Extracorporeal Membrane Oxygenation Outcomes: COVID-19 Pneumonia vs Non–COVID-19 Pneumonia

We used data from the Healthcare Cost and Utilization Project National Inpatient Sample dataset from 2016 through 2021. The National Inpatient Sample is a publicly available database created by the Agency for Healthcare Research and Quality and is the largest all-payer inpatient dataset in the United States.¹⁴  It provides a stratified sample of 20% of discharges from US community hospitals. Unweighted, the database includes up to 7 million hospital stays per year. Weighted, it estimates about 35 million hospitalizations nationally. This database includes primary diagnoses and secondary diagnoses using International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes. The primary or principal diagnosis is the main reason for the hospitalization, and the secondary diagnoses are any comorbid conditions other than the primary diagnosis. We used ICD-10 procedure codes to identify inpatient procedures. Information in this database is deidentified, so no institutional review board approval was required.

Hospitalizations for COVID-19 were identified with the ICD-10-CM code U07.1 as the primary diagnosis. Hospitalizations for non–COVID-19 pneumonia were identified with the ICD-10-CM codes denoted in Supplemental Table 1 (available online only at www.ajcconline.org) as the primary diagnosis. Patients younger than 18 years were excluded. Patients who underwent ECMO were identified with the ICD-10 procedure codes denoted in Supplemental Table 1 (available online only). Before October 2018, all types of ECMO were billed under the same code, so these were labeled “unspecified ECMO.” Patients who did not undergo any type of ECMO were excluded. Patients in the cohort were categorized as those with COVID-19 pneumonia who underwent ECMO and those with non–COVID-19 pneumonia who underwent ECMO.

Sociodemographic data including age, sex, race or ethnicity, median household income for each patient’s zip code, insurance, and hospital size (number of beds) were collected from the database. Comorbidities were identified using the Agency for Healthcare Research and Quality’s Elixhauser Comorbidity Software,¹⁵  which uses a “present-on-admission” indicator to identify preexisting conditions, thereby minimizing confounding of preexisting comorbidities with in-hospital diagnoses.

The primary outcome of interest was inpatient mortality. Secondary outcomes of interest included rate of vascular complications, need for dialysis or continuous renal replacement therapy, total hospital costs, length of stay, and discharge disposition. The diagnosis and procedure codes are detailed in Supplemental Table 1 (available online only).

Baseline characteristics were compared between both groups before matching by using t tests for continuous variables and χ² tests for categorical variables. Propensity scores were estimated from logistic regression models with COVID-19 pneumonia as the dependent variable. Covariates included in the model were age, sex, race or ethnicity, median income, insurance, and Agency for Healthcare Research and Quality’s Elixhauser comorbidities. Patients with COVID-19 pneumonia were matched to patients with non–COVID-19 pneumonia in a 1:5 ratio on the basis of propensity scores using nearest-neighbor matching with replacement. The maximum caliper distance was set at 0.25 SDs of the calculated score.¹⁶  Covariate balance after matching was assessed with the absolute standardized mean difference of the matched cohort, where less than 0.1 was considered satisfactory.¹⁷ 

Discharge-level weights, strata, and cluster elements were applied to account for the complex survey design and clustering as recommended by the Agency for Healthcare Research and Quality.¹⁸  Differences were considered significant at a 2-tailed P value of less than .05. All analyses were performed using the survey suite of commands in Stata 18.0 (StataCorp LLC).

From 2016 through 2021, a total of 2 573 077 adult patients with COVID-19 pneumonia and 633 854 adult patients with non–COVID-19 pneumonia were admitted. Of these, 5680 patients with COVID-19 pneumonia and 430 patients with non–COVID-19 pneumonia underwent ECMO (Figure 1). Baseline patient characteristics are detailed in Table 1.

Patients with COVID-19 pneumonia had a mean (SD) age of 46.0 (11.2) years, and those with non–COVID-19 pneumonia had a mean (SD) age of 45.1 (12.5) years. Patients in both cohorts were predominantly male (71.4% of patients with COVID-19 pneumonia and 61.6% of patients with non–COVID-19 pneumonia). Compared with the non–COVID-19 pneumonia group, the COVID-19 pneumonia group had a greater proportion of Hispanic and African American patients, significantly higher rates of pulmonary circulation disorders and coagulopathy, and significantly lower rates of liver disease and depression.

Among patients with COVID-19 pneumonia, the most frequently used support modality was venovenous ECMO, followed by venoarterial ECMO and central ECMO. Among patients with non–COVID-19 pneumonia, the frequencies of support modality use were similar, but the ECMO modality was unknown for patients billed before October 2018. The types of ECMO used for all patients are shown in Figure 2.

Propensity score matching resulted in 1136 patients in the unweighted COVID-19 pneumonia group paired with 86 patients in the unweighted non–COVID-19 pneumonia group. The mean patient age was 46 years in both groups. Baseline characteristics of the matched cohorts are denoted in Table 2, and the adequate balance achieved after matching is shown in Figure 3.

In the matched analysis, patients with COVID-19 pneumonia had a higher risk of death than did patients with non–COVID-19 pneumonia (odds ratio, 1.98). The mean stay was significantly longer for patients with COVID-19 pneumonia than for patients with non–COVID-19 pneumonia (38.0 days vs 28.5 days). Mean total hospital costs were also significantly higher for patients with COVID-19 pneumonia than for patients with non–COVID-19 pneumonia ($1 278 270 vs $967 866). The need for dialysis or continuous renal replacement therapy and the rate of vascular complications were not different between groups. Patients with COVID-19 pneumonia were less likely to be discharged home than were patients with non–COVID-19 pneumonia (odds ratio, 0.42). Primary and secondary outcomes are shown in Figure 4. Types of vascular complications in both groups are shown in Supplemental Table 2 (available online only).

Our study showed that after groups of patients were balanced by comorbidities, those with COVID-19 pneumonia who underwent ECMO were more likely to die than were those with non–COVID-19 pneumonia who underwent ECMO. Patients with COVID-19 pneumonia also had longer hospital stays and higher hospital costs and were less likely to be discharged home after their hospitalization than were patients with non–COVID-19 pneumonia. The need for dialysis or continuous renal replacement therapy and the rate of vascular complications did not differ.

Our findings affirmed previous research that demonstrated higher mortality rates associated with ECMO in patients with COVID-19 than in patients with other forms of cardiorespiratory failure.^11,12,19,20  For instance, a single-center study showed that 38.5% of patients with COVID-19 treated with venovenous ECMO were alive at 60 days, compared with 63.6% of patients with other infectious causes of respiratory failure.¹⁹  Another study compared venovenous ECMO in patients with COVID-19 pneumonia versus those with non–COVID-19 pneumonia and also revealed a higher mortality rate in the COVID-19 group (49% versus 24%).²⁰  Advanced age was the strongest predictor of poor outcome in patients with COVID-19.^13,21–23 

Our study adds to the existing literature by revealing that elevated mortality rates persist in patients with COVID-19 pneumonia treated with ECMO even after the cohorts are matched for age, sex, race, and underlying comorbidities. This finding shows that the increased risk of mortality cannot be solely attributed to factors like age or underlying comorbidities but rather is related to factors directly associated with COVID-19. In addition, patients with COVID-19 who survive after undergoing ECMO have higher burdens, as reflected in longer hospitalizations, higher costs, and discharge to places other than home.

Several contributors to increased mortality rates in patients with COVID-19 undergoing ECMO have been proposed in other studies. These contributors include higher rates of thrombosis associated with COVID-19,²⁴  lung injury due to prolonged noninvasive ventilation use before ECMO, and a more liberal use of ECMO during the COVID-19 pandemic.²⁵  The use of ECMO was associated with higher survival rates when performed in high-volume centers with substantial ECMO experience.^19,26  As our study revealed, the number of patients treated with ECMO increased after the emergence of COVID-19. For instance, the number of patients with COVID-19 pneumonia treated with ECMO in 2020 and 2021 surpassed the number of patients with non–COVID-19 pneumonia treated with ECMO during the combined 6-year period from 2016 through 2021. The increased use of ECMO most likely required centers with lower prepandemic ECMO exposure, and therefore less experience, to deliver care to a larger population of patients undergoing ECMO during the COVID-19 pandemic, resulting in higher mortality rates.

Our study contributed evidence by assessing a large database of patients with COVID-19 pneumonia treated with ECMO in multiple centers across the nation compared with a broad range of patients with viral pneumonia. An additional strength of our study is that it is broader than previous studies of ECMO in patients with COVID-19 in that it evaluated outcomes of patients who received venovenous, venoarterial, and central ECMO, as opposed to only venovenous ECMO.

The findings of this study should be interpreted in the context of its limitations and at best viewed as hypothesis generating and thought-provoking. First, the study relied on administrative databases, and accuracy and reliability are contingent on the correct classifications within the ICD-10 codes. Second, propensity score matching mitigates bias but never fully replicates the conditions of a randomized controlled trial because only observed variables can be included in the matching algorithm. Unobserved (or unrecorded) variables may remain unbalanced and unaccounted for, even after matching, thus potentially introducing bias into our results. Third, the ICD-10 code for COVID-19 was not developed until late 2020, so some patients treated during that year might not have been captured in the dataset and analysis. Despite these limitations, the use of a nationwide database enabled a large sample size originating from numerous institutions across the country, which we hope improves the generalizability of our results.

In this study, patients with COVID-19 pneumonia who underwent ECMO had a higher mortality rate (independent of sociodemographic factors and comorbidities), longer hospital stays, higher hospitalization costs, and lower odds of being discharged home than did patients with non–COVID-19 pneumonia who underwent ECMO.