Clinical correlates for major bleeding during veno-arterial extracorporeal membrane oxygenation support: a time-dependent perspective

Abstract

Background

Major bleeding is a common and serious complication of veno-arterial extracorporeal membrane oxygenation (VA ECMO). Current perspectives on bleeding risk fail to reflect dynamic changes over time in bleeding incidence and risk factors for bleeding.

Methods

All adults receiving VA ECMO (2016–2023) for cardiac failure in 6 participating Dutch centers were included in a prospective cohort. Major bleeding was defined as bleeding requiring surgical or radiological intervention and/or transfusion of > 3 red blood cell units within 24 h. Bleeding cases were matched up to 4 controls who had not (yet) bled at that timepoint. Generalized additive models were applied to visualize trajectories of coagulation, inflammatory, and perfusion markers, exploring temporal differences between bleeding cases and matched controls.

Results

Among 558 VA ECMO supported patients, 232 (42%) experienced ≥ 1 major bleeding event, totaling 342 episodes (incidence rate 112 per 1000 days at risk). Most bleeding events occurred early (< 48 h; n = 190, 56%), primarily at the cannulation site (37%) and intrathoracically (38%). Fewer events occurred later (> 48 h; n = 152, 44%), predominantly intrathoracically (53%). Female sex and lower body weight (both p < 0.01) were associated with bleeding. Compared with controls, patients who developed bleeding showed increased activated partial thromboplastin time, prothrombin time, heparin ratio, anti-factor Xa and international normalized ratio preceding bleeding, whereas platelet count, hemoglobin and leukocytes were lower. Bleeding was associated with higher intensive care mortality (HR 1.39, 95% CI 1.09–1.76, p < 0.01).

Conclusion

A severe major bleeding developed in nearly half of all patients and was associated with an increased mortality risk. The trajectories of coagulation and inflammatory markers observed in patients with versus without bleeding events underscore the dynamic nature of bleeding risk and highlight future opportunities for dynamic bleeding risk prediction during VA ECMO.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13054-026-05856-7.

Introduction

Veno-arterial extracorporeal membrane oxygenation (VA-ECMO) offers rapid hemodynamic stabilization for patients with severe and acute cardiac failure [1, 2]. However, its benefits seem offset by a substantial risk of complications [3–6].

Bleeding complications affect up to half of VA ECMO patients, ranging from minor cannulation-site oozing to life-threatening gastrointestinal/intracranial hemorrhage [7–10]. Their etiology seems multifactorial, including anticoagulation, shear stress–induced acquired von Willebrand syndrome and platelet dysfunction, and systemic inflammation driving endothelial damage and coagulopathy in critical illness [4, 7, 11, 12]. Despite this complex and multifactorial etiology, bleeding risk management in most centers still largely relies on fixed reference values for activated partial thromboplastin time (aPTT) and/or anti-factor Xa targets [5, 13].

Previous studies have described bleeding incidence and identified risk factors; however, these analyses were cross-sectional and lacked granular longitudinal data [5, 13–17]. This limits the understanding of temporal risk patterns and the development of more personalized strategies. To address this, we aimed to [1] quantify temporal trends in bleeding incidence; [2] identify patient-specific risk factors; and [3] characterize time-varying risk factors by analyzing markers of coagulation and inflammation around bleeding events.

Methods

Study design and population

We used patient data from the PREdiCtIon of weanability, Survival and functional outcomEs after ECLS (PRECISE-ECLS) cohort (NCT05444764) which was established to investigate the clinical course and long-term outcomes of adult patients requiring VA ECMO for acute and severe cardiac failure due to a variety of causes including post-cardiotomy cardiogenic shock, myocarditis, acute myocardial infarction, and refractory cardiac arrest. Patients were enrolled between 2016 and 2023 across six centers in the Netherlands (see Appendix Table 1). The institutional ethics committee approved an opt-out consent procedure (reference 21–604/C).

Baseline data included patient demographics (age, sex, body mass index), relevant comorbidities (such as hypertension and diabetes), and pre-ECMO clinical variables. These comprised illness severity scores (Acute Physiology and Chronic Health Evaluation (APACHE) IV scores), indication for VA ECMO, pre-ECMO complications (such as acute kidney injury and infection), cannulation method (percutaneous or surgical), and use of concomitant mechanical circulatory support (Impella^® or intra-aortic balloon pump). Time-varying data included daily use of anticoagulant and antiplatelet agents (unfractionated heparin (UFH), low molecular weight heparin (LMWH), argatroban, clopidogrel, aspirin and more potent P2Y12-inhibitors (ticagrelor and prasugrel)), laboratory markers of coagulation (aPTT (seconds)), prothrombin time (PT) (seconds), heparin ratio (patient aPTT divided by the normal reference value), international normalized ratio (INR), anti-factor Xa (IU/mL), hemoglobin (Hb) (mmol/L), platelet count (/mm³), D-dimers (mg/L), fibrinogen (g/L), ionized calcium (mmol/L), inflammation (C-Reactive Protein (CRP) (mg/L), leukocytes (/mm³) and lactate (mmol/L). Laboratory data were extracted directly from electronic medical records systems (MetaVision^® (iMDsoft, Tel Aviv, Israel) and HiX^® (ChipSoft, Amsterdam, the Netherlands) in three participating centers (Leiden University Medical Center, University Medical Center Utrecht, and Erasmus Medical Center).

Bleeding complications were recorded using standardized definitions adapted from the Extracorporeal Life Support Organization (ELSO) guidelines [18]. Major bleeding was defined as bleeding requiring surgical or radiologic intervention and/or transfusion of > 3 RBC units within 24 h (2), closely corresponding to the bleeding academic research consortium (BARC) type 3 definition (overt bleeding with ≥ 3 g/dL hemoglobin drop, transfusion with overt bleeding, or bleeding requiring intervention) [19]. Because bleeding may develop insidiously [20], complicating the identification of the exact time of onset, events were categorized into six-hour intervals. This interval length was chosen based on a pilot analysis, which showed that six-hour periods provided the most accurate temporal resolution for identifying bleeding onset. Bleeding sites included cannulation sites, surgical wounds, catheters, pulmonary, cardiac, gastrointestinal, and abdominal sites.

Hemostatic practice during VA ECMO support

Patients undergoing VA ECMO support were treated according to local hospital protocols and guidelines for anticoagulation and hemostatic management (Appendix B). Systemic anticoagulation was generally maintained with UFH (aPTT 50–70 s or heparin ratio 1.7–2.0) and/or anti-factor Xa (0.3–0.7 IU/mL), withheld or adjusted in case of bleeding risk (as assessed by attending physicians). Argatroban was used if heparin-induced thrombocytopenia developed. Red blood cell transfusion targeted Hb > 5.0 mmol/L (8.05 g/dL) and platelet transfusion was considered indicated when platelet count was < 50 × 10⁹/L. Coagulation factor concentrates (e.g., fibrinogen), antifibrinolytic agents and protamine were administered when deemed appropriate by the attending physician. ECMO membranes were replaced in case of circuit dysfunction or progressive consumptive coagulopathy, typically reflected by declining fibrinogen levels, and increasing D-dimer levels.

Statistical analysis

Continuous variables were summarized as mean ± SD or median (IQR), and categorical variables as counts and percentages. Associations between bleeding events and intensive care unit (ICU) mortality were assessed using a time-dependent Cox model with the bleeding event as a time-varying covariate, and survival was illustrated using Kaplan–Meier curves. Baseline characteristics were compared using Student’s t-test or Mann–Whitney U test for continuous variables and the Chi-square test for categorical variables.

Each bleeding case was matched to up to four controls who underwent an equivalent duration (+/- 6 h) of VA ECMO support but had not (yet) experienced a bleeding event. This ratio was chosen because adding more than four controls per case yields minimal further gain in precision [21]. Patients who developed a bleeding event later on could also serve as controls until 24 h before their first event to avoid index-event bias. For each matched set, laboratory values were aligned relative to the bleeding time window of the case—covering 48 h before and after the event—to establish a common timeline. Marker trajectories were then visualized using smoothed trend lines from generalized additive mixed models implemented in the mgcv package in R [22] with natural splines (n = 4) and random effects per patient and matching group to illustrate marker dynamics relative to first bleeding events. Due to the high incidence of bleeding early during ECMO support, sensitivity analyses were performed to plot trajectories separately for early (< 48 h) and late (≥ 48 h) bleeding events. Additionally, daily use of anticoagulants during ECMO and their use around bleeding events in cases and controls were visualized with bar charts. A p-value < 0.05 was considered statistically significant. Analyses were conducted using “R” (version 4.2.2).

Results

A total of 558 patients were included, with a median age of 58 years [IQR 48–68], 175 (31%) females, and a median APACHE IV score of 87 [IQR 65–114]. Among the 558 patients, 232 (42%) experienced at least one major bleeding episode, accounting for a total of 342 bleeding events and an overall incidence rate of 112 per 1000 ECMO-days. Most events occurred early: 190 within the first 48 h after initiation of VA ECMO (188 per 1000 ECMO-days), mainly at the cannulation site (n = 70, 37%) and intrathoracic locations (n = 71, 38%). In contrast, 152 bleeding events occurred later (> 48 h) at a lower rate (68 per 1000 ECMO-days) and were most often located intrathoracically (n = 81, 53%) (Fig. 1). Eight patients experienced intracranial hemorrhage, seven of whom did not survive (these events are not classified as major bleeding according to the ELSO criteria and were therefore not included in the analysis). Bleeding was associated with a significantly higher ICU mortality in time-dependent Cox analysis (Hazard Ratio (HR) 1.39, 95% CI (1.09–1.76), p = 0.008, Fig. 2).

Fig. 1.

Incidence and locations of bleeding complications during VA ECMO. GI = Gastro-Intestinal

Fig. 2.

Kaplan–Meier curve in bleeding vs. non-bleeding VA ECMO patients

Patients with bleeding complications were more often female (38% vs. 26%, p = 0.004) and had a lower body weight (80 [72–90] vs. 85 [75–99] kg, p = 0.003) compared to those who did not develop a bleeding complication (Table 1). VA ECMO run characteristics were comparable between groups (Table 2). When stratified by surgical status (postcardiotomy or post–heart/lung transplantation vs. non-surgical indications), distinct patterns in bleeding locations emerged: intrathoracic bleedings (hematothorax, intrapulmonary bleeding or tamponade) predominated after surgery whereas cannulation-site bleedings were more common in non-surgical patients. In surgical patients, a second peak in bleeding incidence was observed around day 5, mainly due to cardiac tamponade (Appendix Figs. 1 and 2).

Table 1.

Baseline characteristics

Characteristic	Non-bleeding, N = 326	Bleeding, N = 232	p-value1
Age, Median (IQR)	58 (48, 68)	59 (49, 67)	0.8
Weight, Median (IQR)	85 (75, 99)	80 (72, 90)	0.003
BMI, Median (IQR)	27 (24, 30)	26 (24, 29)	0.3
APACHE IV score, Median (IQR)	87 (62, 112)	87 (68, 114)	0.3
Female sex, n (%)	86 (26)	89 (38)	0.004
Hypertension, n (%)	93 (29)	73 (32)	0.3
Myocardial infarction, n (%)	58 (18)	40 (17)	0.2
Diabetes mellitus, n (%)	60 (18)	33 (14)	0.3
PCI, n (%)	44 (13)	39 (17)	0.1
Pacemaker, n (%)	46 (15)	24 (11)	0.2
Chronic kidney disease, n (%)	23 (7)	16 (7)	0.9
CABG, n (%)	16 (5)	13 (6)	0.2
Anticoagulation use at home, n (%) Aspirin, n (%) Acenocoumarol, n (%) DOAC, n (%) Clopidogrel, n (%) Fenprocoumon, n (%) Ticagrelor, n (%)	124 (39) 60 (18) 30 (9) 24 (7) 21 (6) 13 (4) 6 (2)	86 (40) 39 (16) 14 (6) 20 (8) 15 (6) 7 (3) 7 (3)	0.2

¹Wilcoxon rank sum test; Pearson’s Chi-squared test. BMI, body mass index, APACHE IV, Acute Physiology and Chronic Health Evaluation IV, PCI, percutaneous coronary intervention, CABG, coronary artery bypass grafting, DOAC = Direct Oral Anticoagulants

Table 2.

Pre-ECMO characteristics

Characteristic	Non-bleeding, N = 326	Bleeding, N = 232	p-value1
Cannula placement, n (%)			0.3
Percutaneous	219 (67)	146 (63)
Surgical	107 (34)	86 (38)
Indication, n (%)			0.6
ECPR	100 (31)	55 (24)
Post-cardiotomy	97 (30)	91 (39)
Myocardial infarction	41 (13)	28 (12)
Cardiomyopathy	39 (12)	26 (11)
Other	27 (8)	18 (8)
Pulmonary embolism	12 (4)	9 (4)
LVAD dysfunction	8 (3)	4 (2)
ECMO operator, n (%)			0.4
Cardiothoracic surgeon	148 (47)	122 (54)
Interventional cardiologist	108 (34)	68 (30)
ICU doctor	45 (14)	26 (12)
MMT	13 (4)	9 (4)
IABP, n (%)	98 (30)	54 (23)	0.09
Impella, n (%)	19 (6)	23 (10)	0.1
Pre-ECMO cardiac arrest, n (%)	154 (47)	91 (40)	0.1
Pre-ECMO infection, n (%)	29 (9)	23 (10)	0.8
Pre-ECMO AKI, n (%)	39 (12)	34 (15)	0.4
Pre-ECMO CVVH, n (%)	13 (4)	9 (4)	> 0.9

¹Pearson’s Chi-squared test. ECPR, extracorporeal cardiopulmonary resuscitation, LVAD, Left Ventricular Assist Device, ICU, intensive care unit, MMT, mobile medical team, IABP, intra-aortic balloon pump, AKI, Acute Kidney Injury, CVVH, Continuous Veno Venous Haemofiltration

Time-varying markers of coagulation and inflammation during VA ECMO were available in 441 (79%) patients (Appendix Table 2). Compared with VA ECMO support duration-matched control periods, aPTT, PT, heparin ratio, anti-factor Xa and INR levels significantly increased over the 48 h preceding the bleeding event, followed by a subsequent decline. In contrast, platelet count, hemoglobin, and leukocyte levels were lower surrounding bleeding events (Fig. 3). These differences between bleeding and non-bleeding events were more pronounced during early (< 48 h) compared to late bleeding occurrences (Appendix Fig. 3–15).

Fig. 3.

Trajectories of time-varying markers of coagulation and inflammation surrounding first bleeding events in a matched cohort. aPTT = activated Partial Thrombin Time, PT = Prothrombin Time, INR = International normalized ratio, CRP = C-Reactive Protein

Most patients received UFH during the first week of VA ECMO, with its use increasing from approximately 60% on day 1 to nearly 90% from day 3 onward. Acetylsalicylic acid (Ascal^®) and clopidogrel (Plavix^®) use remained around 30–35% and 10–15%, respectively. LMWH, argatroban, and more potent P2Y12-inhibitors were rarely prescribed (Fig. 4). Use of anticoagulant and antiplatelet agents prior to bleeding events was similar between patients with and without bleeding. On the day of bleeding, UFH administration was frequently interrupted after which it was restarted in a substantial proportion of patients during the subsequent days (Appendix Fig. 16).

Fig. 4.

Use of anticoagulants during VA ECMO. UFH = Unfractionated Heparin, LMWH = Low Molecular Weight Heparin

In case of major bleeding complications, red blood cell transfusions were given in 270/342 events (79%), and 179 cases (52%) received surgical interventions (e.g., resternotomy for hematothorax or cardiac tamponade). A chest tube was placed in 41 cases (12%). UFH was (temporarily) stopped in 117 (34%) cases, and radiological interventions occurred in 3 events (0.9%). These interventions were not mutually exclusive.

Discussion

The high burden of bleeding during VA ECMO is well recognized, but a lack of longitudinal data limits insights into how bleeding risk and its accompanying risk factors evolve over time [5, 13, 14]. To address this, we analyzed a multicenter prospective cohort of 558 patients with detailed longitudinal data capturing both baseline and time-varying bleeding-risk factors. In this cohort, major bleeding occurred in nearly half (42%) of patients, most often within the first 48 h after VA ECMO initiation, and was associated with increased ICU mortality. Female sex, lower body weight, and distinct peri-bleeding trajectories of aPTT, PT, heparin ratio, anti-factor Xa, INR, platelets, hemoglobin and leukocytes characterized patients with bleeding complications.

The observed incidence of bleeding in our study corroborates with those from previously reported studies, showing an overall incidence of major bleeding (42%) [4, 6, 9], and confirming associations with lower body weight and female sex [23–25]. The latter could be explained by pharmacokinetic differences in females [26], higher comorbidity burden in women requiring circulatory support [27] and the risk of anticoagulant overdosing in lower-weight patients [25]. A key observation was the change in distribution of bleeding locations: early events (< 48 h) occurred mainly at the cannulation site, particularly in non-surgical patients, and became less common later during support. This pattern mirrors findings from extracorporeal cardiopulmonary resuscitation (ECPR) and veno-venous (VV) ECMO cohorts [13, 28]. Given that a comparable pattern was observed in VV ECMO patients without cardiogenic shock, it is plausible that cannulation-related trauma and the initial bolus of UFH are important drivers of early bleeding events in non-surgical patients. In contrast, surgical patients more often developed intrathoracic bleedings (e.g. hematothorax, cardiac tamponade), a reflection of postoperative bleeding from surgical sites. Notably, in this group, a secondary rise in intrathoracic bleeding around day 5 was observed, possibly due to ongoing administration of anticoagulation and progressive loss of platelet adhesion [29] and von Willebrand factor [30], leading to impaired hemostasis and an increased risk of delayed bleeding at previous surgical sites. Alternatively, difficulties in diagnosing cardiac tamponade in the setting of VA ECMO support [20] and a subsequent delay in its diagnosis could also have contributed to the second peak of bleeding events.

Besides the insights in distributions of bleeding locations over time, our study adds to current knowledge on bleeding complications in VA ECMO by providing a longitudinal characterization of its biochemical risk factors. Bleeding was preceded by a marked prolongation of coagulation times and higher anti-factor Xa and INR, whereas platelet and hemoglobin levels remained consistently lower in cases than in matched controls. These patterns likely reflect a combination of consumptive coagulopathy [11, 31], (too liberal) UFH exposure [4, 7, 32, 33], and ECMO-related acquired von Willebrand syndrome and platelet dysfunction [11]. These observed differences in marker trajectories between cases and controls were most pronounced in the early bleeding events (< 48 h), and are likely a reflection of shock-related and/or post-surgical coagulopathy [11], combined with the UFH bolus administered during cannulation. More stable trajectories in these markers during late bleeding events may reflect alternative mechanisms such as mechanical injury or consumptive coagulopathy [11, 34] (Fig. 5).

Fig. 5.

Bleeding incidence and coagulation marker dynamics during the first week after VA ECMO initiation. Bleeding incidence is highest within the first 48 h and decreases thereafter. We hypothesize that early bleeding is predominantly related to procedural and perioperative factors, whereas bleeding occurring beyond 48 h more likely reflects consumptive coagulopathy and delayed or mechanical surgical-site bleeding

Our findings show that bleeding risk in VA ECMO is associated with both baseline patient characteristics and temporal differences in laboratory parameters during support. Relying on fixed, one-size fits all laboratory thresholds (e.g., aPTT 50–70 s) likely oversimplifies this complexity, particularly in the critically ill where inflammation, sepsis, and coagulopathy can alter assays independent of heparin effect [32, 35]. The trajectories of coagulation and inflammatory markers observed in patients with versus without bleeding suggest that changes may already occur shortly before a bleeding event. Combining the results of multiple coagulation and inflammatory factors and their temporal evolution, rather than relying on isolated thresholds, may offer a more accurate representation of the hemostatic status of a patient and provide a basis for dynamic, individualized anticoagulation management.

Limitations

Our study is limited by a lack of knowledge about the exact onset of bleeding complications. Nevertheless, this reflects clinical practice where often delays in the diagnosis of (internal) bleedings occur. Accordingly, alterations in coagulation marker concentrations may also serve as diagnostic signals of active bleeding. We applied predefined time windows to mitigate the risk for misclassification. Only major bleeding events were included, ensuring clinical relevance and alignment with the ELSO data dictionary, but this may have influenced associations with risk factors and outcomes. Consequently, some clinically relevant events, such as intracranial hemorrhage, were not classified as major bleeding within this framework. We did provide this information in separate. Anticoagulation protocols varied across centers, including monitoring strategies and target ranges, potentially affecting bleeding incidence and coagulation marker trajectories. We also did not assess other markers related to bleeding such as reductions in von Willebrand multimers, FXIII activity, platelet function tests or viscoelastic parameters (e.g., EXTEM, FIBTEM), which may provide additional mechanistic insight. Finally, heterogeneity in diagnosis, surgical status, and illness severity may have contributed to variability in bleeding risk and biomarker dynamics. In addition, patients were managed over a prolonged inclusion period during which advances in critical care and ECMO management occurred, including the gradual implementation of standardized weaning protocols, which may have added further heterogeneity to the case mix. It is however less likely that these changes caused confounding or effect modification.

Future perspectives

Whether trajectories of inflammatory and coagulation markers hold additional prognostic value beyond static measurements remains to be determined in future studies. Joint models that link longitudinal and time-to-event data offer a promising framework for such analyses [36]. Importantly, these models should also account for treatment strategies, including anticoagulation dosing [37]. Integrating baseline risk factors with evolving laboratory data and clinical management may, in time, contribute to more individualized bleeding risk assessment. These assessments may further benefit from the incorporation of additional parameters, such as viscoelastic testing, platelet function testing, and direct thrombin inhibitor specific assays to enhance mechanistic understanding of anticoagulation during ECMO and support the development of optimized monitoring strategies. However, substantial methodological work and clinical validation will be required before such approaches can be implemented in practice.

Conclusion

Major bleeding complications are common during VA ECMO, particularly during the early phase of support, and are strongly associated with increased ICU mortality. The observation that bleeding was preceded by prolonged coagulation times and lower hemoglobin and platelet levels highlights the dynamic nature of bleeding risk and creates an impetus to refine individualized bleeding risk assessment during the course of VA ECMO.

Author contributions

LP, HB, KL, OC, DD, DR and CM contributed to the concept and design of the study. LP and HB contributed to data collection and analysis. DR and KL contributed with statistical/methodological support. LP, HB and CM drafted the manuscript. All authors contributed to a critical revision of the manuscript.

Funding

Stichting Gezondheidszorg Spaarneland (SGS) Fund, Zilveren Kruis Achmea.

Data availability

Data will be available upon reasonable request to corresponding author.

Declarations

Competing interests

The authors declare the following conflicts of interest: D. W. Donker from the institutional research cooperation of the Cardiovascular and Respiratory physiology group of the University of Twente with Maquet Critical Care AB, Solna, Sweden and Sonion Nederland BV, Hoofddorp, The Netherlands (no personal honoraria received). C.L. Meuwese received research funding from “Stichting Gezondheidszorg Spaarneland (SGS)” Fund and a Senior Dekker Clinical Scientist grant from the Dutch Heart Foundation “Hartstichting.” Consultancy work for Getinge and reports having received speaker fees from AOP Medical, Abbot and J&J MedTech. CV reports research funding and speaking/consulting fees from J&J MedTech. N. van Mieghem has received research grant support from Abbott Vascular, Boston scientific, Medtronic, Daiichi Sankyo, Teleflex, PulseCath BV and advisory fees from Anteris, JenaValve, Amgen, Abbott Vascular, Boston scientific, Medtronic, Daiichi Sankyo, Teleflex, PulseCath BV.

Ethical approval and consent to participate

The institutional ethics committee approved an opt-out consent procedure (reference 21–604/C).

Consent for publication

All authors consent to publication of this manuscript.