.0)	60.0 (0.0–75.0)	0.031
Hospital-free days, at 90 d	32.0 (0.0–58.0)	19.0 (0.0–58.0)	40.0 (0.0–58.0)	0.006

Definition of abbreviation: ECMO = extracorporeal membrane oxygenation.

Table 5.

Clinical Outcomes for ICU Survivors and Nonsurvivors

	Survivors (n = 547)	Nonsurvivors (n = 243)	P Value
ECMO duration, d	10.0 (7.00–15.0)	14.0 (6.00–25.5)	<0.001
Mechanical ventilation duration, d	18.0 (11.0–32.0)	21.0 (10.0–36.0)	0.471
Length of stay in ICU, d	24.0 (16.0–39.0)	22.0 (11.0–36.5)	0.001
Length of stay in hospital, d	41.0 (26.0–60.0)	28.0 (13.0–44.0)	<0.001

Definition of abbreviation: ECMO = extracorporeal membrane oxygenation.

Discussion

In this large, multicenter, international, observational, retrospective cohort study of patients receiving ECMO for ARDS, patients with obesity had lower ICU mortality compared with those without obesity. This finding persisted after adjustment for potential confounders and in sensitivity analyses using propensity score matching.

Our findings are consistent with prior single-center reports on the subject. Higher BMI was shown to be an independent predictor of 90-day survival in a 76-patient cohort of patients with COVID-19–associated ARDS receiving ECMO, of whom 29 had obesity (26). Comparably, 77 patients with obesity receiving ECMO for COVID-19–associated ARDS were found to have similar ICU and hospital mortality as patients without obesity (37). Last, in a study of 115 patients with BMI ⩾ 30 kg/m² receiving ECMO for hypoxemic or hypercapnic respiratory failure, an overall mortality rate of 24.7% was reported across all obesity classes, with no association of BMI with mortality (24).

The largest study to date to evaluate this concept using the ELSO registry examined the association of BMI ⩾ 35 kg/m² with in-hospital mortality in 18,529 patients, who were further subdivided into propensity score–matched groups of 4,707 patients with class II or greater obesity matched with 4,707 patients with BMI < 35 kg/m² (27). Similar to our work, although using a different cutoff of BMI of ⩾35 kg/m² to evaluate the impact of obesity, patients with higher BMI in the ELSO registry study had lower in-hospital mortality in propensity score matching, inverse propensity score–weighted, and multivariable analyses. Although our work confirms the central message of the ELSO registry study, there are notable distinctions between the two. As with many registry-based studies, though the sample size is appreciably larger, the quality and validity of the data in the ELSO registry are likely variable, with large percentages of missing data, such as one-quarter of patients missing pre-ECMO PEEP, for example. Our dataset, although smaller, is more granular and validated across multiple centers. Although we included patients receiving ECMO support for ARDS solely, the ELSO registry study based the query of ECMO indications on the International Classification of Diseases, Ninth and 10th Revisions, which inherently may carry inaccuracies, and included other etiologies of respiratory failures such as asthma, pulmonary hypertension, and bridge to lung transplantation. Our dataset is less informative than the ELSO registry data with respect to ECMO complications, discharge destination, and comorbid conditions, although our dataset includes APACHE II and SOFA scores to better characterize the latter. We have chosen to designate BMI ⩾ 30 kg/m² as the definition of obesity, in accordance with World Health Organization definitions, as opposed to a BMI of ⩾35 kg/m², which reflects class II obesity in the ELSO registry study, to be more precise in attributing the observed association to obesity. Furthermore, we focused more on the association of BMI per 1 kg/m² increase to delineate the continuous association with mortality. Interestingly, although our conclusions are similar with respect to the association of obesity with mortality, BMI cutoff differences aside, our GAM plots differ in that the ELSO registry study reported mortality to be highest among patients with BMIs of 30–35 kg/m², whereas we found a narrower confidence interval for that BMI range but not conclusively the highest mortality.

The notion that obesity is protective in surviving critical illness, otherwise known as the obesity paradox, has been previously described (6, 12, 38), with a number of suspected explanations for this phenomenon. The respiratory physiology of patients with obesity, of course, differs on average from that of other patients. In particular, increased chest wall elastance in patients with obesity has been proposed to offer a protective benefit of partially absorbing transpulmonary pressure, thereby decreasing ventilator-induced lung injury (39). In addition, higher plateau airway pressures generated by similar ventilator settings in patients with obesity may prompt lowering of ventilator volumes and pressures by clinicians, even though such pressures may reflect the combination of increased transpulmonary pressure and the increased chest wall elastance.

Although the aforementioned mechanisms may contribute to better survival in patients with obesity and ARDS, our findings specific to ECMO use in patients with obesity and the most severe forms of ARDS warrant further exploration. The difference in survival may be explained, in part, by the fact that patients with obesity are prone to increased pleural pressure, reduced FRC, and greater degrees of atelectasis relative to patients without obesity (40). In this setting, the severity of ARDS may have been overestimated in our patients with obesity relative to the patients without obesity, selecting for a population to be cannulated that, even with comparable Pa_O2:Fi_O2 ratios, may have had less severe pathologic forms of ARDS and a more readily reversible pathology, in particular, greater degrees of atelectasis, thus leading to collider bias. APACHE II scores were also notably lower in our obese group, suggesting the selection of a cohort with lower disease severity among the patients with obesity at the time of ECMO cannulation, thus raising the possibility of indication bias. In addition, despite a wide age distribution (see Figure E8), patients in our cohort were, on average, younger than a “typical” patient with ARDS, as previously reported in a large international cohort (with a mean age of 44 yr compared with a mean age of 61.5 yr), where the average BMI was notably 27.5 kg/m², thus describing a typical patient with ARDS as overweight (41). This age difference, likely explained by strict patient selection before ECMO cannulation, could be contributing to the observed beneficial outcome and may reflect bias by indication as well. In addition, a median PEEP of 15 cm H₂O in our group with obesity, even though significantly higher than in the group without obesity, suggests the potential underuse of PEEP in these patients, with the possibility that some proportion of patients with obesity may not have met indications for ECMO if managed differently. In that context, guiding optimal PEEP settings with bedside tools such as esophageal pressure (42) or electrical impedance tomography before ECMO consideration is desirable (43).

Our study has a number of strengths. First, we report on the largest cohort of patients with obesity receiving ECMO for ARDS, outside of the aforementioned ELSO registry study. Second, by virtue of being multicenter and international, our study is the first of its kind to examine mortality in patients with obesity receiving ECMO exclusively for ARDS across a wide range of settings, making our findings more generalizable. Last, we identified similar associations in propensity score matching analysis.

Our study has limitations. First, we anticipate that the study could be subject to selection bias, as ECMO was offered only to those who were deemed likely to benefit, whereby the selected patients with obesity could be relatively healthier compared with patients without obesity, although such a difference was not seen with the use of traditional scoring systems, with the exception of APACHE II.

Second, we lack data on patients who were deemed not candidates for ECMO support, for both patients with and without obesity. Third, although we report mechanical ventilation parameters, the use of NMBAs, and prone positioning, we do not have the means to assess differences in sedation and mobilization practices across the participating centers. Although practices within a given center should be similar for patients with and without obesity, nonetheless, the degree of obesity might affect the pharmacokinetics of sedatives or the ability to engage in physical rehabilitation, for instance. Notably, the rate of prone positioning before cannulation was lower in our combined cohort, compared with increased use seen in more recent years; increased use of prone positioning may select out a subset of patients with obesity who respond to this maneuver, thus obviating the need for ECMO (44). We used the direct acyclic graph to avoid collider bias but recognize that not all the variables that should ideally be included in the direct acyclic graph were available for analysis.

Fourth, our data do not include esophageal pressure measurements, which could be helpful in assessing the degree to which hypoxemia could be attributed to atelectasis in patients with obesity. Fifth, we lack information about our patients’ net fluid balance, which may be a contributor to hypoxemia in our population, and unfortunately, we lack surrogate measurements such as right and left ventricular function or diastolic dysfunction to approximate fluid status.

Last, we lack data on a comparator group of patients with obesity who received maximal mechanical ventilation in lieu of ECMO to ascertain the contribution of ECMO to the observed association, although such data would always be biased by the patients either not being selected for ECMO or being at a center that does not offer ECMO. In addition, as with all observational studies, our study has limitations intrinsic to the study design. Although we checked our data for inconsistencies and queried all investigators when outliers were detected, measurement error may remain unidentified in this large database. Similarly, unmeasured confounding may exist, as we lack data on, and thus have not been able to adjust for, all potential confounders.

Conclusions

Among patients receiving ECMO for ARDS, patients with obesity had lower ICU mortality than those without obesity, even after propensity score matching. Although unmeasured confounders likely remain, our findings support the notion that obesity should not be considered a general contraindication to ECMO.