ECMO Support for Inuenza A: Retrospective Review of the ELSO Registry Comparing Seasonal and Pandemic Subtypes

Background: While there is substantial published experience of ECMO during the H1N1 pandemic, less is known about the use of ECMO in patients with seasonal inuenza A virus. We hypothesized that the severity of illness and survival of patients supported with extracorporeal membrane oxygenation (ECMO) would differ for those with seasonal inuenza A vs pandemic H1N1 (H1N1) inuenza A. Methods: Retrospective study of ECMO supported adults (>18 years) with inuenza A viral infection reported to the Extracorporeal Life Support Organization (ELSO) Registry between 2009-2019. We describe the incidence and compare characteristics and factors associated with in-hospital survival using a least absolute shrinkage and selection operator regression. Results: Of 2461 patients supported with ECMO for inuenza A, 445 had H1N1 and 2004 had seasonal inuenza A. H1N1 was the predominant subtype between 2009-2011. Pandemic H1N1 patients were younger, with more severe illness at ECMO cannulation and higher reported ECMO complications than those with seasonal inuenza A. Patient characteristics including younger age and higher weight, and patient management including longer ventilation duration before ECMO were associated with worse survival. ECMO complications were associated with reduced survival. There was no difference in survival to hospital discharge according to inuenza subtype after adjusting for other characteristics. Conclusions: Patients supported with ECMO for pandemic H1N1 were younger, with more severe illness than those supported for seasonal inuenza A. Survival to hospital discharge, was associated with patient characteristics, management, and ECMO complications, but was not impacted by the specic inuenza A subtype. membrane oxygenation, ECPR: ECMO cardiopulmonary resuscitation, H1N1: pandemic inuenza A H1N1 subtype, IQR: interquartile range, kg: kilograms, NaH2CO3: sodium bicarbonate, paCO2: partial pressure of arterial carbon dioxide, PIP: positive inspiratory pressure, THAM: tromethamine, VA; venoarterial, VV: venovenous, VVA: veno-venoarterial A H1N1 subtype, IQR: interquartile range, kilograms, NaH2CO3: bicarbonate, paCO2: arterial dioxide, inspiratory cardiopulmonary H1N1: pandemic inuenza H1N1 subtype, IQR: interquartile range, kg: kilograms, odds ratio, paCO2: partial of


Background
In 2009, the H1N1 in uenza A pandemic lead to a surge of extracorporeal membrane oxygenation (ECMO) use in critically ill patients with acute respiratory distress syndrome (ARDS) (1,2,3,4). Prior to this, ECMO use in adults with ARDS was relatively rare because two early randomized clinical trials failed to demonstrate a survival bene t (4,5). A large retrospective review of over 1400 adults with ARDS supported on ECMO before 2006 showed 50% survival (6). The conventional ventilatory support versus extracorporeal membrane oxygenation for severe adult respiratory failure (CESAR) trial in 2009 was the rst to demonstrate the safety of ECMO utilization in patients with ARDS (7). More recently, the ECMO to Rescue Lung Injury in Severe ARDS (EOLIA) trial supports a role for ECMO in adult ARDS management (8, 9).
We hypothesized that the severity of illness and survival of patients supported with ECMO would differ for those with seasonal in uenza A vs pandemic H1N1. Against this background, our aims for this project were to 1) describe the incidence of ECMO use over time for H1N1 vs seasonal in uenza A; 2) compare characteristics of patients supported on ECMO with H1N1 vs seasonal in uenza A; and 3) identify and compare factors associated with survival to hospital discharge in adults with H1N1 vs seasonal in uenza A supported with ECMO.

Study Design
We conducted a multicenter retrospective cohort study using the Extracorporeal Life Support

Variable selection
Predictor variables were determined a priori, with the inclusion of previously identi ed factors associated with mortality (29,30). Variables were identi ed by ICD codes (Online Supplement 1) including central nervous system (CNS) dysfunction, immunocompromised state, and shock (29,30). Age, sex, race, weight, pre-ECMO variables including cardiopulmonary arrest or ECPR, duration of mechanical ventilation, renal dysfunction requiring renal replacement therapy, use of neuromuscular blockade agents or inhaled nitric oxide, metabolic buffer infusions, peak inspiratory ventilation pressure (PIP), mean airway pressure (MAP) and partial pressure of arterial carbon dioxide (paCO2) and any known non-pulmonary coinfections were included as covariates. Year of ECMO support, hours of ECMO support, mode of support, and primary indication for support were incorporated as explanatory variables. Implausible blood gas values were assessed for possible entry in kilopascal instead of millimeters of mercury (mmHg) using an algorithm to calculate the pH according to the Henderson-Hasselbalch equation. If the calculated pH corresponded to the pH of the source, arterial blood gas values were converted to mmHg by multiplying them by 7.5. Missing paCO 2 were replaced by calculated ones if pH and HCO3 were entered; missing pH values were calculated if paCO2 and HCO 3 were available. Variables with more than 15% missing data were excluded from the analysis.

Page 4/24
The primary outcome of interest was survival to discharge from the ECMO center. The secondary outcomes were complications, which were selected by review of the ELSO International Summary Report 2020, where variables showed a proportional survival of less than 50% in adult patients with respiratory ECMO (Online Supplement 2) (30).

Statistical Analysis
Patient and ECMO characteristics were compared between pandemic H1N1 and seasonal in uenza A cohorts using univariate analysis. Variables included demographics, comorbidities, pre-ECMO respiratory status, and complications. Categorical and dichotomous variables were expressed as exact numbers with percentages and analyzed with Fisher's exact or Pearson's chi-square. Continuous variables were expressed as median values with 25th -75th interquartile ranges (IQR) and analyzed with the Wilcoxon-Mann-Whitney test. Univariate unadjusted logistic regression was used to explore the association of patient characteristics against the primary outcome of survival to hospital discharge and reported as odds ratio (OR) with 95% con dence intervals (CI).
We describe the incidence and compare characteristics and factors associated with in-hospital survival by both unadjusted logistic regression and multivariate logistic regression with the least absolute shrinkage and selection operator (LASSO) regularization. We employ the LASSO method to achieve dataadaptive variable selection to build a parsimonious, interpretable model (31). Inference in LASSO is notoriously di cult; to this end, we adopt a recently proposed statistical method to mitigate the randomness in the selection of LASSO that supports principled comparison of mortality between in uenza subtypes in the presence of LASSO. In particular, CIs and p-values for ORs in the multivariate analysis are derived by exact post-selection inference to ensure valid inference after variable selection by LASSO (32). We preselected the following variables for prediction of mortality: H1N1 and seasonal in uenza A, sex, weight (strati ed by interquartile ranges with the highest interquartile weight 110-361kg as reference variable), age by 18-49 years, 50-59 years, and ≥ 60 years, mechanical ventilation prior ECMO < 48 hours, 48 hours to 7 days or ≥ 7 days, CNS dysfunction, diagnosis of shock or immunocompromised state, neuromuscular blockade prior to ECMO, nitric oxide prior to ECMO, metabolic buffer infusions prior to ECMO, other non-respiratory co-infections, cardiac arrest prior to ECMO, and paCO2 ≥ 75 mmHg. For an explanatory model of survival, we added to our prediction model, the year of ECMO support (strati ed by pandemic years 2009-2011, years after 2011), hours of ECMO support (strati ed by interquartile ranges with variable reference < 146.5 hours), and complications while on ECMO support.
Statistical signi cance was de ned as a p-value < 0.05. Statistical analyses were carried out using R software (version 3.6.1, R foundation for Statistical Computing).

Patients supported on ECMO with H1N1 vs other in uenza A subtypes
Patients with pandemic H1N1 were differentiated from seasonal in uenza A and the incidence was determined (Fig. 2). The frequency of reported ECMO support increased during the years 2009-2011 with pandemic H1N1 as the predominant early subtype, but since 2012 other seasonal in uenza A subtypes became the leading viral etiology associated with ECMO support. The number of ECMO centers contributing data to the ELSO registry increased from 164 to 463 during the study period (30

Discussion
This study demonstrates that patients supported on ECMO for pandemic H1N1 had more severe features of critical illness, despite being younger, with higher weight and having fewer comorbidities than those subsequently managed on ECMO for seasonal in uenza A. These ndings support increased virulence associated with novel virus triggering the pandemic, but may additionally re ect resource limitation of this invasive support during the associated abrupt increase in critical care utilization. Importantly, despite differences in severity of illness, there was no difference in survival to hospital discharge for those patients with pandemic H1N1 compared with patients subsequently managed on ECMO with seasonal in uenza A. We did identify patient characteristics, aspects of patient management before ECMO, and ECMO complications that were associated with survival to hospital discharge.
Igniting the surge in ECMO use for adults with ARDS was the success of ECMO during the H1N1 pandemic (2,3,(17)(18)(19)(20)(21)(22)(23)(24)(25) (24,29,42,43). However, the majority of our patients had seasonal in uenza and not speci cally H1N1, and thus factors associated with mortality in our predictive and explanatory models may be more applicable to other viral subtypes causing ARDS. As in previous studies, younger age, higher weight, and lack of reported comorbidities were associated with survival (2,6,19,22,32,34,42). Additionally, those patients who were managed with a shorter duration of mechanical ventilation, who had not progressed to cardiac arrest prior to ECMO cannulation were found to have improved survival, supporting early initiation of ECMO for viral ARDS (3,6,18,29,42). Established ECMO programs with integrated systems to prevent and mitigate complications may be best placed to offer this invasive support, even during times of pandemic-associated resource limitation (37,44).

Study limitations
Our study has the expected limitations inherent in a retrospective observational study. ELSO Registry data is entered voluntarily, without external validation of data in the represented era, however, the institution of a data dictionary, data entry exam, and logic-limited data entry has resulted in improved data quality in the ELSO registry over the duration of this study (45). Our data may be subject to era effect. Some unidenti ed confounding covariates, such as the older population's prior exposure to H1N1, may impact our results. Our application of LASSO regression adjusting for prede ned comorbidities used in the RESP score is a strength of our analysis; however, we did not speci cally include other potential comorbidities (29,46). Additionally, clinically relevant covariates that had more than 15% missing data were excluded from the analysis.

Conclusions
Over the last decade, the utilization of ECMO for viral ARDS has become well established. In this study of patients with In uenza A supported with ECMO, those with pandemic H1N1 were younger, with more severe illness than those supported for seasonal in uenza A. Survival to hospital discharge was associated with patient characteristics, management, and ECMO complications, but was not impacted by the speci c in uenza A subtype. Identi cation of these factors may inform patient selection and pre-ECMO management, which is especially important in the setting of resource limitation.

Declarations
Ethics approval and consent to participate: Permission to analyze the data was granted by the Availability of data and materials: The data that support the ndings of this study are available from the Extracorporeal Life Support Organization Scholarly Oversite Committee with submission of the written request form.
Competing interests: PA reports grants from the National Institutes of Health (NIH) to support research activities not speci c to this study (1R13HD104432-01 Pediatric ECMO Anticoagulation CollaborativE -PEACE). JET is supported by a Career Development Award from the National Institutes of Health/National Heart, Lung, And Blood Institute (K23 HL141596). JET received speaker fees and travel compensation from LivaNova and Philips Healthcare, unrelated to this work. RPB reports grants from National Institutes of Health (R01 HL153519-ASCEND; K12 HL138039-TACTICAL; R01 HD01543-Pediatric Implantable Arti cial Lung) outside the submitted work; RPB also discloses that he is the Extracorporeal Life Support Organization (ELSO) Registry Chair. All other authors have no con icts of interest to disclose (EO, MA, ML, HL,JL, PR, GL, and LS).
Funding: Internal funding was secured for this project.
Authors' contributions: Study conception, design, material preparation, data collection and analysis were performed by MA, EO, and PA. ML conceived and supervised the statistical analysis. HL, JA, and ML performed the statistical analysis. The rst draft of the manuscript was written by EO, and all authors revised the manuscript for important intellectual content. All authors read and approved the nal manuscript.