Upper Gastrointestinal Bleeding in Adults Under Veno-Arterial Extracorporeal Membrane Oxygenation: A Single-Center Retrospective Study

Objective. Upper gastrointestinal bleeding is a common complication in adults treated with veno-arterial Extracorporeal Membrane Oxygenation (VA-ECMO) for refractory cardiogenic shock or cardiac arrest. We aimed to determine risk factors, prevalence and outcomes associated with upper gastrointestinal bleeding (UGIB) in adult patients under VA-ECMO. Design. We conducted a retrospective cohort study (2014-2017) on consecutive VA-ECMO patients. Patients. UGIB was dened as 1) an overt bleeding (hematemesis, melena, hematochezia), or 2) acute anemia associated with a lesion diagnosed on upper gastrointestinal endoscopy. Cause-specic models were used to identify factors associated with UGIB and death, respectively. hazard ratio (CSHR) [95% CI]): 3.06 [1.56; 5.98]), and extracorporeal cardiopulmonary resuscitation (ECPR) (CSHR 2.34 [1.03; 5.35]) were independently associated with an increased risk of UGIB. Conclusions. In adult patients under VA-ECMO, obesity and ECPR were independently associated with UGIB. This study highlights the potential role of obesity and acute ischemia reperfusion injury in the pathophysiology of VA-ECMO-associated UGIB.


Introduction
Veno-arterial Extra Corporeal Membrane Oxygenation (VA-ECMO) is an emergency technique used in adult patients to treat refractory cardiogenic shock or refractory cardiac arrest [1]. There has been a massive increase in its use over the last decade [2,3].
In an multicenter study in 1994 on 2252 critically ill patients, Cook et al. identi ed two independent risk factors for gastrointestinal bleeding, namely respiratory failure and coagulopathy [8]. In the recent SUP-ICU trial conducted on 1034 patients in 2016, variables independently associated with clinically important "stress ulcers" were: three or more coexisting diseases, coexisting coagulopathy, coexisting liver disease, use of renal replacement therapy, acute coagulopathy, and a high organ failure score [6].
The main pathophysiological mechanism of UGIB in critically ill patients is splanchnic hypoperfusion, resulting from both hypotension and release of proin ammatory cytokines at the acute phase. This can notably be increased by an ischemia-reperfusion phenomenon, leading paradoxically to an increase in mucosal ischemia [9]. The clinical presentation of UGIB is non-speci c in the absence of overt bleeding and is usually suspected in the presence of unexplained acute anemia, hemodynamic instability and/or hyperuremia. A de nitive diagnosis is made with upper gastrointestinal endoscopy, which can sometimes lead to a therapeutic action (hemostatic injection, vasoconstrictor injection, clipping…).
Hemorrhagic complications, including UGIB, are frequent during ECMO support and are associated with a poor outcome [10]. Hemorrhagic complications may be observed in VA-ECMO patients for multiple reasons: First, anticoagulation is often used to avoid thrombosis of the oxygenation membrane. Second, ECMO circuits induce a short term loss of high molecular weight von Willebrand factor multimer, which is necessary for primary hemostasis [11]. Last, thrombopenia and low brinogen concentrations are frequent under ECMO [12][13][14].
Pathophysiological mechanisms involved in the development of UGIB under ECMO may include patientrelated factors (i.e. previous ulcer, chronic use of antiplatelet therapy …), illness-related factors (i.e. mechanical ventilation, illness severity anticoagulation, thrombopenia …) [9] and potential ECMO-related factors (i.e. decreased gastric mucosal perfusion [15,16], ischemia-reperfusion injury [17] In the present study, we aimed to identify risk factors, prevalence and outcomes associated with upper gastrointestinal bleeding (UGIB) in adult patients supported with VA-ECMO.

Methods
Design. We performed a single-center retrospective cohort study in the 26-bed medical Intensive Care Unit (ICU) of the Bichat-Claude Bernard University Hospital, APHP, Paris, France.
Ethics. This study was approved by the ethical committee of the French Society of Intensive Care (SRLF). The database was declared to the French National Commission of data processing (CNIL).
Patients. We included consecutive adult patients who received VA-ECMO support for refractory cardiogenic shock or refractory cardiac arrest, between January 1st 2014 and December 31st 2017. Exclusion criteria were 1) age < 18 years; 2) missing data on UGIB, UGE and/or outcome in the ICU. Data collection. Clinical parameters were extracted from the medical charts. These included baseline characteristics at admission (weight, height, comorbidities: chronic use of antiplatelet therapy, history of stroke (ischemic or hemorrhagic cerebral stroke reported in the medical report), history of cancer (cured or ongoing cancer reported in the medical report), peptic ulcer, simpli ed acute physiology score (SAPS II) [19], Charlson score [20], sepsis related organ failure assessment (SOFA) score [21], characteristics at the time of cannulation (i.e., SAVE score [22]), use of antiplatelet therapy during ICU stay, site of implantation, place of implantation, etiology of cardiogenic shock or refractory cardiac arrest, occurrence of pre-ECMO cardiopulmonary resuscitation (CPR), cannulation under chest compression (i.e., extracorporeal CPR (ECPR)) and outcome characteristics (duration of mechanical ventilation (MV), ICU length of stay (LOS), use of renal replacement therapy (RRT), duration and outcome of VA-ECMO, date and cause of death in non-survivors).
Biological parameters were collected at the time of cannulation (hemoglobin, platelets, prothrombin time, creatinine and lactate) and 24 hours before bleeding events (hemoglobin, platelets, prothrombin time, activated partial thromboplastin time) in patients who developed bleeding events.
Transfusion data were extracted retrospectively from the blood bank database of our hospital EFS (Etablissement Français du Sang). For each patient, all red blood cell (RBC) units, fresh frozen plasma (FFP) and platelet units transfused were identi ed and the date of transfusion was noted.
UGIB de nitions and UGE procedures. UGIB was described as an overt bleeding (hematemesis, melena, hematochezia) or acute anemia associated with a lesion found on the UGE. All upper gastrointestinal endoscopy (UGE) procedures performed in the ICU during VA-ECMO support were analyzed. UGE was prescribed by the physician in charge in cases of overt bleeding or acute anemia with a suspicion of GI bleeding. UGE was performed by endoscopy specialists. The results of UGE procedures were classi ed in a pre-de ned list of etiologies. If two or more lesions were found on the same UGE, only the most clinically signi cant one was described in our study.
Stress ulcer prophylaxis, thromboprophylaxis, feeding and transfusion policies. All patients placed on VA-ECMO received a stress ulcer prophylaxis with a proton pump inhibitor (intravenous omeprazole, 40 mg per day) from the day of ICU admission, until ICU discharge. To prevent thrombotic events and membrane clotting during VA-ECMO support, every patient received a continuous infusion of unfractionated heparin, with an anti-Xa target of 0.2 to 0.3 UI/ml that was continued for at least 1 month after ECMO explantation. The decision to stop heparin because of thrombopenia or hemorrhagic events was left to the discretion of the treating physician. Enteral feeding via an oral gastric tube was started within the rst 24 hours of admission, if feasible, in order to achieve appropriate energy needs (25-35 kCal/kg/day) within the rst 7 days. The hemoglobin threshold for transfusion was 8.0 g/dL in the absence of bleeding. In the case of bleeding events, transfusions were determined by the physician in charge.
Statistical analysis. Data are described as numbers (%) for categorical variables and medians (interquartile range, IQR) for continuous variables. Comparisons relied on Fisher's exact test for categorical data, and the Wilcoxon test for continuous data. Using cause-speci c models, we determined risk factors for the occurrence of UGIB in adult patients on VA-ECMO during their ICU stay, considering death as a competing risk. For this purpose, risk factors were explored by univariate analyses. Then, variables yielding a p-value < 0.1 were entered into a multivariate model using a backward selection procedure. Results were expressed as cause-speci c hazard ratios (CSHR) with their 95% con dence intervals (95% CI). Missing data were imputed to the median or to the highest frequency. All analyses were performed using SAS software, version 9.4 (SAS Institute Inc, Cary, North Carolina). A p-value of less than 0.05 was considered statistically signi cant.

Results
Study population. Between January 2014 and December 2017, 259 patients received VA-ECMO support in our ICU. Two patients were excluded because of missing les and 257 patients were nally studied. A study ow chart is provided in additional le 1.
Outcomes are presented in Table 3 Variables associated with UGIB. Figure 1 presents the multivariate analysis (univariate analysis is presented in additional le 2) for UGIB in a cause-speci c manner with death as a competitive risk:  19/74 (26%) revealed at least one gastro-duodenal ulcer, and 10 (14%) revealed gastritis. Nine out of 74 (12%) patients were submitted to an endoscopic therapeutic procedure during the rst UGE (n = 5, epinephrine injection therapy; n = 3, endoclip application and n = 1, thermal coagulation). Among the 74 patients who underwent UGE, 25/74 (34%) patients were submitted to a second endoscopy. Among these patients 9/25 (36%) were found to have gastro-duodenal ulcers, 7/25 (28%) were reported as normal and 3/25 (12%) had signs of gastric erosion. Four out of 25 (16%) patients underwent an endoscopic procedure during the second UGE (n = 3 endoclip application and n = 1 epinephrine injection therapy). 20% (n = 5) of patients with lesions on the second endoscopy, had a rst endoscopy described as normal.
Transfusion characteristics on the day of the UGE procedure are presented in the additional le 2.
Measures of diagnostic accuracy of main clinical and biological characteristics present in the last 24 hours before UGE are presented in the additional le 2. The existence of melena, the transfusion of a least 3 units of red blood cells before UGE, and hematemesis had the highest speci cities (96%, 93%, and 89%, respectively). However, all these three parameters had low sensitivities (31%, 8%, 27%, respectively).

Discussion
In this single center retrospective study that evaluated incidence, risk factors and outcomes associated with UGIB in adult VA-ECMO patients, we found that UGIB occurred in 19% of patients, after a median of 18 [7; 43] days following cannulation. The main ndings during UGE were gastro-duodenal ulcers and a few UGE led to endoscopic therapeutic procedures. A history of obesity and ECPR were the main risk factors. UGIB patients had longer ICU stays, and longer ECMO and mechanical ventilation durations, as compared to non-UGIB patients.
In large studies conducted in the general ICU population, UGIB occurred in 2-5% of cases [6, 8, 23]. The high proportion of UGIB in VA-ECMO patients observed in our cohort could be simply explained by the severe condition of patients on ECMO. These patients often present several risk factors, including high organ failure score, use of renal replacement therapy, coagulopathy, and respiratory failure [6,8]. These variables were not associated with UGIB in our study, as almost every patient received mechanical ventilation, and more than half of the patients received renal replacement therapy.
The population which presented UGIB appeared to be less severely ill at admission (lower SAPS II) which could explain their difference in length of stay, ECMO duration and mechanical ventilation duration. It is noteworthy that UGIB was rarely diagnosed within the rst days following ICU admission. It is thus possible that UGIB was undetected in the most severe patients who died early and in whom diagnostic endoscopic procedures were not performed. This likely explains why UGIB was not associated with death in our analyses.
In our study, we used a competing risk regression model with death as a competing risk to identify risk factors for UGIB. This model allowed us to evaluate the relationship of covariates to cause-speci c events [24], in this case, UGIB and death. In this manner we could identify two independent risk factors for UGIB: BMI > 30 kg/m 2 (CSHR 2.48 [1.47; 5.52]; p < 0.01) and a cannulation under chest compression (CSHR 7.83 [1.59; 38.47]; p = 0.01). These different risk factors could re ect the potential role of obesity, and the importance of ischemia-reperfusion phenomena [25] in the splanchnic hypoperfusion which occurs during shock and VA-ECMO. Obesity is commonly associated with other comorbidities, such as antiplatelet therapy and diabetes. Nevertheless, in our multivariate analysis taking into account these potential cofounders, obesity remained independently associated with UGIB. Obesity is a known risk factor for gastrointestinal complications such as gastroesophageal re ux, erosive esophagitis and gastritis, which is explained by a higher intraabdominal pressure leading to an increase in esophageal acid exposure [26]. Moreover, a decrease in adiponectin has been found in obese patients which is associated with a higher risk of erosive gastritis [27]. To our knowledge, this is the rst study to evaluate obesity as a risk factor for UGIB in ICU patients. Second, the digestive tissue might be very sensitive to ischemia-reperfusion-related damage: intestinal villosities contain the highest concentration in the body of xanthine dehydrogenase which is responsible for the production of superoxide anions, radical oxygen species, in the case of ischemia-reperfusion [28]. Mechanisms involving radical oxygen species that increase the damage caused by ischemia-reperfusion are well-known, including endothelial dysfunction, and platelet and leucocyte activation [29][30][31].
In our study, among the 99 UGE performed, the main ndings were lesions of the digestive mucosa and gastroduodenal ulcer. In a previous study [18], stress gastritis was the most frequent diagnosis among ECMO patients but UGE was only performed in 7 patients. In another study [32], carried out in a medicosurgical intensive care unit without patients on ECMO, 84 UGE were performed: erosion (21%) was the most frequent diagnosis, followed by gastritis (17%) and peptic ulcer (15%) and 25 endoscopies (30%) were described as normal. This cohort seems to present the lesions found in our population in approximately in the same proportion as in our study group. This may suggest that the lesions found in our population may not be speci c to the presence of VA-ECMO. In our cohort, 37 UGE (37%) were described as normal. This high proportion of normal endoscopy may re ect the di culty in diagnosing acute anemia in patients with VA-ECMO. Overt bleeding appeared to be more frequently associated with lesions. Information about increases in norepinephrine and urea -creatinine dissociation was missing and could perhaps be of interest to determine a priori patients with lesions. Although some of these lesions may not result in therapeutic changes, they remain a re ection of splanchnic hypoperfusion and may be responsible for progressive anemia. 9 UGE led to hemostatic endoscopic procedures, a prevalence which, although low, remains high compared to other studies on ICU patients [32].
Our study has several limitations: 1) this is a single-center study, and our results may not be extrapolable to other ECMO centers: it should be noted that we had a relatively high proportion of both post-cardiac surgery VA-ECMO patients and patients with primary graft dysfunction following heart transplantation. However, cardiac infarction and dilated cardiopathy remained the main reasons for ECMO implantation; 2) Our study is based on a retrospective analysis of data. However, all upper gastrointestinal endoscopy reports were computer-based and we can therefore safely assume that no important information about endoscopic procedures was missing; 3) Because the decision to perform UGE was based on clinical suspicion, some asymptomatic gastrointestinal lesions may have been missed; 4) We lacked quantitative information on enteral feeding, which is associated with a decrease in the prevalence of UGIB; we also lacked information concerning anticoagulant therapy and antiplatelet agent management ( the occurrence of UGIB could lead to the discontinuation of the anticoagulants); 5) We did not include lesions found in lower gastrointestinal endoscopy, which is more often prescribed to look for mesenteric ischemia; we aim to investigate the clinical impact of such lesions in a future study.

Conclusion
UGIB is a frequent complication among adult patients under VA-ECMO. Obesity and a cannulation under chest compression are independently associated with UGIB in this population. This study highlights the potential role of obesity and ischemia-reperfusion in the pathophysiology of ECMO-associated UGIB. Organ Failure Assessment