2.1.

Participants

This study included patients 0 to 15 years of age who were placed on ECMO from January 2014 to May 2016 at Children’s Health, Dallas. The reasons for ECMO intervention were persistent pulmonary hypertension of the newborn (PPHN), septic shock and/or acute respiratory distress syndrome (ARDS). Patients with pre-existing neurological injuries, underlying congenital heart disease, and CPR patients were excluded due to their predisposition to pre-ECMO cerebral injuries.^17–19 All the patients were placed on Rotaflow centrifugal pumps. Cannulation for venoarterial (VA) ECMO was through the carotid artery and internal jugular vein. Cannulation for venovenous (VV) ECMO was through the double lumen venous catheters placed in the right internal jugular vein. Most of the patients were sedated with fentanyl and versed. Dexmedetomidine was used as an adjunct in some patients. Heparin was the anticoagulation drug of choice. The loading dose at the time of cannulation was 50 to $100\text{units}/\mathrm{kg}$. Heparin infusion was titrated for anticoagulation goals. Bedside activated clotting time in seconds, partial thromboplastin time (PTT) in seconds, and unfractionated heparin (UH) levels (international units/ml) were the measurement parameters. The study was approved by the institutional review board at the University of Texas Southwestern Medical Center, Dallas, and informed consent was waived.

2.2.

Arterial Blood Gas and Anticoagulation

Measurements of arterial blood gas were obtained from 24-h pre-ECMO to the first 24 h on ECMO as the biggest change occurred during this time period. This included pH, partial pressure of oxygen (${\mathrm{PaO}}_2$), and partial pressure of carbon dioxide (${\mathrm{PCO}}_2$). Similarly, the minimum, maximum, and mean values of the coagulation indices (PTT and UH) throughout the ECMO run were also calculated.

2.3.

Autoregulation Monitoring

Mean arterial pressure (MAP) was continuously measured from an indwelling arterial catheter. As a surrogate for CBF, cerebral tissue oxygen saturation (${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$) was measured on the forehead using a cerebral oximeter (INVOS^™ 4100-5100, Somanetics, Troy, Michigan). Both signals were sampled minute-by-minute and recorded throughout the ECMO run except during patient transportation for imaging or other procedures. The recorded data were inspected for spike-like artifacts, which were removed by linear interpolation between neighboring data points.¹² Then, a second-order polynomial detrending was applied to remove the slow drifts in each signal.

2.4.

Wavelet Transform Coherence

A detailed description of WTC in studying dynamic cerebral autoregulation is outlined in our previous publication.¹⁶ Briefly, it characterizes the squared cross-wavelet coherence and relative phase between two paired signals in time-frequency domain.¹⁵ For a time series $x(n)$ of length $N$, which is sampled from a continuous signal at a time step of $\mathrm{\Delta }t$, the continuous wavelet transform is defined as

In analogy to Fourier analysis, the cross-wavelet transform of two time series, $x(n)$ and $y(n)$, is defined as

The squared cross-wavelet coherence, $R^2(n,s)$, is defined as

The statistical significance of $R^2(n,s)$ between the two paired signals, which were the spontaneous MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ fluctuations in this study, against background noise can be assessed using the Monte Carlo method.¹⁵ Briefly, this method generates a large ensemble of surrogate data pairs ($n=300$) that have the same coefficients as the actual input signals based on the first-order autoregressive (AR1) model. Wavelet coherence is calculated for all of the surrogate data pairs. Then, the significance level of $R^2(n,s)$ of the actual input signals is determined by comparing with those from the surrogate data pairs at each time and wavelet scale. Here, a 95% confidence interval ($p<0.05$) is used for statistical testing. Based on our observation, this confidence interval corresponded to a critical value of $R^2(n,s)$ narrowly between 0.71 and 0.73 across all the patients.

The relative phase between the two paired signals, defined as $\mathrm{\Delta }\phi (n,s)={\tan }^{-1}\{\frac{\text{Im}[W^{XY}(n,s)]}{\text{Re}[W^{XY}(n,s)]}\}$, ranges between $-\pi$ and $\pi$. In the previous publication,¹⁶ we characterized the phase relationship in four separate ranges: $\mathrm{\Delta }\varphi =0\pm \pi /4$, $\pi /2\pm \pi /4$, $\pi \pm \pi /4$, and $-\pi /2\pm \pi /4$. Each phase range represents a distinct pattern of coherence and might be related to different underlying physiological mechanisms: $\mathrm{\Delta }\varphi =0\pm \pi /4$ represents an in-phase pattern where the two signals oscillate in the same directions, $\mathrm{\Delta }\varphi =\pi \pm \pi /4$ represents an antiphase pattern where the two signals oscillate in opposite directions, and both $\mathrm{\Delta }\varphi =\pi /2\pm \pi /4$ and $-\pi /2\pm \pi /4$ represent an asynchronous pattern where the two signals have significant phase differences.

In this study, we focused on a pressure-passive state of cerebral autoregulation,^11,20,21 i.e., the patient’s changes in blood pressure cause simultaneous changes in cerebral oxygenation in the same directions. The pressure-passive state is a vital sign of an impaired autoregulation system²¹ and results in significant in-phase coherence between the MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ signals.¹⁶ Thus, a phase range of $\mathrm{\Delta }\varphi \in 0\pm \pi /4$ was selected as the range of interest. Within this phase range, the percentage of significant coherence, $P(s)$, was quantified as the percentage of time during which the $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence was statistically significant from the background noise ($p<0.05$) via the Monte Carlo method.¹⁶ $P(s)$ was a function of wavelet scale $s$ (i.e., a function of $1/f$) and represented the scale/frequency characteristics of $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence. Cerebral autoregulation is widely recognized as a frequency-dependent phenomenon.^13,14 Previous studies have shown that the $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence with a scale range of $\le 30\min$ was most relevant to the pressure-passive state.^11,20,21 In this study, we calculated the mean value of $P(s)$, $P_{\text{mean}}$, in a scale range of 8 to 32 min as an index of cerebral autoregulation impairment.¹⁶

2.5.

Neuroimaging Assessment and Scoring

For all neonates and children with open fontanels, head ultrasound was done every day for the first three days of their ECMO run and more frequently if clinically indicated. If there was any ongoing concern of their neurological status, an emergent computed tomography (CT) scan was conducted while on ECMO, which consisted of 3-mm contiguous axial images from C1 to the cranial vertex (Somatom, Siemens, Erlangen, Germany). Post-ECMO magnetic resonance imaging (MRI) at 1.5 or 3 T was done on all the patients prior to their discharge (Achieva and Intera, Philips Healthcare, Best, the Netherlands). The standard protocol consisted of T1-weighted sagittal, T2-weighted or fluid-attenuated inversion recovery axial images, T2-weighted coronal, and diffusion-weighted axial images. Some patients also had susceptibility-weighted axial images. Neonates had T1-weighted axial imaging.

Neurological abnormalities on brain imaging were evaluated by a neuroradiologist (M.C.M), who was blinded to the patients’ clinical conditions and autoregulation measurements, based on a scoring system that had been previously validated among ECMO patients.^22,23 It included three categories of abnormalities: bleeding, parenchymal lesions (ischemia or infarction), and ventricular dilatation. Each category was assigned a score ranging from 1 to 3 and then multiplied by a relative weight factor and added to the total score. In case a patient had both CT imaging during ECMO and MRI post-ECMO, the CT imaging was analyzed for scoring purposes. Neuroimaging scores were grouped into the following three categories: normal ($\text{score}=0$), mild to moderate ($\text{score}=0.5$ to 6.0), and severe ($\text{score}=6.0$ to 18.0).²³

2.6.

Statistical Analysis

Both univariate and multivariate linear regression analyses were conducted to investigate: (1) the relationship between the individual cerebral autoregulation indices and neuroimaging scores, which were the primary outcomes in this study and (2) the effects of ECMO type (VV and VA) and duration, blood gas changes (pH, ${\mathrm{PaO}}_2$, and ${\mathrm{PCO}}_2$), and coagulation indices (PTT and UH) on the individual autoregulation indices and neuroimaging scores. These analyses were conducted in two separate age groups (neonates and children) as well as in the entire cohort.

3.1.

Patient Characteristics

Twenty-nine pediatric patients who met the inclusion criteria were enrolled during the study period. Four were not included for data analysis due to loss of MAP data (two) or ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ data (two). The final sample size was 25 (8 males and 17 females). Among these patients, 9 had VV ECMO, 12 had VA ECMO, and the other 4 initially had VV ECMO and then were converted to VA ECMO. Neonates ($\le 4\text{weeks}$ of age) were the biggest subpopulation and had relatively unifying diagnosis (10 out of 11 were placed on ECMO for PPHN). Children had more variable diagnoses. Individual characteristics, primary and secondary diagnoses, ECMO type and duration, and patient survival to hospital discharge are detailed in Table 1.

Table 1

Characteristics, primary and secondary diagnoses, type and duration of ECMO, and survival status (yes or no) of the neonatal patients.

Table 2 shows the neuroimaging modality, timing, and score for each individual patient. In all but one patient, MRI was done within a week of coming off ECMO support. Per this table, 8 patients (32%) had normal neuroimaging, 7 (28%) had mild to moderate neuroimaging abnormalities, and the other 10 (40%) had severe neuroimaging abnormalities. The ECMO type (VA and VV) and duration did not show significant effect on neuroimaging scores.

Table 2

Intra-ECMO autoregulation indices and neuroimaging scores of the patients.

3.2.

Autoregulation Impairment During Extracorporeal Membrane Oxygenation

The spontaneous fluctuations of MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ during ECMO were mostly incoherent among patients who had normal neuroimaging scores. In contrast, patients with significant in-phase $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence always had abnormal neuroimaging scores. Additionally, one neonate and two children had abnormal neuroimaging scores but low in-phase $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence. The neonate (patient 17) had PPHN secondary to group B Streptococcus (GBS) infection with concern for possible central nervous system GBS infection, which could have predisposed the infant to abnormal neuroimaging. The two children (patients 8 and 15) had leukemia with significant brain volume loss from their underlying immunocompromised state.

For instance, results from two representative patients are shown in Figs. 1 and 2 respectively, and described in detail below.

Fig. 1

Autoregulation and neuroimaging results from a patient (patient 6: neonatal male) who was placed on VV ECMO for meconium aspiration secondary to PPHN. (a) Partial enlarged figure of WTC between the spontaneous fluctuations of MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$. In this graph, the $x$-axis represents the time, the $y$-axis represents the wavelet scale (in inverse proportion to Fourier frequency), the color scale represents the squared cross-wavelet coherence ($R^2$) that ranges from 0 to 1, and the black line contours designate the areas of significant coherence ($p<0.05$) identified through Monte Carlo simulation. The arrows designate the relative phase between MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$: a rightward-pointing arrow indicates in-phase coherence and leftward-pointing arrow indicates antiphase coherence. (b) A segment of real-time MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ data. (c) MRI brain image acquired 2 days after ECMO.

Fig. 2

Autoregulation and neuroimaging results from a patient (patient 5: 8-year-old female) who was placed on VV ECMO for pulmonary contusion secondary to ARDS s/p motor vehicle collision. (a) Partial enlarged figures of WTC between the spontaneous fluctuations of MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$. In this graph, the $x$-axis represents the time, the $y$-axis represents the wavelet scale (in inverse proportion to Fourier frequency), the color scale represents the squared cross-wavelet coherence ($R^2$) that ranges from 0 to 1, and the black line contours designate the areas of significant coherence ($p<0.05$) identified through Monte Carlo simulation. The arrows designate the relative phase between MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$: a rightward-pointing arrow indicates in-phase coherence and leftward-pointing arrow indicates antiphase coherence. (b) A segment of real-time MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ data. (c) CT brain image acquired during ECMO.

Patient 6 was a neonatal male who was placed on VV ECMO for meconium aspiration secondary to PPHN. Incoherent MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ changes were seen during the majority of the time of ECMO run. This was confirmed in a segment of real-time MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ data. The patient survived, and his post-ECMO MRI was normal.

Patient 5 was an 8-year-old female who was placed on VV ECMO for pulmonary contusion secondary to ARDS s/p motor vehicle collision. Her initial head CT on admission had no evidence of intracranial bleeding. She was found to have a large intracranial bleeding on day 6 and was emergently weaned off from ECMO and decannulated. Intermittent in-phase coherence between the MAP and ${\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ changes was seen in a scale range of 8 to 64 min throughout her ECMO run. It became much more apparent on day 6, $\sim 10\mathrm{h}$ prior to decannulation. Her CT image after decannulation showed extensive intraventricular and right frontal parenchymal hemorrhage with obstructive hydrocephalus.

At the group level, the most predominant in-phase $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence was seen in a scale range of 8 to 32 min (Fig. 3), which was consistent with previous findings on pressure-passive autoregulation.^11,20,21 Therefore, the mean percentage of significant in-phase $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence was calculated in this scale range as an index of intra-ECMO autoregulation impairment.

Fig. 3

Percentage of significant coherence, $P(s)$, derived from the in-phase $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence (i.e., $\mathrm{\Delta }\phi \in 0\pm \pi /4$). In this graph, the $x$-axis represents the wavelet scale, $s$, which is in inverse proportion to Fourier frequency. The $y$-axis represents the percentage of time during which the $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence was statistically significant over the background noise ($p<0.05$). Therefore, $P(s)$ represented the scale/frequency characteristics of the $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence. For ECMO patients, predominant in-phase $\mathrm{MAP}\to {\mathrm{S}}_{\mathrm{ct}}{\mathrm{O}}_2$ coherence was seen in a wavelet scale range of 8 to 32 min (the shaded area), which corresponded to a frequency range of 0.0005 to 0.002 Hz. This range was selected to calculate the autoregulation index.

3.3.

Autoregulation Index versus Neuroimaging Score

As shown in Fig. 4, significant correlation between the autoregulation indices and neuroimaging scores was found over the entire cohort ($R=0.66$; $p<0.0001$). Thus, high degrees of cerebral autoregulation impairment during ECMO were indicative of severe neuroimaging abnormalities.

Fig. 4

Correlation between autoregulation index and neuroimaging score.

3.4.

Effects of Arterial Blood Gas and Anticoagulation Parameters on Cerebral Autoregulation

Since the biggest changes in pH, ${\mathrm{PaO}}_2$, and ${\mathrm{PCO}}_2$ generally occurred from pre-ECMO to first 24 h of ECMO (Table 3), we analyzed the effects of these changes on cerebral autoregulation impairment. Specifically, the individual pH and ${\mathrm{PaO}}_2$ changes were calculated from the minimum values in the 24-h pre-ECMO to the maximum values in the first 24 hours of ECMO, which showed no significant correlation with autoregulation index. Similarly, the individual ${\mathrm{PCO}}_2$ change was calculated from the maximum value in the 24-h pre-ECMO to the minimum value in the first 24 h of ECMO, which showed no significant correlation with autoregulation index either.

Table 3

Arterial blood gas changes from 24-h pre-ECMO to the first 24 h of ECMO.

Finally, the anticoagulation parameters (PTT and UH) (Table 4) were not significantly associated with the patients’ neuroimaging scores.

Table 4

Anticoagulation parameters throughout ECMO run.