Extracorporeal membrane oxygenation (ECMO) is a method of external mechanical circulatory support (MCS) designed to oxygenate and circulate blood, thus supplementing insufficient respiratory and cardiac functions during critical illness. Its use has increased in the preceding decade as technology and techniques have improved. However, approximately 13% of ECMO recipients are found to have neurologic injuries either from underlying pathology or the ECMO therapy itself (Migdady et al., 2019; Shoskes et al., 2020; Kietaibl et al., 2021). Performing accurate neurologic examinations in this population is challenging due to sedation and neuromuscular blockade (NMB) requirements to prevent cannula dislodgement, optimize hemodynamics, or to force ventilator compliance (Debacker et al., 2018; Patel et al., 2020). Neuroimaging requires resource-intensive, multi-disciplinary transportation to imaging suites (Prodhan et al., 2010), while invasive neuromonitoring can further increase the risk of cerebral hemorrhage due to anticoagulation requirements while on ECMO (Thomas et al., 2018). Thus, the brain’s perfusion and oxygenation requirements are not routinely monitored in current clinical practice, furthering the potential for injury.

Diffuse correlation spectroscopy (DCS) provides non-invasive, continuous bedside neuromonitoring using near-infrared light, modeled with the correlation diffusion equation (Durduran et al., 2010a), to measure blood flow index (BFI), which is directly proportional to cerebral blood flow (CBF) (Durduran and Yodh, 2014). Its use to monitor CBF has been described in a number of brain-injured cohorts (Durduran et al., 2009; Delgado-Mederos et al., 2018; Baker et al., 2019) and has been validated against other CBF measurement modalities including transcranial Doppler (TCD) (Durduran et al., 2010b; Kim et al., 2010; Mesquita et al., 2011). DCS is closely related to near-infrared spectroscopy (NIRS), a commercially available technology that measures cerebral oxygenation, but DCS offers the advantage of measuring CBF directly rather than using oximetry as a surrogate marker.

Veno-arterial (VA) ECMO is often configured to supply oxygenated blood in retrograde flow up the aorta via a cannula placed in the femoral artery. This is contrasted with veno-venous (VV) ECMO, in which oxygenated blood is returned to the body via the venous system and perfuses the arterial system in a normal physiologic manner. This distinction is important as asymmetric cerebral and somatic oxygenation and perfusion have been described in VA ECMO (Cove, 2015; Acharya et al., 2017; Nezami et al., 2021). The presence of disrupted cerebrovascular autoregulation in comatose patients with hypoxic brain injury furthers the possibility of asymmetric perfusion arising from the brain’s interaction with the ECMO circuit and the unique changes in circulatory physiology ECMO produces (Van Den Brule et al., 2018). We used DCS to record hemispheric CBF in adult ECMO patients with the hypothesis that comatose patients will exhibit more interhemispheric asymmetry than in non-comatose patients.

Out of 62 of eligible patients undergoing ECMO treatment for different indications, thirteen (21%) were consented for participation (five females; mean age 46 years [range 21–78]). Patient demographics and pertinent clinical parameters are shown in Table 1. Seven subjects required ECMO for cardiac arrest, four for cardiogenic shock and two for acute respiratory distress syndrome (ARDS). VA ECMO was performed on 11 patients and VV ECMO on the two patients with ARDS. The mean length of hospital stay was 27 days. The mean duration of ECMO treatment was 266 h. Mean pre-ECMO SOFA score was 12. Nine subjects (69%) were classified as “awake” and four (31%) as “comatose” as defined above. Nine subjects survived to ECMO decannulation and seven to hospital discharge. A total of 63 monitoring sessions were conducted, with an average of five sessions per subject. Data from seven sessions were omitted from analysis due to low SNR (subject 3, days 1 through 6 and subject 11, day 3). Data for the first 6 days of Subject 3’s recording were discarded due to jaundice, which confounded DCS signal strength. Subject 11 had low signal quality that could not be corrected from the right hemisphere on day 3 which was also discarded.

Daily GCS-M scores recorded during ECMO treatment for each subject are shown in Supplementary Figure 1. Subjects 1, 2, and 4 were awake and remained so, while subjects 8, 9, 11, and 13 remained comatose throughout monitoring. GCS-M scores for subjects 3, 5, 6, 7, 10, and 12 fluctuated but were able to improve to follow commands at some point (GCS-M = 6). Notably, subjects 7 and 10 awoke to follow commands after paralytics and sedation were weaned, while the exam for subject 5 improved spontaneously on day 4.

Table 2 details the clinical characteristics for all subjects. These include cumulative sedation the subjects received during the monitoring period, ABG results, and ECMO circuit pCO₂ measured both pre and post-oxygenator on the day of maximum observed ASYM_rBF. Supplementary Table 1 shows the mean values of these clinical characteristics for the comatose and non-comatose groups. None of the values were significantly different between the two groups.

As archetypes of the observed asymmetry phenomenon, Figure 2 highlights ASYM_rBF vs. GCS-M data for subjects 1 (awake, conversant) and 11 (comatose, severely brain injured due to cardiac arrest with hypoxic-ischemic encephalopathy confirmed by CT and MRI neuroimaging). Daily ASYM_rBF is shown for all subjects in Supplementary Figure 2. For each subject’s day of maximum ASYM_rBF, Supplementary Figure 3 shows correlation between hemispheric rBF and MAP and Figure 3 displays |ΔrBF_MAP| . Figure 4 summarizes these data with a group-level comparison: comatose subjects experienced significantly higher ASYM_rBF values compared to awake subjects (29% [IQR 23–34%] vs. 11% [IQR 8–13%], one-tailed p = 0.009, two-tailed p = 0.018).

Time-courses of ΔrBF_MAP during the total monitoring period on the day of maximum asymmetry are shown for all patients in Supplementary Figure 4. Among comatose patients, subject 8 showed rBF skewed toward the right hemisphere at lower MAP values while subject 9, 11, and 13 showed rBF values skewed toward the left hemisphere. However, these patients showed equal rBF of both hemispheres at around a MAP value of 80 mmHg. On the contrary, non-comatose patients show an equal rBF throughout the entire MAP range compared to the comatose patients.

Regarding TCD and DCS comparison analysis, an example of the difference in MAP range experienced by patients is illustrated in Figure 5A. Due to the limitations in obtaining continuous TCD measurements in the MAP range, only values over a limited MAP range were acquired compared to DCS. The shaded region denotes the overlap between the asymmetry quantified by DCS and TCD values for subject 1. A Bland–Altman plot comparing |ΔrBF_MAP| and |ΔrCBFV_MAP| for all patient data is shown in Figure 5B, revealing most data to fall within the 95% confidence interval. A non-parametric Wilcox rank sum test resulted in a p-value of 0.54, indicating that the two methods do not significantly differ in their measurement of CBF asymmetry. A boxplot showing the individual subjects’ average difference between DCS and TCD asymmetry is shown in Supplementary Figure 5.

Univariate linear modeling with Pearson correlation analysis was performed with ASYM_rBF vs. ABG and post-oxygenator pH, pCO₂, and pO₂ values, as well as LVEF. Subsets of these analyses are shown in Figure 6. There were no significant associations between ASYM_rBF and pO₂ or LVEF in either comatose or non-comatose groups and are not shown. Neither were any associations found to be significant for ABG data: ASYM_rBF vs. ABG pCO₂, comatose slope = 0.26 %/mmHg (R = 0.22, p = 0.78), non-comatose slope = 0.12%/mmHg (R = 0.53, p = 0.15); ASYM_rBF vs. ABG pH, comatose slope = 51.04%/(mmol/L) (R = 0.24, p = 0.76), non-comatose slope = −2.00%/(mmol/L) (R = −0.03, p = 0.93). In addition, there was no significant association found for post-oxygenator data although excellent R values were noted for some cases: ASYM_rBF vs. post-oxygenator pCO₂, comatose slope = 0.75%/mmHg (R = 0.91, p = 0.09), non-comatose slope = 0.19%/mmHg (R = 0.60, p = 0.09); ASYM_rBF vs. post-oxygenator pH, comatose slope = −105.27%/(mmol/L) (R = −0.89, p = 0.11), non-comatose slope = −11.87%/(mmol/L) (R = −0.26, p = 0.50). However, comparison of linear model slopes between the comatose and non-comatose subgroups did reveal a significant difference both for post-oxygenator pCO₂ (p = 0.03) and post-oxygenator pH (p = 0.04).

Two of four comatose subjects did not survive to ECMO decannulation and underwent withdrawal of care due to septic shock (subject 9) or severe neurologic injury (subject 13). Subject 8 survived decannulation and regained consciousness, but had care withdrawn in context of subsequent recurrent cardiogenic shock. Subject 11 survived decannulation, remained comatose and was discharged to a coma rehabilitation program but died from recurrent cardiac arrest within 1 week of discharge. Three of the nine non-comatose subjects died prior to discharge due to septic shock (subject 2), hemorrhagic shock (subject 3), or recurrent cardiogenic shock (subject 12).

In this pilot study of 13 patients undergoing ECMO, those who remained comatose throughout their duration of treatment were found to have relatively high hemispheric asymmetry in rBF over a range of blood pressures. In these four subjects, the degree of asymmetry appears to change with MAP, but rBF was elevated on the left hemisphere in 3 (75%) of them. Though ASYM_rBF was found to be elevated in comatose subjects, this metric did not associate with survival to discharge. To our knowledge, this is the first study to demonstrate this phenomenon using continuous, non-invasive CBF monitoring via DCS.

Multiple possible physiologic reasons may explain the presence of asymmetric perfusion, though our study was not powered to elucidate specific justifications of this phenomenon. All four comatose subjects underwent peripheral VA ECMO, whereby blood is ejected from the ECMO circuit directly into arterial circulation retrograde up the aorta from the femoral artery. Under the pressure of ECMO device, the blood flows against intrinsic cardiac outflow in a mixing cloud (Alwardt et al., 2013). Increasing cardiac contractility pushes this cloud distally, supplying the brain with native cardiac blood via the carotid arteries which diverge from the proximal aorta. Conversely, increasing ECMO flow pushes the cloud proximally toward the heart, perfusing the brain with ECMO circulation. Native and ECMO blood typically have different pH levels, oxygen saturations, and amounts of dissolved carbon dioxide. If the mixing cloud lies in between the brachiocephalic trunk and left common carotid artery, the two sides of the brain may receive blood under different pressures and with different levels of oxygen, CO₂, pulsatility, and pressure (Stevens et al., 2017; Nezami et al., 2021). We did not find significant differences in comatose vs. non-comatose patients in average pH, pCO₂ from multiple sources, ECMO flows, MAP range, or pulsatility (Supplementary Table 1). Comatose subjects did experience a lower range of pulse pressures, but this did not correlate with ASYM_rBF and may have been because they were unconscious and less responsive to external stimuli. A previous study looking at angiography of the brain in VA ECMO patients illustrated this asymmetric blood flow via filling time of injected iodine contrast (Acharya et al., 2017). Thus, it is possible that the left hemisphere receives a higher proportion of blood from the ECMO circuit, and at higher flow rates, than the right hemisphere.

Underlying cardiogenic shock, cardiac arrest, and ARDS may predispose the brain to hypoxic-ischemic brain injury (HIBI) before ECMO is initiated, which can prime the brain for secondary injury (Inoue et al., 2020). Cerebrovascular dysregulation is a feature of HIBI (Sundgreen et al., 2001) and, in the setting of asymmetric ECMO blood flow, may lead to interhemispheric differences in CBF due to impaired cerebral autoregulation. In the setting of disrupted cerebral autoregulation post-arrest, optimal MAPs for cerebral perfusion are often higher than the clinically accepted norm of 65 mmHg (Sundgreen et al., 2001; Peberdy et al., 2010; Sekhon et al., 2019). Furthermore, interhemispheric communication may be disrupted resulting in regional independence of function and thus BFI oscillations that are independent on each side of the brain. Disruption in interhemispheric communication is seen in patients with traumatic brain and concussion (Schmidt et al., 2021; Wang et al., 2021), and thus may play a role in other global brain injuries such as HIBI. In addition, asymmetry in the intrinsic ischemic damage may result in differential dysregulation and cerebrovascular coupling, leading to asymmetry in the cerebral metabolic rate of oxygen consumption (CMRO₂) and thus oxygen extraction fraction and local CO₂ production. CO₂ vasodilates cerebral arterioles and precapillary sphincters by increasing local [H⁺], triggering voltage-gated [K⁺] channels to hyperpolarize endothelial cells leading to vascular relaxation (Battisti-Charbonney et al., 2011). There are many other contributing factors affecting diameter of the blood vessels, including reduced cerebrovascular reactivity to vasomotor stimuli (Girouard and Iadecola, 2006), reduced CMRO₂ in regions of cell death (Duckrow et al., 1981) and reduced CBF in proportion to neural activity (and thus CMRO₂) (Bundo et al., 2002). Augmented by asymmetric perfusion resulting from the ECMO circuit, these differences may become even more pronounced as noted in our comatose patient cohort.

Our data showed potential for an increased sensitivity to the effects of post-oxygenator pH and pCO₂ on rBF asymmetry with significant differences in linear model slopes between comatose and non-comatose patients. However, the associations themselves were not found to be significant. This discrepancy is likely due to low degrees of freedom secondary to small sample sizes when assessing each individual association’s significance, vs. increased degrees of freedom when comparing the two models against one another. ABG pH and pCO₂ obtained from the patient’s right radial artery, representing native circulation, were not correlated with ASYM_rBF, and importantly, nor were any differences noted between the two subgroup models. Of note, comatose patients had a higher, though nonsignificant, difference in PCO₂ between pre- and post-oxygenator blood (Supplementary Table 1), implicating the degree of gas exchange in our findings. One possible reason for comatose patients having increased sensitivity to post-oxygenator pH and pCO₂ is that these patients are more likely to be outside the zone of cerebral autoregulation (Van Den Brule et al., 2018). As implied above, it is possible that the left hemisphere is being exposed to the post-oxygenator blood more than the right side, and thus the effects of pCO₂ on brain perfusion and autoregulation may be asymmetric with the difference being highlighted by an unstable neurovascular unit. While this evidence is insufficient to claim that differential association exists, it warrants further investigation. Should this finding hold true in larger future studies, it could provide novel evidence that post-oxygenator parameters should be controlled more tightly and in a more individualized manner, particularly for patients with suspected brain injury.

Previous studies have measured cerebral autoregulation using cross-correlation analysis between continuously measured blood pressure and either invasive (intracranial pressure, brain tissue oximetry) or non-invasive (NIRS-based cerebral oximetry, TCD-based CBF velocity) (Zweifel et al., 2008; Aries et al., 2010; Andresen et al., 2018; Montgomery et al., 2020) neuromonitoring modalities. We did not use this method because MAP variation was low (<5 mmHg) for long periods during monitoring, thus did not yield wide enough MAP ranges to generate full autoregulation curves. We did not utilize a circuit manipulation protocol to challenge MAP as has been performed in other studies (Papademetriou et al., 2012; Busch et al., 2020) because of the possible danger of secondary brain injury from hypo- or hyperperfusion. Though our methodology was unable to explore the upper and lower limits of autoregulation for each patient, optimal MAPs that minimized asymmetry can be identified in Supplementary Figure 4. The variability of this optimal MAP may represent the customized ranges of individual blood pressure targets for optimal cerebral perfusion observed in ECMO patients with HIBI (Rikhraj et al., 2021). Past clinical trials of increased MAP goals for cardiac arrest survivors resulted in improved cerebral oxygenation, but not improved outcomes (Jakkula et al., 2018; Ameloot et al., 2019). However, none of these studies utilized individualized, autoregulation-based MAP targets (Russo et al., 2017, 2018). The hypothesized benefit of CBF augmentation in this population is predicated on patchy hypoperfusion due to the no-reflow pathophysiology (Kloner et al., 2018). In contrast, the physiology by which asymmetric dysregulation may occur in our cohort remains to be elucidated.

Several limitations to our study must be considered when interpreting these data. Our study captured a small population with heterogeneous pathologies ranging from cardiac arrest to acute heart or lung failure. Not all patients were treated with the same modality of ECMO, and the physiology of VA ECMO differs significantly from VV ECMO as the latter relies on the native heart for forward uniform circulation. We analyzed ASYM_rBF in comatose vs. non-comatose patients excluding VV ECMO recipients in order to evaluate the effect of ECMO type on our findings and found comatose patients still had higher ASYM_rBF (n = 10, 28.7 ± 8.2 vs. 10.2 ± 2.5, two-tailed p = 0.04). We chose to keep VV ECMO patients in our study design as comparators with a priori risk of ischemic injury. Subjects varied by other factors such as cardiac function, level of sedation, receipt of NMB, and severity of illness (e.g., pre-ECMO SOFA score). Our small sample size precluded the ability to analyze these factors as covariates for association between ASYM_rBF and GCS-M. Furthermore, our monitoring methodology was not equipped to differentiate the source of cerebral perfusion (ECMO circuit vs. native circulation). Future studies should include specific ECMO populations with a single ECMO subtype and similar etiology necessitating ECMO.

One challenge we faced in designing our study was the need to dichotomize patients into “comatose” vs. “non-comatose” groups. Fundamentally, coma is a disordered state of consciousness in which the individual has no interaction with their environment (Provencio et al., 2020). A robust clinical definition of coma in the acute setting has not yet been developed and has often been disease-specific in prior research (Provencio et al., 2020). We relied on prior clinical trials (Nielsen et al., 2013; Lascarrou et al., 2019; Dankiewicz et al., 2021), to establish parameters for cohort dichotomization.

To derive rBF, the median BFI value was defined as the baseline for each monitoring session as was described in the method section. However, previous DCS studies define baseline differently (Durduran et al., 2009; Kim et al., 2010; Busch et al., 2019). In these studies, the baseline BFI is defined as a period of the measurement where the subject is at rest. This period of rest precedes a defined perturbation, allowing comparison of the relative change of BFI between the baseline and perturbation. Since this study did not include controlled perturbations, defining such a baseline period for these subjects is difficult and arbitrary. Additionally, there is no indication that the selected baseline region corresponds to when the subject is truly at rest, thus introducing user selection bias. We did not observe statistical significance in |ASYM_rBF| between the two groups when using this method of normalization.

Alternatively, the mean value of the BFI time series could be used as a baseline for calculating rBF (Giovannella et al., 2020) but this would be disproportionately affected if there are spontaneous changes in BFI. The median is less affected by such changes than the mean. We compared the median with the mean resting baseline values calculated in the following manner. First, each monitoring period was smoothed using a moving average filter and regions where drastic flow changes occurred were removed. This modified signal was then averaged to define a baseline to calculate rBF. The difference between the median and the mean resting baseline varied case-by-case (i.e., sometimes they matched well, and other times they differed significantly). Despite the difference, ASYM_rBF computed using the mean resting baseline for normalization showed statistical significance between the two groups (comatose vs. non-comatose) as was shown using the median value (p < 0.006 for one-tailed and p < 0.01 for two-tailed). This affirms that using the median value as the baseline definition did not alter the major findings of this study. This might be due to our data having relatively few incidents of drastic changes in BFI as many patients were sedated or kept stable. While this mean resting baseline method seems to capture the baseline well, it does require user input when deciding on regions of drastic BFI changes. Thus, we chose to use the median as the baseline to allow for consistent selection for all the measurement periods. Future studies will continue to investigate the optimal method to estimate baseline values.

Measuring blood flow with DCS also presents its own set of limitations. As with all near-infrared spectroscopic methods, DCS is limited in its depth of penetration to cortical tissue (Buckley et al., 2014). Blood flow in subcortical regions is not well-captured. Moreover, spatial resolution was limited to the forehead because DCS is much more sensitive to interference from hair than continuous wave NIRS, which has been employed with multiple detectors throughout the head in other studies (Papademetriou et al., 2013). The lack of spatial resolution is less likely to be a limitation in whole-brain injury, including states of reduced consciousness or coma, as the entire hemisphere may be affected to a similar magnitude (Kaloostian et al., 2012; Ohshima et al., 2012; Foley et al., 2013; Sato et al., 2014; Baker et al., 2019).

Our simultaneous measurements of CBFV with TCD enabled us to compare the asymmetry in CBF measured by our experimental device vs. clinical gold standard. Using Bland–Altman analysis, we found that the two methods did not significantly differ in their measurement of asymmetry and thus lent further support to the validity of DCS measurements. However, it must be noted that this comparison was done over a range of MAP values limited to those the patient experienced during simultaneous TCD and DCS measurement, a limitation of manually placed TCD probe measurement. Future use of either phased array Doppler ultrasound (Pietrangelo et al., 2018) or headframe-assisted probe placement would facilitate longer-term measurements and enable a more robust comparison between DCS and TCD measurements of CBF.

Another limitation of our DCS measurements was the use of constant optical coefficients of scattering and absorption. These coefficients were based on data in the literature (Brady et al., 2010; Parthasarathy et al., 2018; Selb et al., 2018). Incorrect values can lead to error in the BFI calculation, particularly the scattering coefficient. It is reasonable to assume the scattering coefficient does not change during brain monitoring. However, absorption may change based on the changes in total hemoglobin concentration or blood oxygen saturation. Relative changes in blood flow are less sensitive to absorption changes, compared to absolute BFI values. Our results remain valid since we are quantifying the difference between hemispheres based on relative changes in blood flow. Nevertheless, concurrently quantifying absorption and scattering coefficients will improve the accuracy of the BFI. In the future, a frequency domain NIRS system will be added to extract coefficients of absorption and scattering alongside DCS measurements.

The manner in which DCS was utilized in this study does not allow for measurements of absolute blood flow, but instead uses a relative measure of CBF. Our asymmetry measure is based on the rBF normalized to the median value for each day, which allows us to better compare between days and subjects. Absolute blood flow calibration can require imaging modalities such as CT perfusion (Verdecchia et al., 2016) or MRI arterial spin labeling (Durduran et al., 2004; Carp et al., 2010). We decided against using these techniques for this pilot study to minimize transport to CT scanners and avoid nephrotoxic exposure to intravenous contrast dye; furthermore, the ECMO circuit is not MRI compatible. A recent study utilized time-resolved dynamic contrast enhanced NIRS (DCE-NIRS) in conjunction with a one-time bolus of indocyanine green to calibrate DCS measurements in critically ill adults (He et al., 2018). The use of indocyanine green may be contraindicated in patients with renal dysfunction, which is common among patients with severe cardiac disease (Hu et al., 2016; Sandroni et al., 2016). Nevertheless, absolute CBF calibration may be important in quantifying differences in CBF as the cause of asymmetric perfusion.

In this study of adult ECMO recipients, comatose patients demonstrated increased cerebral hemispheric perfusion asymmetry as measured with DCS when compared to non-comatose patients. Future studies will include larger sample sizes, other modalities of neuromonitoring, and patients with specific conditions to validate hemispheric asymmetry as a biomarker of neuronal dysfunction in this population. Ultimately, it remains to be seen whether this marker is a predictor of coma recovery, informs pathophysiological basis of asymmetric perfusion, and provides a goal for targeted ECMO therapy.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the University of Rochester Research Subjects Review Board. The patients/participants provided their written informed consent to participate in this study.

TJ contributed to study concept and acquisition, analysis, and interpretation of the data, and played a significant role in drafting this manuscript. ID contributed to study design and acquisition, analysis, and interpretation of the data, and played a significant role in drafting this manuscript. KD, YX, and ES contributed to data acquisition and analysis. OS contributed to study concept and design, participated in patient recruitment, as well as data interpretation, and revised this manuscript for intellectual content. MM and RM contributed to study concept and design, as well as data interpretation, and revised this manuscript for intellectual content. TW contributed to statistical analysis. IG contributed to data acquisition and study design, and revised this manuscript for intellectual content. GS contributed to study concept and design, and revised this manuscript for intellectual content. RC contributed to study concept, design, data acquisition, analysis and interpretation, and revised this manuscript for intellectual content. IK contributed to study design, concept, data acquisition, interpretation, and drafted and revised this manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.