Ischemic and hemorrhagic brain injury during venoarterial-extracorporeal membrane oxygenation

A clinical neurological complication was defined as any clinical event occurring on ECMO support, including any clinical sign suggestive of stroke (hemiplegia, mydriasis, anisocoria, asymmetry on physical examination), but also confusion, delirium, seizures, coma despite sedation withdrawal. Patients were categorized according to brain-injury presence or absence on cerebral computed-tomography images (i.e., no damage, ischemic stroke, intracranial bleeding), and groups were compared [11]. Patients with hemorrhagic transformation of ischemic stroke were arbitrarily classified as ischemic stroke: Because the first injury is ischemic stroke, we assumed the risk factors associated with hemorrhagic transformation of ischemic stroke were the same than those of ischemic stroke, rather than risk factors for intracranial bleeding. Neurological events occurring pre-ECMO or 7 days post-ECMO removal were not considered as having occurred on ECMO, and those patients were classed as having no damage. Events occurring within 7 days post-ECMO were arbitrarily considered to be on-ECMO brain injuries because a neurological event is sometimes diagnosed only several days after its occurrence, mainly because patients are sedated, making their evaluation difficult. Patients with diffuse microbleeds were not included because this particular condition’s pathophysiology differs from that of ischemic stroke or intracranial bleeding [12]. Central VA-ECMO refers to cannulation performed within atria and ventricles during open-chest surgery, and peripheral VA-ECMO refers to cannulation performed within femoral vessels, either percutaneously or after surgical cut-down.

Anticoagulation protocol [11, 13, 14], and membrane oxygenator and its circuitry management are reported in the Additional file 1.

Patients underwent daily routine neurological examinations by ICU physicians and nurses at least once daily after sedation withdrawal, including Glasgow coma scale calculation, response to verbal orders or pain, tendon reflexes, brainstem reflexes and plantar reflex, eye opening and pupil examination, with pupil sizes and their light reactivity assessed every 4 h. Moreover, any unexpected event (e.g., seizures, delirium confusion, no awakening after sedation withdrawal…) was recorded in the medical chart. Once a neurological symptom was observed (including but not restricted to change in neurological examination findings, mydriasis, anisocoria, seizures, delirium, confusion, coma despite sedation withdrawal…), a cerebral computed-tomography scan was obtained within 6 h.

Data are expressed as medians [25th;75th percentile] or means [± standard deviation (SD)], as appropriate. Between-group comparisons were analyzed using Student’s t test, the Mann–Whitney U-test or Kruskal–Wallis test for continuous variables and Chi-square test for categorical variables. A logistic-regression model was used to test the univariable association of patients’ clinical characteristics and ICU events with the development of intracranial bleeding or ischemic stroke. Thereafter, multivariable logistic-regression models using backward-stepwise variable elimination (with the variable-exit threshold set at P > 0.05) compared the factors that were significant in the univariable analyses (P ≤ 0.10), and those previously reported to be strongly associated with intracranial bleeding or ischemic stroke were entered into each model. Interactions were tested in the models; variables strongly associated with other(s) were not included in the multivariable model. For univariable and multivariable analyses, continuous variables were dichotomized according to their median values, except for those suspected of being associated with ischemic stroke or intracranial bleeding. To analyze factors associated with ischemic stroke, the following platelet count, fibrinogen and prothrombin time cutoffs at ECMO onset were retained: 350 giga/L, 6 g/L and 70%, respectively, rather than their median values, assuming that patients with “normal” or “supranormal” coagulation parameters could be more at risk of ischemic stroke than patients without. To analyze factors associated with intracranial bleeding, the following platelet count, fibrinogen, aPTT and prothrombin time cutoffs at ECMO start were chosen: 100 giga/L, 1.5 g/L, 3% and 30%, respectively, rather than their median values, postulating that patients with impaired coagulation parameters could be more at risk of intracranial bleeding than patients without.

Because there may be competing risks between the risk of developing neurological complications on ECMO and death, we performed 3 supplementary multivariable analyses: the first one to investigate factors associated with hospital mortality; the second to investigate factors associated with a composite endpoint of death and ischemic stroke; and the third one to investigate factors associated with a composite endpoint of death and intracranial bleeding.

Two nested case–control studies were designed to explore the role of specific risk factors in intracranial bleeding or ischemic stroke. Controls were patients with no damage matched for age ± 5 years, SAPS II ± 5, ENCOURAGE mortality-risk score ± 5 and ECMO duration. See the Additional file 1 for their methodological details.

Specific univariable and multivariable analyses of the 42 ischemic stroke patients attempted to identify risk factors associated with this complication (Table 3). Multivariable analyses retained only central versus peripheral ECMO and platelets > 350 giga/L at ECMO start as being significantly associated with ischemic stroke, with respective OR [95% CI] of 3.2 [1.5–6.6] and 3.8 [1.4–10.7]. None of these factors were associated with hospital mortality (see Additional file 1: Table S5) or with the composite endpoint of death and ischemic stroke.

Because patient groups were heterogeneous, we designed a nested case–control study to evaluate potential associations of hemostasis parameters during ECMO support and ischemic stroke (Additional file 1: Table S3). Forty of our 42 cases could be matched with 86 controls, at least one per case. No relevant association could be established between anticoagulation use (with aPTT considered a surrogate marker of heparin dosage), fibrinogen level or platelet counts during ECMO and ischemic stroke. Figure 2A shows a brain CT-scan of an ECMO-related cerebral infarction.

Multivariable logistic-regression analysis of the risk factors of 20 patients with intracranial bleeding (Table 4) retained female sex, central versus peripheral VA-ECMO and platelets < 100 giga/L at ECMO onset as being significantly associated with intracranial bleeding, with respective OR [95% CI] of 2.9 [1.1–7.5], 3.8 [1.1–10.2] and 3.7 [1.4–9.7], and notably excluded hemostasis disorders during ECMO (particularly low platelet count). Whereas platelets < 100 giga/L at ECMO start were associated with death (see Additional file 1: Table S5) and with the composite endpoint of death plus intracranial bleeding (OR 1.7; 95% CI 1.2–2.3), female sex and central versus peripheral ECMO were not associated with these outcomes.

Because it was been previously shown that blood-gas change at ECMO onset is associated with intracranial bleeding in patients receiving VV-ECMO [11], a nested case–control study was undertaken to evaluate this association in VA-ECMO-treated patients (see Additional file 1). Nineteen of our 20 cases with intracranial bleeding could be matched with 58 controls. Cases’ PaCO₂ decreased more after ECMO onset, and they had lower platelet counts at ECMO start compared to the controls (Additional file 1: Table S4). No pattern of a specific location or aspect of ECMO-related brain hemorrhage was identified among all brain imaging, and within the 20 patients with intracranial bleeding, we identified, 13 patients (65%) with intraparenchymal hemorrhage (among which 4 patients with multiples lesions), 4 patients (20%) with subarachnoid hemorrhage and 3 patients (15%) with acute subdural hematoma. Figure 2B shows a brain CT-scan of an ECMO-related intraparenchymal hemorrhage.

Our study has several limitations. The first is its retrospective, single-center design. However, this large study covered a long period and analyzed previously unexamined risk factors for ischemic stroke and intracranial bleeding. Therefore, we think our results may be transposable to other ICUs with other patients. Second, because patients with clinical symptoms underwent brain imaging, it is highly likely that we may have missed some subclinical events. Indeed, Rastan et al. [17] identified during autopsy a high percentage of patients with thromboembolic events that were not clinically apparent, and a recent study on VV-ECMO-treated patients showed that systematic cerebral computed-tomography scans may find clinically unapparent bleeding events [22]. Third, we tried to find risk factors associated with ischemic stroke and intracranial bleeding, but other potentially confounding factors may not have been considered in our analyses. More specifically, we examined some, but not all, hemostasis parameters known to be modified during ECMO as risk factors for intracranial bleeding, e.g., von Willebrand factor [23]. Thus, our conclusion that hemostasis disorders during ECMO were not involved in neurological events (mainly intracranial bleeding) might be inaccurate. In addition, the potential role of laminar versus pulsed flow in the pathogenesis of neurological complications, mainly ischemic stroke, was not considered. Indeed, some patients may have residual left ventricular ejection on VA-ECMO that might be implicated, especially in ischemic stroke. Unfortunately, because of our study’s retrospective design, this information is not available. Fourth, some variables included in the analyses are not baseline characteristics and may only identify the sickest VA-ECMO patients rather than true risk factors. Fifth, we have no data on long-term quality of life. Although mortality was not different between patients without brain injury and those experiencing ischemic stroke, differences in quality of life may exist. This should be explored in future studies. Sixth, we were unable to have the number of patients who underwent cerebral imaging because of abnormal neurological findings and whose imaging was normal. Last, we decided to exclude patients who developed neurological complication 7 days after ECMO removal, which could underestimate the incidence of brain injury. We hypothesized that a neurological complications occurring more than 7 days after ECMO removal had a low probability to be directly related to the ECMO itself. Since it is our policy to stop sedation after ECMO removal for neurological evaluation, it is highly probable that we are able to diagnose brain injury in the first week after ECMO removal, if this injury occurs during ECMO course. Indeed, there are many reasons for these patients to develop brain injury after ECMO removal: emboli from a failing heart, anticoagulation, etc. Moreover, median time from ECMO removal to diagnosis of brain injury in the patients who were excluded was 13 (IQR 11–21) days, reinforcing our hypothesis.