Brain Histopathology of Adult Decedents After Extracorporeal Membrane Oxygenation

Abstract

Objective

To test the hypothesis that brain injury is more common and varied in patients receiving extracorporeal membrane oxygenation (ECMO) than radiographically observed, we described neuropathology findings of ECMO decedents and associated clinical factors from 3 institutions.

Methods

We conducted a retrospective multicenter observational study of brain autopsies from adult ECMO recipients. Pathology findings were examined for correlation with demographics, clinical data, ECMO characteristics, and outcomes.

Results

Forty-three decedents (n = 13 female, median age 47 years) received autopsies after undergoing ECMO for acute respiratory distress syndrome (n = 14), cardiogenic shock (n = 14), and cardiac arrest (n = 15). Median duration of ECMO was 140 hours, most decedents (n = 40) received anticoagulants; 60% (n = 26) underwent venoarterial ECMO, and 40% (n = 17) underwent venovenous ECMO. Neuropathology was found in 35 decedents (81%), including microhemorrhages (37%), macrohemorrhages (35%), infarctions (47%), and hypoxic-ischemic brain injury (n = 17, 40%). Most pathology occurred in frontal neocortices (n = 43 occurrences), basal ganglia (n = 33), and cerebellum (n = 26). Decedents with hemorrhage were older (median age 57 vs 38 years, p = 0.01); those with hypoxic brain injury had higher Sequential Organ Failure Assessment scores (8.0 vs 2.0, p = 0.04); and those with infarction had lower peak Paco₂ (53 vs 61 mm Hg, p = 0.04). Six of 9 patients with normal neuroimaging results were found to have pathology on autopsy. The majority underwent withdrawal of life-sustaining therapy (n = 32, 74%), and 2 of 8 patients with normal brain autopsy underwent withdrawal of life-sustaining therapy for suspected neurologic injury.

Conclusion

Neuropathological findings after ECMO are common, varied, and associated with various clinical factors. Further study on underlying mechanisms is warranted and may guide ECMO management.

Extracorporeal membrane oxygenation (ECMO) is a type of temporary mechanical circulatory support that artificially provides blood gas exchange and circulation for critically ill patients. The use of ECMO to treat acute respiratory distress syndrome (ARDS), cardiogenic shock, or cardiac arrest has grown by 7-fold worldwide in the past 20 years.¹ Nine percent of patients receiving ECMO experience neurologic injury, including intracranial hemorrhage (ICH), ischemic stroke, and seizure, excluding prior underlying injury.² Patients with neurologic injuries who undergo ECMO experience a higher rate of mortality than those without ECMO (83% vs 42%).² Published rates of neurologic injury have been largely dependent on neuroimaging and EEG performed in the setting of clinical suspicion. Clinical neurologic evaluation is often confounded by sedation or neuromuscular blockade required to prevent the dislodgement of ECMO cannulas, and neuroimaging is limited to CT because the ECMO circuit is incompatible with MRI. This can leave patients with undetected neurologic injury that may occur before ECMO due to hemodynamic instability and hypoxia or during ECMO from a host of mechanisms such as thromboembolism, hemodynamic changes, anticoagulation, and endothelial inflammation.^3,4

We examined neuropathologic findings of adult ECMO decedents with brain autopsies from multiple institutions to understand the span of structural brain injury in this population. Three prior studies have described brain autopsy findings in single-center cohorts, which do not consider practice variations among ECMO centers.^5–7 We hypothesized that brain injury is more common than detected radiographically and more varied than previously described.

Methods

An observational retrospective chart review was conducted for adult patients (age ≥18 years) who received an autopsy after admission for venovenous (VV) or venoarterial (VA) ECMO. The study was conducted across 3 academic medical centers in the United States (University of Rochester Medical Center [UR], NY; University of Maryland Medical Center [UMD], Baltimore; and University of Michigan Medical Center [UMI], Ann Arbor).

Decedent Selection

All patients receiving ECMO during admission from January 2013 to January 2018 were reviewed. All patients met the criteria for ECMO therapy at their respective institutions: (1) they were previously independent with activities of daily living; (2) they had no known underlying neurologic disorder; (3) heart or lung dysfunction was deemed to be possibly treatable at initiation of ECMO; and (4) withdrawal of life-sustaining therapy was not expected within the first 48 hours. Decedents who received an autopsy after ECMO therapy were selected for inclusion in the study.

ECMO Therapy

The initiation of ECMO at all 3 sites was at the discretion of the attending cardiac surgeon, and no center followed a strict institutional protocol. ECMO requires the placement of large-bore cannulas into the artery or vein of the patient. Cannulation for ECMO was performed by a dedicated ECMO team consisting of cardiac surgeons, intensivists, or vascular surgeons.

VV ECMO is defined by the removal of blood from a large vein, circulation through a centrifugal pump and artificial semipermeable membrane (oxygenator) for gas exchange, and return to the body via a different vein. This was initiated on patients who exhibited refractory hypoxemia, defined as Pao₂ of 50 to 60 mm Hg with peak inspiratory pressures >35 cm H₂O and fraction of inhaled oxygen 100% despite optimal medical optimization. Hemodynamically stable patients were transported to an operating room or cardiac catheterization laboratory for placement of a dual-lumen internal jugular vein cannula. In hemodynamically unstable patients, percutaneous cannulas were placed at bedside in both femoral veins or a femoral and jugular vein.

VA ECMO is similar to VV ECMO in circulation through a centrifugal pump and oxygenator, but blood is returned into an artery instead of a vein. This was initiated for patients who could not be weaned from cardiopulmonary bypass after cardiotomy, cardiac arrest, or severe cardiogenic shock. Locations for cannulation were the femoral artery and vein (femoral-femoral) or right atrium and ascending aorta (central, used only in patients with cardiopulmonary bypass). Extracorporeal cardiopulmonary resuscitation was initiated at the cardiac surgeon's discretion for cardiac arrests lasting >10 minutes. Only patients with ventricular dysrhythmia underwent extracorporeal cardiopulmonary resuscitation, and both in- and out-of-hospital arrests were cannulated.

ECMO was performed with 1 of 3 devices, depending on site: the Maquet Cardiohelp (Getinge, Gothenburg, Sweden), Maquet Rotaflow (Getinge), or Thoratec Centrimag (Abbott, Pleasanton, CA). All patients used the Maquet Quadrox-i oxygenator unit (Getinge). The decision to decompress the left ventricle after initiation of ECMO via additional devices such as intra-aortic balloon pump or Impella Ventricular Assist Device was at the discretion of the clinical care team, as were the timing and method of weaning from ECMO.

Pathology Examination

Gross and microscopic specimens obtained as part of postmortem examination were retrospectively reviewed by a single neuropathologist at each site. Sites sampled for histologic examination included the medulla, pons, midbrain, anterior hypophysis, hippocampus, cerebellum, basal ganglia, frontal neocortex, and watershed zones; supratentorial sites were sampled bilaterally. Microscopic specimens were fixed in formalin and stained with hematoxylin and eosin for histopathologic examination at the time of autopsy. For all decedents, total occurrences by anatomic location in each histopathologic category were counted.

Definitions of histologic abnormalities were defined as the following: microhemorrhages were red blood cells outside the cerebral vasculature visible only by microscope; macrohemorrhage was hemorrhage grossly visible on visual examination; acute infarction was area of dead tissue with increased edema and hypoxic (red) neurons; subacute infarction was characterized by the presence of liquefactive necrosis and peripheral neovascularization. Hypoxic ischemic injury was defined by the presence of hypoxic (red) neurons and global parenchymal edema. Other pathologies consisted of axonal degeneration (n = 3), remote infarction (cystic space and microscopic glial scarring, n = 2), endothelial necrosis (n = 2), central pontine myelinolysis (n = 1), fat emboli (n = 1), spinal cord infarction (n = 1), and Duret hemorrhage (hemorrhage grossly visible in medulla or pons with concomitant herniation, n = 1).

Clinical Data and Neuroimaging

Data collected on each decedent included baseline demographics and clinical characteristics on admission, before ECMO, and during ECMO. Clinical characteristics included ECMO settings, neurologic examination findings, and laboratory values. ECMO flow was defined as the rate of blood circulating from the ECMO circuit to the body (liters per minute), and ECMO sweep was defined as the flow rate of oxygen through the oxygenator unit (liters per minute). Severity of illness was characterized by Sequential Organ Failure Assessment (SOFA) scores. Glasgow Coma Scale (GCS) scores were assessed by nursing staff during pauses in sedation whenever safe. For all time points, data closest to that point in time were recorded. CT or MRI neuroimaging that was performed on patients receiving ECMO at any time point during admission was reviewed by board-certified radiologists, and reports were categorized (I.K., Y.G., F.F.). Neuroimaging was performed at the clinician's discretion for suspicion of neurologic injury.

Outcomes

Nursing and discharge documentation was reviewed to determine final GCS scores before death. Survival to ECMO decannulation and withdrawal of life-sustaining therapy were documented. Cause of death was determined from physician review of discharge summary (I.K., Y.G., L.M.) and categorized as neurologic, cardiovascular, respiratory, or other on the basis of the most proximal listed cause of death.

Statistical Analysis

Data are reported descriptively with mean and SD or median and interquartile range (IQR) according to variable distribution and analyzed by pathology type on autopsy with Wilcoxon rank sum for continuous variables and Fisher exact test for categorical variables. Individual pathologies were compared to all other decedents without that pathology; those with normal pathology were compared to those with abnormal pathology. Statistical significance was set a priori at p < 0.05 using 2-sided hypothesis testing. The analysis was performed with Stata version 14.2 (StataCorp, College Station, TX).

Standard Protocol Approvals, Registrations, and Patient Consents

Approval to conduct this study was received from ethics standards committees at UR, UMD, and UMI.

Data Availability

Study data will be made available at the reader’s request to the corresponding author for purposes of replicating procedures and results.

Results

A total of 1,526 patients underwent ECMO at the 3 sites between 2013 and 2018 (516 at UR, 238 at UMI, 772 at UMD). Forty-three decedents received brain autopsy (10 at UR, 14 at UMI, 19 at UMD).

Histopathologic Examination

Neuropathology was noted on 35 of 43 decedents (81%). Postmortem analysis revealed 20 decedents with acute or subacute infarction, 17 with hypoxic-ischemic brain injury (HIBI) including 10 with a diffuse pattern of cerebral edema, 16 with microhemorrhage, and 15 with macrohemorrhage (figure 1). Nine decedents had a variety of other pathologies, described above. The mean number of pathologies per decedent was 1.8 (SD 1.3, median 2, IQR 1–3). The mean number of pathologies for those with other pathology is inherently greater (3.1 pathologies [SD 1.3] vs 2.0 pathologies excluding those with normal pathology [SD 1.3], p = 0.01). Among the 16 decedents with microhemorrhages, 9 (56%, p = 0.04) also had macrohemorrhage. Two decedents had both acute and subacute stages of infarction in different anatomic locations. Representative examples of infarction, hemorrhage, and hypoxic ischemic brain injury are shown in figure 2.

Figure 1. Pathology Type and Neuroimaging by Decedent^a.

^aPathologies are not mutually exclusive. Individual decedents often have multiple pathologies in different anatomic locations. This graphic demonstrates the overlap of pathologies by decedent. ^bSee Methods section for definition of other pathology types. Asterisk indicates decedents (site B decedent 14, site C decedent 10) with 2 other pathology types.

Figure 2. Examples of Pathologies Noted on Brain Autopsy From Decedents With ECMO.

Top row: (A) basal ganglia neuronal necrosis and (B) pontine microinfarctions (red arrows) in a 21-year-old man who underwent venoarterial extracorporeal membrane oxygenation (ECMO) for cardiac arrest and (C) pituitary infarction in a 47-year-old man who underwent venovenous ECMO for acute respiratory distress syndrome. Bottom row: (D) pontine microhemorrhage and (E) pontine periventricular edema in a 69-year-old man who underwent venoarterial ECMO for acute cardiogenic shock and cardiac arrest. All samples stained with hematoxylin and eosin.

Across the predefined 9 anatomic locations within the brain, there were 170 occurrences of pathology among the 43 decedents in the study. The frontal neocortex (n = 43 occurrences), basal ganglia (n = 33), and cerebellum (n = 26) were the most affected areas of the brain (figure 3). HIBI occurred most frequently by anatomic location (n = 56 occurrences), followed by infarction (n = 38) and macrohemorrhage (n = 30). HIBI demonstrated the most diffuse damage with pathology in multiple anatomic locations. Examination of pathologies by location demonstrated a correlation with macrohemorrhages and infarctions in watershed zones (n = 3 of 3 watershed infarcts with hemorrhage, p < 0.001).

Figure 3. Heat Map of Pathology Type by Anatomic Location^a.

^aThe n values represent the number of occurrences for each pathology type by anatomic location. ^bSee table 1 for definition of other pathology types.

Demographic and Clinical Characteristics

The median age of all decedents was 47 years (IQR 30–58 years) (table 1). Thirteen decedents were female, and the median length of stay was 24 days (IQR 5–51 days). Individuals with either microhemorrhages or macrohemorrhages (n = 22 accounting for overlap) tended to be older (median 57 years, IQR 42–60 years vs median 38 years, IQR 25–48 years, p = 0.01). Indication for ECMO was split among extracorporeal cardiopulmonary resuscitation for cardiac arrest (n = 15, 35%), cardiogenic shock (n = 14, 33%), and ARDS (n = 14, 33%). The most common comorbid conditions included smoking, hypertension, and congestive heart failure. Those with microhemorrhage were more often smokers and had congestive heart failure. Preexisting venous thromboembolism was infrequent (n = 3) but present exclusively in those with macrohemorrhage. A larger proportion of microhemorrhages were identified within site A compared to sites B and C (n = 8 of 10 vs 4 of 14 and 4 of 19, respectively, p = 0.01).

Table 1.

Demographic and Clinical Characteristics of ECMO Decedents by Pathology Type^a

ECMO Characteristics

Twenty-six (60%) decedents in this study underwent VA ECMO; 17 (40%) underwent VV ECMO (table 1). Most were cannulated in the operating room (n = 22, 54%). More common cannula locations included femoral-femoral (n = 19, 44%), central (n = 10, 23%), and femoral-jugular (n = 9, 21%). The large majority were on anticoagulation during ECMO (n = 40, 93%). Median ECMO duration was 140 hours (IQR 48–618 hours). Characteristics of ECMO initiation, type, place of cannulation (i.e., emergency department, operating room, intensive care unit, other hospital), cannula location (i.e., femoral-femoral, femoral-jugular, dual lumen jugular, central), anticoagulation (yes/no), duration of ECMO, and circuit flow and sweep were similar across pathology subgroups (data not shown). When the 5 patients with normal pathology and known maximum ECMO sweep gas flow were compared to the 27 patients with abnormal pathology and known maximum sweep, those with normal pathology had a higher maximum sweep (median 7.0 L/min, IQR 6.0–9.0 L/min vs median 5.5 L/min, IQR 4.5–6.0 L/min, p = 0.04).

Laboratory and Clinical Data

On admission, decedents with microhemorrhages had greater GCS scores (mean 14.2, SD 3.1 vs mean 10.4, SD 5.6, p = 0.02) compared to decedents without microhemorrhage (table 2). Decedents with hypoxic brain injury had lower admission GCS motor scores (mean 4.1, SD 2.4 vs mean 5.2, SD 1.9, p = 0.04) and higher SOFA scores (median 8.0, IQR 3.5–14 vs 2.0, IQR 1–12, p = 0.04) compared to those without hypoxic injury. Patients with other pathology, who had a greater number of pathologies on postmortem examination, had lower average platelet counts (132,000 [IQR 73,000–189,000]) and higher average bilirubin (2.9 mg/dL [IQR 0.9–3.9 mg/dL]) and arterial oxygen tension (138 mm Hg [IQR 97–229 mm Hg]) compared to other groups. During ECMO, those with acute or subacute infarction had lower maximum Paco₂ compared to those without infarction (median 53 mm Hg, IQR 45–62 mm Hg vs median 61 mm Hg, IQR 50–96 mm Hg, p = 0.04). Twenty-six decedents (61%) underwent CT (n = 26) or MRI (n = 3), of which 17 were abnormal. All decedents with abnormal neuroimaging had abnormal autopsies (n = 17 of 17), while 67% with normal imaging (n = 6 of 9) also were found to have abnormal autopsies. The most common findings on imaging were cerebral edema (n = 8) and hemorrhage (n = 7).

Table 2.

Clinical and Laboratory Values on Admission, Before ECMO, and During ECMO for Decedents by Pathology Type^a

Decedent Outcomes

Final GCS score (mean 7.4, SD 4.5 vs mean 4.0, SD 2.1, p = 0.008) was greater for decedents with normal autopsies (table 3). Eleven decedents survived to decannulation of ECMO. Most decedents died after withdrawal of life-sustaining therapy (n = 32, 74%). The most common causes of death were cardiovascular (n = 21, 49%) and neurologic (n = 11, 26%). These categories included refractory hypoxemia (n = 10, 23%), multiorgan failure (MOF) due to cardiac arrest (n = 8, 19%), MOF due to cardiogenic shock (n = 8, 19%), coma from cardiac arrest (n = 5, 12%), brain death (n = 5, 12%), acute ischemic stroke (n = 2, 5%), MOF due to septic shock (n = 3, 7%), massive hemorrhage from ECMO cannula (n = 1, 2%), and repeat cardiac arrest after decannulation (n = 1, 2%). Two decedents with normal postmortem examination of the brain had a cause of death listed as neurologic, and both of these decedents died as a result of withdrawal of life-sustaining therapy.

Table 3.

Decedent Outcomes^a

Discussion

To the best of our knowledge, this is the largest ECMO brain autopsy study to date to describe brain autopsy findings from multiple ECMO centers. We sought to conduct a multicenter observational study because considerable practice variation exists among ECMO centers, and it is unknown whether this correlates with neurologic injury.⁸ Microhemorrhages and macrohemorrhages were our most common finding, followed by acute ischemic strokes and HIBI. We found that the occurrence of cerebral hemorrhage differed among treatment sites. Regardless of variation in histotype, we found brain injury to be common among those with ECMO (81%), including patients with normal CT scans (67%). Finally, autopsy findings were not uniformly congruent with clinical manifestations, and 2 individuals for whom life-sustaining therapy was withdrawn for suspected brain injury were found to have normal autopsies.

Radiographic incidence of ICH in patients receiving ECMO varies from 2% to 19% in the published literature.⁹ Cerebral microhemorrhages have been far less described radiographically because the use of MRI is not routine in adult ECMO survivors.¹⁰ Prior studies found a correlation between ICH in patients receiving ECMO and women, thrombocytopenia, and duration of ECMO.^11,12 In our study, patients with ICH were older and more likely to be smokers compared to those without ICH. Older patients may be at higher risk of ICH due to cerebral amyloid angiopathy, but the median age for patients with microhemorrhages in our cohort (59 years) was relatively young compared to the typical age group in which cerebral amyloid angiopathy is observed.^13,14 Older age may also decrease platelet function.¹⁵ In addition, smoking is a risk factor for spontaneous ICH and may have contributed to the risk of ICH in our cohort. Although the use of anticoagulation has long been associated with hemorrhage in patients receiving ECMO, almost all patients in our study were heparinized, and partial thromboplastin times were no different among ICH cohorts.^16,17

We noted 1 site to report a significantly higher number of hemorrhages than the others, which could be due to differences in ECMO technique, equipment, or clinical characteristics. Differences in ECMO equipment used may possibly account for this discrepancy, and there is literature to support mechanical differences between the 2 most popularly used centrifugal pumps.¹⁸ However, to the best of our knowledge, no literature exists suggesting differences in occurrence of neurologic injuries. Another source of discrepancy among sites is clinician-driven variability in ECMO management. Significant variability exists among ECMO clinicians worldwide in weaning strategies, ECMO flow management, cannulation strategies, and ventilator management.¹⁹ We did not notice a difference between sites in the use of anticoagulation, laboratory values, SOFA scores, or demographics. However, site A had fewer decedents with hypotension on admission, and site C cannulated more often in the operating room.

Acute ischemic strokes were the next most commonly found pathology in our cohort. Possible etiologies of infarctions in patients receiving ECMO include an inflammatory prothrombotic response to contact with foreign circuit material, decreased cerebral perfusion due to cardiogenic shock, and arterioarterial thromboembolisms.^2,20 We found our stroke cohort to have a lower peak Paco₂ level, possibly implicating impairments in cerebral autoregulation leading to vasoconstriction and ischemia.^21–23 Elevations in lactate are associated with the occurrence of ischemic strokes in patients receiving ECMO, but this finding was not recapitulated in our cohort.² Our stroke cohort had lower total and motor GCS scores before death than individuals with other pathologies, implying deeper comatose states. This finding is aligned with the morbidity that acute ischemic strokes impart on patients receiving ECMO, causing significantly longer lengths of stay, higher rates of discharge to nursing homes, and greater health care costs.²⁴

HIBI was a common finding in our cohort and has previously been described in autopsies of ECMO decedents.^5,6 In our cohort, patients with HIBI appeared to be more severely ill on admission with higher SOFA scores and lower GCS motor scores. The occurrence of HIBI was not associated with a particular indication or type of ECMO (VV or VA). These findings are similar to findings of a prior ECMO autopsy study in that HIBI likely occurred before ECMO initiation.⁶ However, it is unclear whether ECMO-driven pathophysiologies play a role in the development of this injury. Differential hypoxia, one such pathology, can occur when the heart begins regaining function but pumps poorly oxygenated blood into the aorta and cerebral vessels, overpowering the retrograde perfusion of oxygenated blood of the ECMO circuit.²⁵ ECMO can also be associated with cerebrovascular dysregulation, leading to vasoconstriction and ischemia.^22,26

We noted a preponderance of hemorrhages, infarctions, and HIBI to occur in the cerebellum, basal ganglia, and frontal neocortex and a cluster of infarction and ischemic injury in the anterior hypophysis and hippocampus. Many individuals were seen to have the co-occurrence of these pathologies, suggesting a common pathway of HIBI that can lead to infarctions from the no-reflow phenomenon or hemorrhages from reperfusion injury.²⁷ This pathway is also suggested by our cohort's injury distribution; HIBI particularly injures neurons in the cerebellum, basal ganglia, frontal neocortex, and hippocampus. Surprisingly, no association was noted between types of pathology and indication for ECMO, which suggests that our cohort was much more susceptible to HIBI than clinically suspected.

We also noted a cluster of infarctions and HIBI in the anterior hypophysis, which has not been previously described in the literature. Pathology at this location was seen in all 3 institutions. None of the patients in our cohort exhibited clinical hypopituitarism. Possible etiologies of pituitary necrosis could include embolic infarcts or cerebral vasospasm from subarachnoid hemorrhage because blood flow to the anterior pituitary is supplied solely by the superior hypophyseal artery.^28,29 However, further study will be required to examine a possible pattern of pituitary infarction in patients on ECMO.

Very few common indicators of critical illness were associated with types of pathology in our cohort (tables 2 and 3). Patients with normal autopsies had higher platelet levels, suggesting a lower burden of platelet consumption and shear, which has been associated with infarctions and hemorrhages.^30,31 They also had higher maximum ECMO sweep flows, suggestive of an ARDS population with higher gas exchange needs. Patients with ARDS on VV ECMO typically have fewer incidences of neurologic injury than patients on VA ECMO with cardiac failure.³² However, neither type of ECMO nor underlying etiology necessitating ECMO was associated with pathology in our cohort. Likewise, other typical indicators of critical illness such as lactate, SOFA scores, and hemoglobin levels were not associated with particular pathology types. It is possible that these typical indicators may not be strong biomarkers of the pathophysiology that patients on ECMO encounter. If HIBI were a common pathway for ischemia and hemorrhage in our cohort, the mechanism for injury could include reperfusion and oxidative stress, associated with cascading inflammation, leading to neuronal injury.^33,34 This pathway might be common to various etiologies necessitating ECMO such as ARDS or cardiac arrest. Particularly in patients receiving ECMO, arterial pulsatility and CO₂ vasoreactivity may play a role in brain injury.³⁵ Future studies of these pathways may lead to better biomarkers of neuropathology in critically ill adults who undergo ECMO.

Autopsy is an important tool to evaluate for missed neurologic injuries in patients with ECMO in whom serial clinical examination is limited by heavy sedation and analgesia requirements.^7,36 Accuracy in diagnosis becomes especially important when discussing prognosis with family members who ultimately make decisions about goals of care for these patients. In our cohort, 32 of 43 (74%) patients underwent withdrawal of life-sustaining therapy, 11 of whom were withdrawn for suspected devastating brain injury. Of these patients, 2 of 11 had normal brain autopsy results, leading one to suspect diagnostic error that could have been due to inadequate sedation clearance or other reversible causes of encephalopathy. Conversely, 67% of patients with normal neuroimaging were found to have some sort of abnormality on autopsy, highlighting the limitation of CT to detect neurologic pathology. Notably, the majority of patients in our study did not survive to decannulation and underwent withdrawal of life-sustaining therapy while undergoing ECMO. Diagnostic accuracy can be improved by performing close neurologic examination, minimizing sedation, and mobilizing patients on ECMO whenever possible, realizing that cannula dislocation, ventilator dyssynchrony, and pain may be caveats to this strategy.³⁷ However, it remains unknown whether some pathologies affect only finer neurocognitive function and would escape the detection of clinical surveillance.^38,39

While many hypotheses can be generated by our findings, several limitations must be considered. We conducted an observational retrospective study of ECMO recipients only, leaving open the possibility of selection bias due to the omission of a non-ECMO control group. Many findings such as microhemorrhages and HIBI could be caused by cardiac arrest, sepsis, or other illnesses common in critically ill patients.^27,40,41 Future studies examining the association of these pathologies with ECMO should include a control group matched by severity of illness and demographic factors, which has proved to be difficult in ECMO research.⁴² Gross and microscopic specimens were reviewed by pathologists local to each site rather than a centralized, blinded pathology team, which can impart interrater variability. We strived to limit this by grading pathology specimens by prespecified anatomic locations and definitions agreed on by all pathologists. The timing of neuroimaging was not protocolized because it was obtained at the discretion of the clinician, but all images were obtained during or after ECMO. As a result, it may not precede pathology in all cases, leading to a false-negative finding. Finally, the pathologies described herein could have occurred in the time period after discontinuation of ECMO and before death. However, the majority of our patients died while on ECMO, decreasing the impact of this shortcoming on our results.

Various forms of neuropathology, including hemorrhages, infarctions, and hypoxic-ischemic injury, were noted in autopsies of adults who received ECMO before death. While it remains unknown whether ECMO causes these injuries, numerous mechanisms make this a possibility. Further study on cerebral physiology in patients receiving ECMO may identify interventions to prevent neurologic injury.

Glossary

ARDS

acute respiratory distress syndrome

ECMO

extracorporeal membrane oxygenation

GCS

Glasgow Coma Scale

HIBI

hypoxic-ischemic brain injury

ICH

intracranial hemorrhage

IQR

interquartile range

MOF

multiorgan failure

SOFA

Sequential Organ Failure Assessment

UMD

University of Maryland Medical Center

UMI

University of Michigan Medical Center

UR

University of Rochester Medical Center

VA

venoarterial

VV

venovenous

Appendix. Authors

Study Funding

No targeted funding reported.

Disclosure

I.R. Khan receives research funding from the UR (University Research Award). Y. Gu, B.P. George, L. Malone, K.S. Conway, F. Francois, J. Donlon, N. Quazi, and A. Reddi report no disclosures. C.Y. Ho receives research funding from the National Institute of Neurologic Disorders (NS102468-03). D.L. Herr, M.D. Johnson, and G.Y. Parikh reports no disclosures. Go to Neurology.org/Nhttps://n.neurology.org/lookup/doi/10.1212/WNL.0000000000011525 for full disclosures.