Neonatal extracorporeal membrane oxygenation (ECMO) is the final treatment option for patients with severe cardiopulmonary failure. ECMO temporarily supports cardiopulmonary function by circulating blood through an external circuit for oxygenation and carbon dioxide removal before reinfusion into the body. This intervention is typically considered when conventional therapies such as mechanical ventilation, fluid resuscitation, and inotropic support are ineffective. First introduced in neonates by Bartlett in 1975, ECMO use has steadily increased worldwide. The most recent data from the Extracorporeal Life Support Organization registry indicate that approximately 50,000 neonatal ECMO cases are conducted annually worldwide, with an overall survival rate of approximately 65%. Among the neonates with respiratory failure, the survival rate increased to 72% [1]. Although ECMO is used in adults and older children for both cardiac and respiratory failure in similar proportions, it is more frequently used in neonates with reversible respiratory illnesses, leading to improved outcomes in this population. However, the small body and vessel sizes of neonates often necessitate surgical cannulation and careful management, placing them at an increased risk of complications such as bleeding.

According to recent reports, including meta-analyses, the incidence of brain injury in neonates receiving ECMO is approximately 50% [2-4]. These brain injuries commonly manifest as ischemic or hemorrhagic lesions associated with ECMO therapy but may result from hypoxic-ischemic injury due to severe cardiopulmonary failure during ECMO support [5]. Unlike adults, in whom percutaneous cannulation is widely used, neonates typically require surgical cannulation for ECMO initiation, which often delays hemodynamic stability. Moreover, the cerebral autoregulatory capacity of neonates is narrower than that of adults, making them more susceptible to brain injury from blood pressure fluctuations before and during ECMO [6].

Neonates who experience ECMO-associated brain injury remain at a high risk of long-term neurodevelopmental impairment, even if they survive. Therefore, strategies aimed at optimizing neurodevelopmental outcomes are required, particularly in populations vulnerable to ECMO-related brain injury. To develop such strategies, it is essential to understand the patterns and long-term consequences of brain injury in neonatal ECMO survivors. However, studies addressing this issue are limited [7,8] and to date, no such research has been conducted in Korea, where neonatal ECMO is relatively uncommon.

This study aimed to characterize the clinical patterns and severity of brain injury in neonatal survivors who underwent ECMO for respiratory failure, evaluate their short-term neurological outcomes, and identify the factors associated with these outcomes.

We retrospectively reviewed the medical records of neonates who were admitted to the neonatal intensive care unit (NICU) at Asan Medical Center between February 2018 and February 2024 and who survived discharge after receiving ECMO therapy for acute cardiopulmonary failure within the first 28 days of life. The exclusion criteria included ECMO initiation due to postoperative deterioration following congenital heart disease surgery, the presence of major genetic abnormalities known to affect neurodevelopment, and the absence of documented post-discharge developmental assessments in the medical records.

ECMO was initiated when patients presented with severe hypoxemia unresponsive to maximal mechanical ventilation (oxygenation index [OI] >40 or PaO₂ <30–40 mm Hg), severe pulmonary hypertension accompanied by right or left ventricular dysfunction, or hypotension and shock refractory to vasopressor support. All patients received venoarterial (VA)-ECMO. The ECMO system (Xenios AG) was used in conjunction with the Minilung Kit (Xenios AG), which included a Deltastream DP3 (Medos Medizintechnik AG) centrifugal pump and a Hilite oxygenator (Medos Medizintechnik AG). When this system was unavailable, the RotaFlow system (Getinge), equipped with a cone-type centrifugal pump, was used in combination with a Lilliput2 oxygenator (LivaNova).

Continuous anticoagulation with heparin is required during ECMO to prevent thrombus formation within the extracorporeal circuit. The heparin dose was adjusted to maintain an activated clotting time of 180 to 200 seconds and an activated partial thromboplastin time of 1.5 to 2.5 times the normal range. Anticoagulation targets were modified according to the bleeding risk and coagulation status of each patient.

Weaning from ECMO was considered when the sweep gas flow could be reduced to ≤0.3 L/min, the ventilator minute ventilation exceeded 0.2 to 0.3 L/kg/min, and the ECMO flow was decreased to <30 to 50 mL/kg/min. In addition, echocardiographic evidence of recovered or normal left ventricular function and the absence of only mild-to-moderate pulmonary hypertension are required.

To identify the factors associated with brain injury and developmental delay, data on perinatal history, demographic characteristics, initial arterial blood gas results after birth, duration of mechanical ventilation, and use of high-frequency oscillatory ventilation or inhaled nitric oxide (iNO). ECMOrelated variables included pre-cannulation arterial blood gas values, OI, duration of lactic acidosis (serum lactate >5.0 mmol/L), duration of severe hypoxemia (OI >40), timing of ECMO initiation and termination, occurrence of cardiopulmonary resuscitation before or during ECMO, catheter malfunction and replacement, failure to wean from ECMO, and the use of continuous renal replacement therapy (CRRT). The complications and follow-up indicators included seizure occurrence, length of NICU stay, neurodevelopmental outcomes, and hearing loss. Brain injury severity was classified as normal, mild, moderate, or severe based on brain magnetic resonance imaging (MRI) performed before discharge, according to the criteria described by Bulas and Glass [9]. Mild brain injury was defined as the presence of a small (<1 cm) extra-axial hemorrhage, scattered petechial hemorrhage, periventricular or focal white matter injury, mild cerebral edema, or mild ventricular dilatation. Moderate brain injury included findings such as a single large parenchymal hemorrhage (>1 cm), patchy periventricular leukomalacia or hypodensity, a combination of mild hemorrhagic and non-hemorrhagic abnormalities, mild generalized brain atrophy, or mild thinning of the corpus callosum.

Severe brain injury was defined as the presence of multiple large parenchymal hemorrhages (>1 cm), diffuse periventricular leukomalacia, large parenchymal infarcts (>1 cm), moderateto-severe generalized brain atrophy, or diffuse thinning of the corpus callosum.

Short-term neurodevelopmental outcomes were assessed using the Bayley Scales of Infant Development (BSID), either in the second or third edition, between the ages of 12 and 40 months. When BSID data were unavailable, developmental status was assessed using the Korean Developmental Screening Test (K-DST). For infants who underwent more than one developmental assessment, the most recent evaluation was considered for the analysis. Developmental delay was defined as a mental developmental index (MDI) <70 or a psychomotor developmental index (PDI) <70 on the BSID-II. For BSID-III, based on the criteria proposed by Yi et al. [10], a cognitive language composite (CLC) score of <78 or a motor composite (MC) score of <80 was considered indicative of developmental delay [10]. In the case of the K-DST, developmental delay was defined as the presence of at least one domain requiring ‘further in-depth evaluation required’ [11].

All statistical analyses were performed using SPSS Statistics software version 21.0 (IBM). To identify factors associated with the severity of brain injury and developmental delay, patients were categorized into two groups: those with moderate-tosevere brain injury and those with no or mild injury. These groups were compared using univariate analyses of perinatal history, demographic characteristics, treatment variables, ECMO-related factors, complications, and follow-up outcomes. Similarly, patients were divided into groups with and without developmental delay, and univariate comparisons were conducted for the same variables in addition to brain MRI findings.

Non-parametric statistical methods were used because the sample size was small, and the data were not normally distributed. Categorical variables were analyzed using Fisher’s exact test, and continuous variables were compared using the Mann–Whitney U-test. Statistical significance was set at P<0.05.

During the study period, 37 neonates received ECMO for acute cardiopulmonary failure, of whom 22 survived, resulting in a survival rate of 59.5%. Among the survivors, one patient diagnosed with genitourinary and/or brain malformation syndrome via whole genome sequencing due to congenital diaphragmatic hernia (CDH) and craniofacial anomalies and two patients who did not undergo developmental assessments in the outpatient setting were excluded. Consequently, 19 patients were included in the final analysis (Figure 1).

The mean gestational age of the study population was 37.7± 1.0 weeks, and the mean birth weight was 3,027±537 g. The underlying diseases included CDH (n=16), persistent pulmonary hypertension in the newborn (n=1), pulmonary hemorrhage of unknown etiology (n=1), and massive pulmonary hemorrhage following surgical resection of a congenital pulmonary airway malformation (n=1). The median age at ECMO initiation was 23 hours after birth, and the median duration of ECMO therapy was 12 days. The perinatal history, demographic characteristics, and clinical course of the patients are summarized in Table 1.

All patients underwent brain MRIs before discharge. Brain abnormalities were identified in 18 of the 19 patients (94.7%). Based on the MRI findings, brain injury was classified as mild in 12 patients, moderate in one patient, and severe in five patients (Figure 2). MRI analysis of patients with severe brain injury revealed various lesions, including infarction and encephalomalacia in the middle cerebral artery territory; large hemorrhagic infarctions involving the frontal, parietal, and temporal lobes; multifocal intracerebral hemorrhages in the cerebral hemispheres and cerebellum; encephalomalacic changes in the left thalamus; and T1 hyperintensities in the frontal and parietal lobes suggestive of white matter injury. In addition, a few patients showed evidence of diffuse brain atrophy, diffuse thinning of the corpus callosum, and dilatation of the bilateral lateral and third ventricles. Compared with the group with no or mild brain injury, the moderate/severe brain injury group had significantly longer durations of ECMO support and mechanical ventilation, as well as a significantly higher rate of CRRT (Table 2).

Developmental delay was identified in seven patients (36.8 %) based on BSID-II (n=5) and BSID-III (n=2) assessments. Among the patients evaluated using only the K-DST (n=3), no developmental delay was observed. All patients with developmental delay exhibited delays in the cognitive (MDI or CLC) and motor (PDI or MC) domains. The demographic characteristics, duration of mechanical ventilation, ECMO support, use of CRRT, and brain MRI findings of patients with developmental delay are provided in Table 3. The developmental delay occurred approximately three times more frequently in the moderates/evere brain injury group (66.7%, 4/6) than in the normal/mild injury group (23.1%, 3/13); however, the difference was not statistically significant (P=0.129) (Table 2). Patients with developmental delay had significantly longer durations of mechanical ventilation and NICU stay than those without developmental delay. Seizures were observed only in the developmental delay group, and this difference was statistically significant (Table 4).

Brain MRI findings and short-term neurodevelopmental outcomes were analyzed in neonates who received ECMO during the neonatal period. At the time of discharge, brain injury of varying degrees was observed in 94.7% of survivors, and approximately 37% of the cohort demonstrated a significant developmental delay.

The reported incidence of brain injury in neonates on ECMO varies depending on the imaging modality used. Studies based on brain ultrasound or computed tomography have reported brain injury rates ranging from 24% to 52% [2,12,13]. In contrast, MRI-based studies have demonstrated substantially higher detection rates of up to 79% [14]. This discrepancy in the reported incidence is likely attributable to the higher sensitivity of MRI in detecting brain injury among ECMO survivors. However, the frequency of brain injury can vary depending on factors such as whether deceased patients are included and the timing of imaging. Additional research is required to compare the detection rates of brain injury between brain ultrasonography during ECMO and MRI after discharge.

The duration of ECMO was associated with an increased risk of moderate-to-severe brain injury. This finding is consistent with the results of a retrospective multivariate analysis by Melbourne et al. [14], which remains the largest single-center study on neonatal ECMO to date. In that study, which was conducted over an 11-year period, ECMO duration was significantly associated with the severity of brain injury [14]. Prolonged ECMO support may increase the risk of intracranial hemorrhage (ICH) due to extended exposure to anticoagulants such as heparin and may increase the risk of cerebral infarction due to thrombus formation within the ECMO circuit or mechanical complications [15,16]. In a study of adult patients on ECMO, Omar et al. [15] reported that the duration of anticoagulation during ECMO was significantly longer in patients who developed ICH, with a mean duration of 9.4 days compared to 5.7 days in the control group. In a neonatal study, Biehl et al. [17] reported that 85% of patients with ICH occurred within the first 72 hours of ECMO initiation. Prospective studies are required to investigate the relationship between anticoagulation therapy duration or dosage and the risk and timing of ICH in neonatal patients receiving ECMO.

Organ dysfunction associated with brain injury may have contributed to prolonged ECMO duration. Severe ICH or hypoxic-ischemic injury is considered a contraindication to ECMO. Therefore, the brain injuries identified in this and other studies are presumed to have occurred after ECMO initiation [18]. Certain lesions, particularly mild brain injuries, were missed during pre-ECMO screening or bedside brain ultrasonography but were later detected on post-ECMO MRI [12,19].

In the current study, the frequency of CRRT was significantly higher in the group with moderate-to-severe brain injury than in the group with no or mild injury. Brain injury and acute kidney injury (AKI) are serious complications of neonatal ECMO and often share similar risk factors and clinical courses. The incidence of AKI during neonatal and pediatric ECMO ranges from 60% to 74%, with most cases occurring within the first 48 hours of ECMO initiation [20]. The incidence of severe AKI requiring CRRT due to significant fluid overload is approximately 40%, and the mortality rate in this group is nearly twice as high as that in patients with AKI not requiring CRRT [21]. Fluid overload in patients receiving ECMO has been associated with increased mortality [22]. The causes of fluid overload in this population are multifactorial and may include ischemic kidney injury secondary to an underlying disease, insufficient cardiac support, and reduced renal perfusion and urine output caused by non-pulsatile ECMO flow, particularly in patients with excessive dependence on VA-ECMO [23]. Further studies are needed to determine whether active management of AKI and fluid overload, such as fluid restriction and diuretic therapy before the initiation of CRRT, can reduce the risk of brain injury and mortality in neonates receiving ECMO.

Although not statistically significant, developmental delays occurred approximately three times more frequently in the moderate- to-severe brain injury group than in the normal or mild injury groups. However, 43% of patients with developmental delay showed no evidence of moderate-to-severe brain injury, whereas 33% of those with moderate-to-severe brain injury did not exhibit developmental delay. In term neonates with hypoxic-ischemic brain injury, the sensitivity and specificity of brain MRI for predicting developmental delay at 12 months were 88.9% and 70.8%, respectively [24]. However, the predictive value of MRI is considerably lower for mild or moderate brain injury than for severe injuries [25]. In addition, predictive accuracy varies depending on the location of the brain injury [26,27]. According to a study by Vojcek et al. [27], injury to the parietal lobe and thalamus/basal ganglia in patients with perinatal hemorrhagic stroke was associated with a higher risk of developmental delay than injury to other regions. Future studies with long-term follow-ups and analyses based on the pattern and location of brain injury may more clearly define the relationship between brain injury and developmental delays.

We observed that patients with developmental delay had significantly longer durations of mechanical ventilation and NICU stay than those without developmental delay. This suggests that in addition to brain injury, environmental factors related to treatment, such as prolonged mechanical ventilation, nutritional deficiency, and separation from parents, which are known to affect neurodevelopment in preterm and critically ill neonates, may influence long-term neurodevelopmental outcomes in ECMO-treated infants [28,29]. Limitations exist to the use of MRI alone to predict long-term neurodevelopmental outcomes in neonatal patients undergoing ECMO. In neonates, the prognostic accuracy of MRI has been shown to improve when used in conjunction with other modalities such as amplitude-integrated electroencephalography and magnetic resonance spectroscopy [25,30]. However, this study did not include additional neurodiagnostic tools in addition to conventional MRI. Future studies should seek to predict neurodevelopmental outcomes by integrating the newly defined classifications of brain injury with a patient’s clinical condition and environmental factors.

Reports on neonatal ECMO in Korea are limited. A recently published single-center study demonstrated that the implementation of a multidisciplinary ECMO team significantly improved survival in neonates with CDH [31]. In this study by Lee et al. [31], quality improvement initiatives were undertaken in four key areas: in-NICU initiation of ECMO for CDH with severe pulmonary hypertension, use of iNO, application of VA-ECMO, and delayed surgical repair of the diaphragmatic defect following ECMO weaning. As a result, CDH survival significantly increased from 66% to 83% compared to the previous period. ECMO was initiated via the VA method at an average of 29 hours after birth and continued for a mean duration of 6.4 days. Diaphragmatic repair was performed after ECMO in 90% of patients. With future improvements in neonatal transport systems and broader implementation of ECMO across centers in Korea, its utility may expand beyond CDH to other forms of severe neonatal respiratory failure.

This study has certain limitations. First, due to the small sample size, multivariate analysis could not be performed, and the statistical power to detect associations between the severity of brain injury and short-term neurodevelopmental outcomes was limited. Second, the study included only patients who survived ECMO. As a result, data on brain injury severity assessed using brain ultrasound in deceased patients were excluded, which may have limited the evaluation of the risk factors for brain injury. For example, among the excluded cases, one patient diagnosed with alveolar capillary dysplasia discontinued ECMO after genetic confirmation, and another patient with CDH who underwent prolonged ECMO died of hepatic failure. In both patients, brain lesions were not observed on ultrasonography; however, MRI was not performed. A mildto-moderate brain injury could have been detected if an MRI had been performed on these patients. Third, the assessment of brain injury severity on MRI was not based on a quantitative scoring system that incorporated lesion size, type, or specific brain regions affected. Future studies involving larger cohorts with standardized MRI protocols and quantitative classification systems are required to improve brain injury evaluation and its clinical significance. Fourth, the follow-up period for neurodevelopmental assessment was insufficient. Due to the retrospective nature of the study, the developmental evaluation tools used were not standardized across the study period.

In conclusion, this study identified that various types and severities of brain injury can be observed in neonatal survivors on ECMO, which is consistent with the findings of previous studies. Further studies are required to investigate the risk factors for developmental delay by integrating a structured classification of brain injury patterns with a thorough evaluation of clinical and environmental treatment factors.