We set out to investigate the incidence and predictors of thrombosis and thromboembolism following ECMO therapy. Sixty-three patients, treated on ECMO in our center, were included in this analysis. The main findings of this retrospective analysis are: (1) at a rate of 46.1 %, thrombotic and thromboembolic events are frequent, (2) time on ECMO and level of anticoagulation influence the incidence of thrombotic and thromboembolic events and (3) these events tend to influence the outcome of these patients. Patient with a bicaval double-lumen cannula had no higher incidence of VT/VTE than other patients.
Critically ill patients are per se vulnerable for thrombotic and thromboembolic events. Kaplan et al.  published a VTE incidence of 37.2 % in ICU patients during severe sepsis and septic shock. Patients with viral ARDS are also at risk as suggested by data derived from the pandemic H1N1 . Regarding non-infectious etiologies, pulmonary fibrosis is, for example, associated with a significantly elevated risk of thromboembolic disease .
Over and above these baseline risks, there are some certain characteristics for patients treated on ECMO. During extracorporeal assist, modification of blood composition as well as foreign surface and pump-mediated coagulopathy is one part of the problem. Cannulation is another. Starting with vascular injury at cannula entry site, the cannula splints the vessel until central areas are reached. This generates long areas of low flow and stasis along the cannula and thus provides an ideal location for thrombus formation. Possibly, additional indwelling catheters further obstruct the vessel. Taken together, these ICU patients are at high risk of VT/VTE.
However, there are still few data concerning the incidence of VT following ECMO and the aspect of VTE following respiratory ECMO therapy has not been investigated so far.
Cooper et al.  recently reported a retrospective analysis concerning the incidence of DVT in survivors of respiratory ECMO diagnosed by postdecannulation venous Doppler ultrasound. They used a 25 F multistage access cannula and a 23 F single-stage return cannula (Biomedicus, Mineapolis, MN) as standard cannulation configuration and a systemic anticoagulation with heparin aiming at a PTT of 1.5–2-times normal values. Surprisingly, DVT was found in merely 13/72 (18 %) of cases. This low incidence might be explained by a lack of reporting regarding V. cava thrombosis. Additionally, with a reported survival rate of 79 %, this cohort might be different from other cohorts.
We observed VT/VTE in 30/63 patients (47.6 %) that were mainly femoral/jugular cannulated. Due to the fact that thrombosis was mainly cannula associated, the IJV (46.7 %) and the IVC (50 %) were the most affected veins. A recent publication by Shafi et al.  reported upper-extremity deep vein thrombosis in 80 % of cases in a case series of 10 patients who were put on vvECMO with a dual-lumen cannula. In our present analysis, the rate of thrombosis associated with dual-lumen cannulas was not significantly different from other cannulation types and occurred in 4/11 cases (36.4 %).
To the best of our knowledge, this analysis is the first showing a high incidence of IVC thrombosis in ECMO patients. IVC thrombosis might account for roughly 2 % of lower limb deep vein thrombosis (DVT) . IVC thrombosis harbors a 12–30 % risk of PE and a relevant risk of long-term complications such as chronic venous insufficiency, venous congestion and postthrombotic syndrome in up to 20 % of non-resolved thrombosis [18, 19]. PE as a major complication of VT was consequently diagnosed in 7/63 cases (11 %). We found one case of suspected PE directly following cannula removal and two cases of incidental PE diagnosed by CT scan. In most cases, VTs at cannulation sites are adherent to the venous wall and do not cause clinical symptoms. Up to now, there is no relevant literature on long-term complications following cannula-associated VT.
One of these patients died 24 days after weaning from device due to multiseptic organ failure following LTX and underwent postmortem examination.
The autopsy did not report VT or VTE possibly due to thrombus dissolution under prolonged anticoagulation. PE was consequently diagnosed in 5/21 deceased patients (24 %). Pulmonary embolism was only systematically evaluated for in deceased patients, so no adequate incidence for PE in the whole population can be given from this data. Reasons for PE in deceased patients might also be associated with circumstances and treatment modalities prior to end of life.
The management of anticoagulation, especially in prolonged ECMO support and in patients with sepsis and septic shock, is unclear. A recent ELSO survey showed that anticoagulation policies vary widely by center . Bleeding is common in ECMO and cerebral bleeding, occurring at a rate of 4–15 %. In the majority of these patients, the outcome is deleterious [21–25]. Bleeding risk and transfusion requirements need to be balanced against VT/VTE risk, and this might likely end up in less anticoagulation. Mostly, aPTT around 50 s is recommended to minimize the risk of bleeding [4, 21, 22, 26]. This might provoke an increase in VT and VTE. We targeted a mean aPTT of 50–60 s, and our average aPTT was slightly above 50 s. The group of VT/VTE patients had a lower mean aPTT, possibly in line with the fact that they received more packed red blood cells. We assume that we missed our anticoagulation target due to bleeding complications. In fact, VT/VTE patients received more RPBC in total even though these differences did not reach statistical significance. Bleeding is often difficult to assess—a loss of 0.3 mL of blood per minute might be difficult to quantify but will lead to a loss of 480 mL per day. Additionally, there is no consensus on how to assess bleeding severity, especially if bleeding occurs at sites that are not easily accessible. Melena can be such an example; patients can lose significant amounts of blood, but quantification other than with the number of transfused RPBCs is impossible.
Our data are in line with the work of Rastan et al.  who analyzed autopsy results of patients undergoing veno-arterial ECMO support after cardiac surgery. They identified cardiac ECMO support > 2 days as an independent risk factor for systemic thromboembolic events and found systemic thromboembolic events in 36/78 patients (46.2 %) . Compared to cardiac ECMO support, respiratory ECMO runs are often much longer. Cardiac ECMO runs were 3.5 ± 3.2 days in the study of Rastan et al.  versus 22.4 ± 17.4 in our pulmonary patients study. We also observed a trend toward longer duration of support in VT/VTE patients (p = 0.040). However, VT/VTE was also detected in short ECMO runs.
Our study has several limitations, which we would like to address. Many limitations are due to the retrospective nature and the low number of patients included in this work. One major limitation is based on the fact that we did not have a standardized screening protocol. The group of ECMO survivors has not been systematically analyzed. This might lead to an underestimation of the overall incidence.
Nevertheless, it has to be assumed that VT and VTE is an important and underdiagnosed complication of respiratory ECMO support and has impact on mid- and long-term outcomes. While we met common recommendations for anticoagulation on ECMO, the VT/VTE incidence was still high. Our results tend to underestimate the true incidence of VT/VTE in ECMO patients. Hence, it might be important to aim at higher aPTT times to prevent VT/VTE in a cohort of patients with a high-risk profile for VT/VTE.