Ventilator-associated pneumonia in patients assisted by veno-arterial extracorporeal membrane oxygenation support: Epidemiology and risk factors of treatment failure

Introduction Ventilator-associated pneumonia (VAP) is frequent in Intensive Care Unit (ICU) patients. In the specific case of patients treated with Veno-Arterial Extracorporeal Membrane Oxygenation Support (VA-ECMO), VAP treatment failures (VAP-TF) have been incompletely investigated. Methods To investigate the risk factors of treatment failure (VAP-TF) in a large cohort of ICU patients treated with VA-ECMO, we conducted a retrospective study in a Surgical ICU about patients assisted with VA-ECMO between January 1, 2013, and December 31, 2014. Diagnosis of VAP was confirmed by a positive quantitative culture of a respiratory sample. VAP-TF was defined as composite of death attributable to pneumonia and relapse within 28 days of the first episode. Results In total, 152 patients underwent ECMO support for > 48h. During the VA-ECMO support, 85 (55.9%) patients developed a VAP, for a rate of 60.6 per 1000 ECMO days. The main pathogens identified were Pseudomonas aeruginosa and Enterobacteriaceae. VAP-TF occurred in 37.2% of patients and was associated with an increased 28-day mortality (Hazard Ratio 3.05 [1.66; 5.63], P<0.001), and VA-ECMO assistance duration (HR 1.47 [1.05–2.05], P = 0.025). Risk factors for VAP-TF were renal replacement therapy (HR 13.05 [1.73; 98.56], P = 0.013) and documentation of Pseudomonas aeruginosa (HR 2.36 [1.04; 5.35], P = 0.04). Conclusions VAP in patients treated with VA-ECMO is associated with an increased morbidity and mortality. RRT and infection by Pseudomonas aeruginosa appear as strong risks factors of treatment failure. Further studies seem necessary to precise the best antibiotic management in these patients.


Introduction
Ventilator-associated pneumonia (VAP) is frequent in Intensive Care Unit (ICU) patients. In the specific case of patients treated with Veno-Arterial Extracorporeal Membrane Oxygenation Support (VA-ECMO), VAP treatment failures (VAP-TF) have been incompletely investigated.

Methods
To investigate the risk factors of treatment failure (VAP-TF) in a large cohort of ICU patients treated with VA-ECMO, we conducted a retrospective study in a Surgical ICU about patients assisted with VA-ECMO between January 1, 2013, and December 31, 2014. Diagnosis of VAP was confirmed by a positive quantitative culture of a respiratory sample. VAP-TF was defined as composite of death attributable to pneumonia and relapse within 28 days of the first episode.

Results
In total, 152 patients underwent ECMO support for > 48h. During the VA-ECMO support, 85 (55.9%) patients developed a VAP, for a rate of 60.6 per 1000 ECMO days. The main pathogens identified were Pseudomonas aeruginosa and Enterobacteriaceae. VAP-TF occurred in 37.2% of patients and was associated with an increased 28-day mortality (

Introduction
As illustrated with the most recent data [1], the incidence of extracorporeal membrane oxygenation (ECMO) support has increased for over a decade. In-hospital mortality of patients assisted with veno-arterial (VA-ECMO) or veno-venous (VV-ECMO) ECMO remains high, secondary to the severity of illness leading to ECMO support, but also to frequent and severe complications, most being infectious [2]. Hence, nosocomial infections under ECMO support (from 14.8 to 75.5 infectious episodes per 1000 ECMO days) are associated with longer duration of ECMO support, mechanical ventilation, and ICU and hospital lengths of stay [3][4][5][6].
Although pneumonia appears to be the most frequent infection under ECMO, no specific data exist on this specific issue. Epidemiology, risk factors, microbiology, specific outcome or risks of treatment failure (TF) of ventilator-associated pneumonia under ECMO support remain unknown. However, severity of patients assisted with ECMO, immunosuppression secondary to extracorporeal circulation [7][8][9] as well as changes in antibiotics pharmacokinetics under ECMO [10][11][12] make us think that VAP under ECMO is a distinct entity.
To address this issue, we aimed to investigate the epidemiology, risk factors, microbiology and outcomes of VAP, and to identify risk factors of TF in a large and well-defined cohort of ICU patients treated with VA-ECMO.

Design and setting
This observational, retrospective and monocentric study was conducted in the Surgical Intensive Care Unit of Cardiology Institute in La Pitié-Salpêtrière University Hospital (Paris, France). We included all patients assisted with VA-ECMO and admitted in our unit.

Standard care procedure for patients on ECMO
Procedures for implantation, daily management and weaning were realized according to the recommendations of the Extracorporeal Life Support Organization (ELSO) and are described in S1 File. For patients with peripheral cannulation, prophylactic antibiotic therapy was performed with second-generation cephalosporin as a single intravenous injection (cefazoline 2g) at the time of ECMO implantation. Patients with central cannulation received prophylactic antibiotic for the surgery (cefazoline 2 g within the hour prior to incision, then 1g every 4h), but no antibiotic prophylaxis was continued during the course of ECMO.

Prevention of VAP
VAP prevention is based on a combination of measures: orotracheal intubation, endotracheal tube cuff pressure maintained between 20 and 30 cm H 2 O, oral chlorhexidine application four times per day, semi recumbent body positioning (30 to 45˚) when possible. Sedation is patienttargeted, relying on assessments eight times per day of Richmond Agitation-Sedation Scale (RASS). Patients are started on enteral tube feeds within 48 h of intubation whenever possible.
Ventilator-associated pneumonia under ECMO support was defined post hoc by 2 independent infectious experts (AB, CB) as a VAP occurring in patients assisted with VA-ECMO for more than 24 hours or withdrawn for less than 48 hours. Immunosuppression was defined as HIV, immunosuppressive therapy, or corticosteroids> 0.5 mg/kg per day for more than a month. In case of persistence, worsening or recurrence of clinical criteria, a new respiratory sample was performed. After 6 days of antibiotic therapy against the identified pathogen, and up to 48 hours after the end of antibiotic therapy, we considered a positive quantitative culture of the same pathogen as a persistence of VAP (S1 Fig). If this new respiratory sample was performed after ! 48 hours without antibiotic treatment, we considered positive quantitative culture of the same pathogen as a relapse of VAP. During antibiotic therapy of initial VAP, and up to 48 hours after the end of antibiotic therapy, we considered a positive quantitative culture of the new pathogen as a superinfection. After ! 48 hours without antibiotic treatment, we considered positive quantitative culture of the new pathogen as a new VAP. Nevertheless, because the occurrence of a new VAP, documented with a new pathogen is part of the prolonged ventilation of a severe ICU patient, and does not appear to be directly related to the presence of ECMO, we did not consider it as a treatment failure. Attributable death to VAP was defined as death under ECMO during antibiotic therapy for VAP without any other evident cause of death, or death under ECMO secondary to septic shock of pulmonary origin.
Our primary endpoint was treatment failure, a composite of attributable death to pneumonia or persistence or relapse of pneumonia within 28 days of the first episode.

Data collection
The following information were recorded retrospectively for each patient: demographic data; clinical variables; comorbidities; reason for ECMO; duration of ECMO support, mechanical ventilation, ICU stay, and hospital stay; presence and type of infectious complications, hemorrhagic complications, digestive complications, neurologic complications and acute kidney injury.

Ethics approval and consent to participate
The ethics committee Ile de France 5 approved the collection of data compiled in our ECMO registry (Reference B-7-15). Due to the retrospective design of the study, and in accordance with the decision of the ethics committee, no consent was needed.

Statistical analysis
Results are expressed as number of patients (%), or mean (± Standard deviation). When indicated, 95% CIs were calculated. Influence of the occurrence of VAP on overall survival, ECMO assistance duration and mechanical ventilation duration were assessed with univariate Cox regression models, with ECMO-associated pneumonia as a time-dependant covariable. Kaplan-Meier representation of the overall survival according to ECMO-associate pneumonia occurrence was obtained using the Simon and Makuch method [13].
Univariate and multivariate Poisson regression survival models with the log of ECMO duration as an offset term were used to evaluate risk factors of ECMO-associated pneumonia. Univariate and multivariate Cox analyses were used to identify risk factors of treatment failure among patients with VAP. In both cases, all variables associated (p < 0.10) with the outcome in the univariate analysis were introduced into the multivariate model. Relative risks (RR) for Poisson model and hazard ratios (HR) for Cox model were calculated with their 95% confidence intervals (CI).
Significance was defined as p-values less than 0.05. Statistical analyses were performed using R 3.3.3 (http://www.R-project.org).

Population
Between January 2013 and December 2014, 163 patients were assisted by VA-ECMO and retrospectively included in our database. Of these 163 patients, 11 died within the 48 first hours and were not included in the study. Demographic and clinical characteristics of the 152 patients with ECMO support > 48 hours are described in Table 1
Occurrence of a VAP was associated with increased length of ECMO assistance duration (HR 1.   Table 3). Due to too many missing data, SOFA score at admission was not included in the multivariate analysis, which found only the female gender as associated with a decrease in the incidence of VAP (0.52 [0.30;0.91], p = 0.02). Due to the small number of patients with central  Table 4).

Discussion
In this large retrospective cohort study in VA-ECMO patients, VAP occurs in 56% of patients, early after VA-ECMO implantation, and increases significantly the 28-day mortality. Treatment failure was identified as a frequent event, associated with the documentation of Pseudomonas aeruginosa and the presence of renal replacement therapy at diagnosis. The previously published studies about infectious complications under ECMO support are either pediatric patients-focused [14], from large international registries [4], or interested in all types of infections, and did not address this specific issue, whereas VAP remains the main infection in intensive care [15] as in ECMO patients [4][5][6]. Incidence of infections under ECMO and particularly VAP varies greatly depending the studies, ranging from 14.8 to 75.5 infectious episodes per 1000 days of ECMO [3][4][5][6]. With a rate of 60.6 per 1000 ECMO days, our results are in the high range of previously published studies, probably linked to the severity of our patients (SAPS 2 score at admission of 54.4 (22.3)) and to the extensive review of hospitalization reports, including daily observation, biological and microbiological data.
Moreover, since cardiopulmonary bypass (CBP) is know to induce an immunosuppression, decreasing monocyte response to pathogens [7][8][9] and significant reduction of HLA-DR and TLR4 expression on alveolar macrophages [16], ECMO could induce the same alterations that CBP and thus predispose patients to infectious complications, in particular pneumonia. However, to the best of our knowledge, there is no pre-clinical data to support this hypothesis so far. Finally, it is interesting to note that the main microorganisms documented in our cohort Ventilator-associated pneumonia in patients assisted by VA-ECMO were Enterobacteriaceae (39.4%) and Pseudomonas aeruginosa (18.2%), even though pneumonia was an early event during assistance. This proportion is similar to that found in the post cardiac surgery context [17], and could be linked with a bacterial translocation in this context of circulatory failure. In our cohort, Renal Replacement Therapy (RRT) and documentation of Pseudomonas aeruginosa were associated with treatment failure. The increase of volume of distribution as much as the antibiotic drug sequestration in the circuit secondary to ECMO could accentuate the pharmacokinetic alterations seen in critically ill patients, leading to poor achievement of target concentrations of antibiotics, especially hydrophilic antibiotics used against P. aeruginosa (e.g., aminoglycosides, beta-lactams) [18]. Moreover, RRT is known to add a significant variability in antibiotic concentrations [19,20]. Hence, it was shown that the combination of ECMO and RRT could alter meropenem pharmacokinetic, especially for P. aeruginosa [11].
Our study suffers from several limitations. The diagnosis of VAP is challenging for clinicians [21], and even more in patients assisted with ECMO, since the parameters usually used for VAP diagnosis are not interpretable in this setting [22,23]. Hence, to our knowledge, the  Ventilator-associated pneumonia in patients assisted by VA-ECMO and analyzed, the retrospective design of the study as well as the relatively low sample size of the cohort limits the strength of our findings. In addition, even though SOFA score was significant in the univariate analysis of risk factors for pneumonia, the large number of missing data did not allow us to integrate it into the multivariate analysis. Finally, the monocentric design of the study limits its external reproducibility, even if a large cohort obtained on a short period and in a single center guarantees homogeneity of management. Ventilator-associated pneumonia in patients assisted by VA-ECMO

Conclusion
VAP is a frequent and severe complication in patients treated with VA-ECMO, and is associated with an increased morbidity and mortality. RRT during antibiotic therapy and infection by Pseudomonas aeruginosa appear as strong risks factors of treatment failure. The best antibiotic management in patients with RRT and with ECMO remains to be determined.
Supporting information S1