Optimizing anticoagulation during extra-corporeal membrane oxygenation: clinical and methodological considerations towards a future clinical trial

The first successful use of extracorporeal membrane oxygenation (ECMO) was reported in 1972 for acute post-traumatic respiratory failure. From the outset, the challenge of balancing the risks of bleeding and thrombosis was evident, with daily transfusions of packed red blood cells, plasma, and platelets during a 3-day ECMO run.1 Over the subsequent two decades, widespread uptake of ECMO was inhibited by technological limitations and by limited understanding of the key underlying physiological drivers of mortality such as ventilator-induced lung injury. Following publication of the CESAR trial,2 and the coincident arrival of the H1N1 influenza pandemic,3 annual adult ECMO volume rose four-fold in 2009. Publication of EOLIA further accelerated adoption,4 the COVID-19 pandemic saw a surge of VV-ECMO use, and in recent years we have also seen expansion of VA-ECMO for cardiogenic shock and broader implementation of extra-corporeal cardiopulmonary resuscitation (eCPR) in cardiac arrest.5 The core component of any ECMO intervention is the membrane oxygenator, a hollow-fibre gas-exchange device. Blood flows along one side of a semipermeable membrane while sweep gas flows on the other, allowing oxygen and carbon dioxide to diffuse across their concentration gradients, analogous in layout to a dialysis circuit. Although ongoing developments have decreased the inflammatory impact of non-native material, the blood-surface contact can nevertheless trigger coagulation and circuit thrombosis, which creates both a bleeding and clotting diathesis. To counteract this, patients typically receive systemic unfractionated heparin (UFH) titrated to protocolized targets, such as an activated partial thromboplastin time (aPTT) of 1.5-2 times the laboratory control value. In the multicenter PROTECMO observational study heparin was the most commonly used anticoagulant in ECMO patients (73% of anticoagulated ECMO days), but bleeding was common; over 50% of patients experienced at least one event, including 1.6% fatal bleeds.6 In addition, for the small number of patients with proven or suspected heparin-induced thrombocytopenia, heparin is obviously contraindicated. Bivalirudin and argatroban (direct thrombin inhibitors) are used as alternative anticoagulants, but randomized comparative data are lacking. In this issue of the Journal, Buchtele and colleagues contribute a research letter reporting a pilot single-center randomized clinical trial comparing argatroban with unfractionated heparin in patients receiving ECMO.7 Initial cannulation and stabilization was performed using unfractionated heparin, following which patients were randomized to continue unfractionated heparin or switch to an argatroban infusion. A total of 40 patients were included, 20 in each group, with approximately half being VV-ECMO and half being VA-ECMO for eCPR. They found no difference in the primary endpoint of clinically relevant bleeding or thrombosis, which occurred in 5/20 (25%) and 6/20 (30%) patients in the heparin and argatroban groups, respectively. This study demonstrates that enrolment and protocolized anticoagulation are possible, with enrolment of close to 1.5 patients per month; however, it is underpowered to demonstrate comparability, superiority, or non-inferiority. A common error is to read a nonsignificant result as proof of no difference. Small differences can rarely be excluded. Non-inferiority trials can address this but need 3 attributes to be present: 1) the treatment under consideration should exhibit therapeutic non-inferiority to the active control; 2) the treatment would exhibit therapeutic efficacy in a placebo-controlled trial if such a trial were to be performed; and 3) the treatment offers ancillary advances in safety, tolerability, cost, or convenience.8 Such trials specify a clinically acceptable margin (Δ). For example, tolerating at most a + 5% absolute increase in bleeding/thrombosis with argatroban versus heparin, and declaring non-inferiority only if the confidence interval for the treatment difference lies entirely within Δ. Because crossovers and protocol deviations are inevitable, especially when heparin intolerance prompts switching, two analyses are required: intention-to-treat (ITT), which tends to dilute differences, and per-protocol (PP), which excludes major deviations. The non-inferiority conclusion must be supported by both ITT and PP; lack of concordance raises concern.8 In this instance, the authors acknowledge the limitations of the small sample size and heterogeneous sample. We would like to elaborate on a few points for incorporation into a future larger RCT. Randomization occurred approximately one day after cannulation, which likely excluded the most critically unwell, early deaths, and severe coagulopathy. This may explain the favourable outcomes reported across both groups: weaning 33/40 (83%), in-hospital deaths 6/40 (15%), and 90-day survival 75%–metrics inconsistent with typical adult eCPR survival to discharge (∼20-30%).9 Late randomization builds immortal-time bias and suppresses event accrual. Pooling eCPR/VA with VV under a single anticoagulation endpoint mixes populations with different baseline hazards and time-at-risk. eCPR’s short runs produce unstable rate-per-ECMO-day estimates because a few events dominate small denominators. The median study-drug exposure of 5 days [IQR 3-10] is short for adult ECMO and would undermine any comparative claim, especially a non-inferiority one. In contemporary cohorts, the median run length is approximately 10 days for VV, 5 days for VA, and 3 days for eCPR.5 Fewer drug-days mean fewer opportunities to detect true differences in bleeding or thrombosis. Importantly, oxygenator/circuit thrombosis often clusters after Days 5-7, so brief runs may not detect differences in anticoagulation-related events. The resulting effect is that signal is suppressed and apparent similarity is inflated. We believe that a multicenter randomized-controlled trial comparing argatroban to heparin in ECMO would be of significant value. Given the increased cost of argatroban and the long clinical experience with unfractionated heparin, a clinical superiority framework would be appropriate as practice would unlikely change with non-inferiority alone. To maximize success, we would advise future trialists to consider enrolling a much larger number of patients across diverse centres and: Randomize as soon as possible after cannulation Stratify by configuration (VV vs VA), and eCPR status–perhaps even running technically distinct parallel trials in these 3 populations Control important co-interventions including transfusion thresholds, oxygenator exchange criteria, and imaging schedules Conduct blinded central adjudication for clinically important bleeding or thrombosis Anchor inference to a fixed Day-28, patient-level composite (major bleed, clinically significant thrombosis, or death) In conclusion, we congratulate the authors on their successful pilot study, and we recognize that much work remains to address this important clinical question. Supplementary material is available at American Journal of Respiratory and Critical Care Medicine online. Please see the ICMJE disclosure forms, which have been provided as supplementary material.