The Role of Excessive Anticoagulation and Missing Hyperinflammation in ECMO-Associated Bleeding

Extracorporeal membrane oxygenation (ECMO) is increasingly used in carefully selected patients with cardiac or respiratory failure. However, complications are common and can be associated with worse outcomes, while data on risk factors and outcomes are inconsistent and sparse. Therefore, we sought to investigate potential risk factors and predictors of haemorrhage and adverse events during ECMO and its influence on mortality. We retrospectively reviewed all patients on ECMO support admitted to intensive care units of a tertiary university centre in Austria. In a period of ten years, ECMO support was used in 613 patients, with 321 patients meeting the inclusion criteria of this study. Haemorrhage, occurring in more than one third of the included patients (123, 38%), represented the most common and serious ECMO complication, being associated with an increased one year mortality (51% vs. 35%, p = 0.005). The main risk factors for haemorrhage were severity of the disease (hazard ratio (HR) = 1.01, p = 0.047), a prolonged activated partial thromboplastin time (HR = 1.01, p = 0.007), and lower values of C-reactive protein (HR = 0.96, p = 0.005) and procalcitonin (HR = 0.99, p = 0.029). In summary, haemorrhage remained the main ECMO complication with increased mortality. Moreover, we reported a possible association of lower inflammation and bleeding during ECMO support for the first time. This generated a new hypothesis that warrants further research. Finally, we recommend stricter monitoring of anticoagulation especially in patients without hyperinflammation.


Introduction
The use of extracorporeal membrane oxygenation (ECMO) in patients with refractory cardiac or respiratory failure is increasing [1][2][3]. ECMO is a temporary mechanical extracorporeal support, aiming to support pulmonary or cardiac functions until recovery. It can be established as a venovenous (vv-ECMO) or venoarterial (va-ECMO) configuration. Additional possible uses of ECMO support include bridging to lung or heart transplant and rewarming of patients with severe hypothermia. Moreover, initiation of ECMO support through emergency medical service for out-of-hospital cardiac arrest has been popularised in the last decade [4,5]. According to the Extracorporeal Life Support Organization (ELSO) recommendations, ECMO support should be considered in cardiorespiratory failure with a mortality risk above 50%, and is indicated in a selected patient population with a mortality risk exceeding 80% [6].
Based on the data from 521 international ELSO registry centres, more than 154,000 ECMO runs were registered until the end of 2020, with 18 260 runs during the last year. The overall survival to hospital discharge or transfer was 54% [7].
While this type of extracorporeal support is lifesaving in selected patients, complications are still common and both haemorrhage and thrombosis are associated with reduced outcome [8,9], independent of the initial ECMO indication. Recently, a group of ECMO pioneers from the University of Michigan reported an overall bleeding incidence of 39% [10]. Intracranial haemorrhages or infarction represented a total of 8%. Therefore, improvement in prevention and early recognition of haemostatic complications including both haemorrhage and thrombosis remain pivotal to improve patient outcomes through better management of anticoagulation therapy.
In this study, we aimed to investigate potential risk factors and predictors of haemorrhage during ECMO support. Moreover, we provided a summary and comparison of the demographic and clinical characteristics of patients undergoing ECMO support, while focusing on adverse events and outcomes.

Patient Selection
We retrospectively reviewed the electronic medical records of all patients admitted to the trauma intensive care unit (ICU) and the general and surgical ICU of the Department of Anaesthesiology and Critical Care Medicine, Medical University Innsbruck, Austria. These tertiary ICUs treat surgical, post-trauma, and medical patients. The observation period included 10 years, from January 2010 to December 2019.
All patients undergoing ECMO were assessed for eligibility. Exclusion criteria were patients younger than 15 years, patients having their second ECMO initiation, and patients with incomplete data sets.

Data Collection
We obtained (1) socio-demographic data including age, sex, body weight, height, body mass index, and information regarding ICU ward; (2) data on disease severity as the simplified acute physiology score III (SAPS III) and sequential organ failure assessment (SOFA) score on ICU admission, mechanical cardiopulmonary resuscitation before or during ECMO implantation, underlying disease, indication and type of ECMO support, and duration of ECMO support; (3) detailed information on complications; (4) use of anticoagulation, transfusion of blood, and coagulation products; (5) coagulation status including platelet count (g/L), fibrinogen (modified Clauss method, mg/dL), factor XIII (%), rotational thromboelastometry (ROTEM ® ), prothrombin time (%), activated partial thromboplastin time (aPTT, seconds), international normalised ratio and antithrombin (%); (6) other laboratory parameters such as haemoglobin (g/L), haematocrit (l/L), erythrocytes (T/L), white blood cells (g/L), C-reactive protein (CRP, mg/dL), and procalcitonin (µg/L); and finally (7) cause and date of death with delivered data on ICU mortality, in-hospital mortality, and up to one year mortality.
Laboratory data were recorded starting 24 h before ECMO initiation and daily during the whole support period until a maximum of 14 days. The time frame was chosen based on the median duration of ECMO support (6 days) and percentage of patients (95%) having ECMO terminated within the 14 days.
Two authors independently checked each electronic medical record and extracted the data in a predesigned case report form.
This retrospective study was approved by the Ethics Committee of the Medical University of Innsbruck, Austria (#1274/2019).

Anticoagulation Protocol
Management of elective patients under antithrombotic drugs was performed according to the national and international guidelines [11,12]. All patients receiving elective surgical procedures and on a dual platelet therapy continued using acetylsalicylic acid while P2Y12 inhibitors were paused before intervention. If Multiplate ® (Roche, Basel, Switzerland) intraoperatively showed presence of significant platelet blockade and satisfactory haemostasis could not be achieved, platelet concentrates were transfused during surgery.
Patients with new oral anticoagulants underwent elective procedures only with blood levels below threshold. In case of emergency procedures, the effects of these drugs were antagonised if possible. All patients with suspected acute coronary syndrome received loading with acetylsalicylic acid and 70 IU/kg unfractionated heparin (UFH) during prehospital treatment. P2Y12 inhibitors were administered-after consultation with the cardiologist on duty-either prehospital or immediately before percutaneous coronary intervention. Due to local protocols, none of our patients received ticagrelor.
Unfractionated heparin-coated ECMO circuits were used, and all patients received a loading dose of 50-100 IU/kg UFH before cannulation for ECMO, if they were not already on cardiopulmonary bypass. After ECMO initiation, anticoagulation was adapted according to the low-range activated clotting time (LR-ACT), aPTT, CT INTEM in the ROTEM ® , and blood drug concentration or an anti-factor Xa assay activity. Patients with still inadequate haemostasis received substitution of coagulation factors in order to achieve adequate coagulation for ECMO support.
Anticoagulation of patients receiving ECMO support was conducted according to the local standard operating procedure protocol and the ELSO Anticoagulation Guideline [13]. Unfractionated heparin (initiated with 5-20 IU/kg/hour, with the targeted aPTT), was used as the first choice for anticoagulation. In the case of inadequate anticoagulation with UFH or suspected or proven heparin-induced thrombocytopenia type 2 (HIT 2), anticoagulation was changed to argatroban. In rare cases of increased coagulation (e.g., enhanced turn-over of concomitant renal replacement filters), epoprostenol (4 ng/kg/min) was used. In presence of severe coagulopathy, anticoagulation was paused. Continuous administration of UFH or argatroban was titrated to an aPTT value of 50-70 s and argatroban blood concentration of 0.3-0.5 µg/mL, respectively. Directly after anticoagulation initiation, monitoring was performed every 30 min until a stable aPTT or argatroban blood concentration was reached. After reaching a steady state, routine controls were performed every six hours and after every change of dosing until reaching stable conditions again. With every blood gas analysis, point-of-care ACT was measured, and aPTT was repeated if deteriorated.

Objectives and Outcomes
The primary endpoint of our work was identification of potential risk factors and predictors of haemorrhage during ECMO support. Secondary endpoints included the comparison (bleeding versus no bleeding event) of demographic and clinical characteristics as well as the incidence and type of adverse events during ECMO support. Finally, we evaluated the effect of haemorrhage on mortality and reported on subgroup analyses based on the type of ECMO, presence of surgical intervention, and major haemorrhage.
Reported outcomes comprised bleeding, thromboembolic events, sepsis, and mortality. Haemorrhagic complications were only observed during the period of ECMO support. Thereafter, bleeding events were considered as not being ECMO-related. Haemorrhagic events were defined as major or minor, according to the ELSO definition [13]. A major bleeding event was defined as clinically overt bleeding associated with a haemoglobin decrease of at least 2 g/dL over 24 h or administration of two or more red blood cell concentrate units over the same period [13]. Any pulmonary or retroperitoneal bleeding involving the central nervous system or requiring surgical intervention was also considered as major bleeding. Minor bleeding events were defined as any other noticeable bleeding [13]. Severe coagulopathy was defined as clinically significant bleeding with impaired clot formation and need for blood product substitution. We recorded only the date of the first bleeding event, in case if multiple bleeding events or sources appeared. Computed tomography was performed before ECMO support initiation (except in the case of ECMO implantation under resuscitation). Additional imaging for bleeding or thromboembolic events were performed if clinically indicated, e.g., in clinically suspected acute intracranial pathology.
Information on thromboembolic events (date of identification, localisation, and type of thromboembolic complication) was gathered from the medical documentation and radiological reports during the ECMO support and within the two weeks after ECMO termination. The retrospective collection of data on the presence of thrombosis was only possible if radiological investigation was performed during the whole observation period. Thromboembolic events were confirmed using computed tomography or ultrasound. Thrombosis was stratified into central arterial and venous (heart, pulmonary artery, and aorta) or peripheral thrombus formation (all peripheral veins and arteries), embolization (i.e., ischaemic stroke), ECMO cannula or central vascular catheters, and mixed arterial and venous thrombosis.
We recorded data on the date of death, and therefore calculated mortality in different periods. The information on patient death was collected from the hospital records.

Statistical Analyses
A statistician not involved in the study procedures or patient assessment performed the statistical analyses using SPSS (Version 22.0. Released 2013, Armonk, NY, USA: IBM Corp.) and R version 4.0.2 (free software for statistical computing and graphics-R Core Team 2020: a language and environment for statistical computing; R Foundation for Statistical Computing, Vienna, Austria). All statistical assessments were two-sided, and a significance level of 0.05 was used. Depending on the type of variables and the normality of the distribution, results were presented as frequency (percent), median (range), or mean with standard deviation. For parametric data, independent samples t-test was used, and for numeric data with non-normal distribution and ordinal data, Mann-Whitney U test was used. To test differences between nominal data (frequencies), chi-square test and Fisher's exact test were used. In the univariate Cox proportional hazards model, we analysed the effect of each potential predictor of bleeding, and all significant covariates were assessed for the multivariate model. The significance level for the model was set to 0.1. The Kaplan-Meier method was used to estimate the time to bleeding event during ECMO support. To estimate the variability of observed laboratory values over the time (from ECMO support initiation until the event of interest), the coefficient of variation was calculated.

Results
During the observed period, 613 patients needed ECMO support. After screening all electronic medical charts, 415 patients met inclusion criteria, with 321 patients showing complete data sets. More than one third of included patients (123, 38%) experienced haemorrhage.

Factors Associated with Bleeding Events
Patients with bleeding events had higher SAPS III (69 vs. 65, p = 0.023) and SOFA (13 vs. 12, p = 0.005) scores, lower inflammatory markers (C-reactive protein (5.5 vs. 7.7, p = 0.002) and procalcitonin (3.4 vs. 7.0, p < 0.001)), and a prolonged aPTT (58 vs. 56.5). ECMO support was needed longer (7 vs. 6 days), and fewer patients were successfully weaned from support or bridged to another support (Table 2). Furthermore, these patients received more blood products and coagulation factors (Table 3). Finally, patients with bleeding events had a higher mortality in all registered periods with cardiac failure being the main cause of death within 90 days (41, 33%), followed by multiple organ failure (38, 31%) and brain death (26, 21%) ( Figure 3 and Table S1).   In the analysis of laboratory parameter fluctuations over the time on ECMO, higher variability of prothrombin time (HR = 1.02, p = 0.016) was associated with bleeding (Table S2).
Following univariate analyses (Table S3), SAPS III score, CRP, Fibrinogen, and aPTT were included in a multivariate analysis. The final model consisted of four variables, as the rest were excluded due to multicollinearity or large numbers of missing values. A higher SAPS III score (HR = 1.01; p = 0.047), lower values of CRP (HR = 0.96; p = 0.005), and a prolonged aPTT (HR = 1.01; p = 0.007) were associated with increased risk of bleeding (Table 4). In a further multivariate analysis, SOFA score and procalcitonin were included in the model (Table S4).

Subgroup Analyses
Comparison of critically ill patients regarding the ECMO configuration (va-ECMO or vv-ECMO) is shown in Tables S5-S7. In short, patients with respiratory failure were significantly younger (48 vs. 60, p < 0.001), less often resuscitated before support initiation (10% vs. 22%, p = 0.017), and had a longer ICU stay (21 vs. 17 days, p = 0.010). There was no difference in disease severity scores or ICU mortality. The univariate analysis of va-ECMO identified the same risk factors for bleeding as in the main analysis (Table S8), opposite to the vv-ECMO (Table S9). The multivariate model identified lower C-reactive protein (HR = 0.96; p = 0.020) and prolonged aPTT (HR = 1.01; p = 0.029) to be associated with an increased risk of bleeding (Table S10).
Moreover, comparing patients with major bleeding to those without or with only minor bleeding events revealed higher SAPS III (69 vs. 66) and SOFA (13 vs. 12, p = 0.004) scores, ICU mortality (53% vs. 32%, p = 0.002), and more frequent coagulopathy and sepsis in patients with major bleeding (Table S13). In the Cox model, a higher SOFA score (HR = 1.08; p = 0.021) and presence of surgical intervention (HR = 1.88; p = 0.026) were associated with increased risk of bleeding (Table S14).
Finally, to analyse the influence of bleeding on the first ECMO day (to exclude direct postoperative bleeding), or ECMO duration less than two days on outcomes, a subgroup analysis was performed. However, this analysis did not show any significant differences as compared to the main analysis (data not shown).

Discussion
In this retrospective study from a Central European university centre, we reported on patient characteristics, risk factors for haemorrhage, and clinically important outcomes of critically ill patients needing ECMO support. Almost 40% of our patients experienced bleeding events during ECMO. These patients (1) were sicker, (2) spent more time on extracorporeal life support, and (3) had an increased mortality. Moreover, a prolonged aPTT and a weaker inflammatory response were associated with an increased bleeding rate. Finally, in our cohort, almost two thirds of the reasonably young patients survived utilising extracorporeal life support, with 61% of patients being discharged from the hospital.
Although ECMO support is lifesaving in many circumstances, complications are common and associated with the potential of permanent injury or even death. In our study, haemorrhage was the most frequent adverse event (38%), which is in line with a meta-analysis of 1763 patients reporting any kind of haemorrhage (40%) [14]. Major haemorrhage, as defined by the ELSO, occurred in 19%, and minor haemorrhage occurred in 20% of our patients. The cannulation and surgical area were the most common sites of bleeding, which is in line with current data [15][16][17]. From the 14 patients with an intracranial haemorrhage, 50% did not survive, which is comparable to findings from centres in the United Kingdom [8,18].
The overall survival to discharge was 61%, with a one year survival of 59% being in the higher share of the reported range (34-67%) [8,10,[14][15][16][17]19]. Patients with haemorrhage had a higher mortality in all registered periods, corresponding to the literature [8,15,16] and being the highest in the case of intrapulmonary or intracranial bleeding.

Factors Associated with Haemorrhage
Several factors may be causative for haemorrhages during ECMO support, including vessel damage at the cannula or surgical sites, greater surgical complexity, or longer cardiopulmonary bypass times. Furthermore, coagulopathy due to reduced coagulation factors, thrombocytopenia, platelets dysfunction, acquired von Willebrand syndrome, or increased fibrinolysis may contribute to bleeding occurrence. Even fungal pneumonia or centrifugal ECMO pump have been shown to be associated with increased risk for haemorrhage [16,[20][21][22][23][24][25]. Only some of these factors are modifiable, and whether their impact can be influenced by adaption of systemic anticoagulation or special therapeutic regime remains unclear [22].
In our study, we were able to identify a number of risk factors for haemorrhage during ECMO support. Higher SAPS III and SOFA scores, decreased clotting capability, and lower inflammation markers increased the risk of haemorrhage in the univariate analysis. Finally, the multivariate Cox regression model identified a higher SAPS III score (HR = 1.01, for every increase in one unit of measurement, hazard ratio increased 1%), prolonged aPTT (HR = 1.01), lower CRP (HR = 0.96), and procalcitonin (HR = 0.99) as predictors of bleeding in our retrospective study.
Prolonged aPTT has been inconsistently reported as a risk factor for bleeding during ECMO support [8,15,16,18], and our findings confirm earlier reports from Aubron et al. [16]. However, the ideal parameter for anticoagulation monitoring is still a matter of discussion, as different factors can influence anticoagulation monitoring in critically ill patients [13,26,27].
Interestingly, patients with a weaker inflammatory response had a higher risk of bleeding, which has not been reported in the literature. The association of inflammation and thrombosis (thromboinflammation) is well-established and discussed extensively in COVID-19 patients [28,29]. However, literature on the impact of inflammation on bleeding is still missing. Clearly, the activation of host defence results in activation of coagulation and a prothrombotic state [30][31][32][33][34]. The increased levels of both CRP and procalcitonin are allied with coagulation activation and consequent thrombosis [35,36].
Surgical trauma and exposure of blood to the artificial surface of the ECMO initiate and propagate the inflammatory response [37]. Inflammation further initiates clotting, decreases the activity of natural anticoagulant mechanisms, and impairs the fibrinolytic system in many different ways. Acute inflammation leads to an extensive elevation of the acute phase proteins and activation of diverse molecules. Within these processes, endotoxin, IL-1β, tumour necrosis factor-α (TNF-α), and neutrophil elastase reduce thrombomodulin on endothelial cell surfaces [38,39]. Furthermore, endothelial cell leucocyte adhesion molecules (P-selectin and E-selectin) are expressed on endothelial or platelet surfaces. Tissue factor is induced by endotoxin, TNF-α, or CD40 ligand on the cell surface of leucocytes, particularly monocytes, and it further binds factor VIIa, activates factor X, and forms complexes with factor Va to generate thrombin [40]. Additionally, inflammation decreases protein C levels, and inflammatory mediators increase the production of new and even more thrombogenic platelets [41,42]. This leads to the release of ultra-large von-Willebrand factor multimers and inhibits its cleavage by ADAMTS13 [43]. Increased CRP levels facilitate monocyteendothelial cell interactions and promote plasminogen activator inhibitor-1 (PAI-1) and tissue factor formation with subsequent complement activation [44][45][46]; antithrombin is consumed and/or inactivated, and the concentration of vascular heparin-like molecules can be reduced by inflammatory cytokines and neutrophil activation products [47][48][49]. Finally, the role of the fibrinolytic system in ECMO patients is still not well-investigated. Recent data has shown that increased fibrinolytic activity during ECMO support is associated with an increased risk of bleeding [25,50]. In contrast, McVeen et al. reported on the normalisation of increased fibrinolytic enzymes within a few days after ECMO initiation [51]. In our study, we did not find an association of InTEM lysis index as measured by ROTEM ® and bleeding; however, further research in this area is warranted.
Given the above, the systemic anticoagulation of critically ill patients is complex and difficult, especially if done by UFH, a mixture of heterogeneous glycosaminoglycans of different molecular weights. Heparin's polyanionic nature limits its interaction specificity with antithrombin, forming the heparin-antithrombin complex, which is a direct inhibitor of thrombin and factor Xa [52]. Hence, the steady-state plasma concentration is in no direct linear relation to dosage under continuous infusion [53]. Hyperinflammation can lead to a limitation of the anticoagulant effect of UFH by increasing the heparin-binding acute phase proteins, including factor VIII and fibrinogen, or decreasing the antithrombin levels [47][48][49]. Traditionally, anticoagulation monitoring is based on aPTT assays, despite the known heterogeneous results and the risk of interference in critically ill patients [26,48,54].
We hypothesised that in cases of a weak inflammatory response, prothrombotic activation by contact with artificial surfaces in the extracorporeal circuit would be less pronounced.
This would lead to an increased susceptibility of bleeding under standard anticoagulation protocols which do not consider inflammatory response but rely only on aPTT, anti-factor Xa, or the blood concentration of the anticoagulant used. Lastly, this was limited by the fact that the ideal parameter for anticoagulation monitoring is still missing.
Finally, the extent of the inflammatory response to the ECMO circuit showed a high interpatient variability without a physiological rationale [37]. Hyperinflammation has a major impact on coagulation and heparinisation, and our hypothesis was that the unintended excessive anticoagulation in case of "non-hyperinflammation" would lead to the outcome-relevant bleeding.

Further Directions
Future research on risk factors for bleeding in regard to surgical techniques, ECMO systems, anticoagulation possibilities, and patient factors is warranted. The ongoing technological development of ECMO circuits, through different coating materials or modifications of extracorporeal circulation systems, resulted in reduced levels of thrombin [37]. Research on ECMO-induced inflammation and its association with clinical outcomes may lead to novel anti-inflammatory therapies with a more stable anticoagulation [55][56][57]. The ideal anticoagulant, with reduced or even eliminated risk of thrombosis and absent haemorrhage risk, is still not available. Emerging preclinical data suggest that antibodies targeting factor XI and XII may have improved safety and efficacy, but data in humans are still missing [55][56][57]. Whilst it is certainly difficult to isolate the relative contribution of specific patient factors to bleeding during ECMO support, failure to do so can result in a missed opportunity for intervention. Patient factors were extensively investigated, but due to the diversity in reporting and missing of guidelines on minimal reporting criteria for ECMO complications, a comparison of studies was challenging.
Further studies should investigate whether the consideration of the inflammatory response can farther personalise anticoagulation and reduce the incidence of bleeding events during ECMO. This is even more interesting given the rise of ECMOs in coronavirus disease 2019 (COVID-19) patients with a tendency towards a hypercoagulatory state. The ongoing prospective trials (ECLS-SHOCK, NCT03637205; EUROSHOCK, NCT03813134; ECMO-CS, NCT02301819; and ANCHOR, NCT04184635) aim to provide prospective information on the benefits of extracorporeal circulation in cardiac shock and respiratory failure. However, the gap between the association of inflammation and bleeding is still to be closed. We therefore recommend cautious monitoring (as per existing protocols) and adaptation of anticoagulation in patients with a weak inflammatory response until further evidence is available.

Limitations
This study was limited in several aspects. We reported on a mixed population of critically ill, all with a common attribute of a very high mortality risk, where ECMO support was initiated as the last resort. Due to the retrospective nature, selection bias could not be excluded. Furthermore, our analyses may be confounded by undisclosed or undiscovered factors such as the plasma levels of laboratory parameters being diluted by other drugs or other similar factors. However, as all of our patients experienced a distinct decrease in the majority of parameters, the chance for a significant confounding factor in only one of the groups should be rather small. We used the widely accepted ELSO bleeding definition to classify bleeding [13], but some complications may have been overlooked or missed. Given the liberal approach to diagnostic imaging and post-mortem examinations, this chance seems rather small. Although our study involved a quite large cohort of patients as compared to the literature, larger samples and prospective studies are needed to further elucidate the interaction of inflammation and haemostasis during ECMO support.

Conclusions
To the best of our knowledge, this was the largest European study investigating risk factors for bleeding events during ECMO support. Haemorrhage remains the most frequent and serious ECMO complication with an increased mortality. We confirmed previous findings of a prolonged aPTT and disease severity as risk factors for haemorrhage. For the first time, we reported on the association of inflammation and bleeding during ECMO support, generating a new hypothesis and warranting further research. Finally, based on our findings, we recommend stricter monitoring of anticoagulation especially in patients without hyperinflammation, until the further evidence is available.