Neuro-anatomical localization of EEG identical bursts in patients with and without post-anoxic myoclonus

Abstract

Background:

The electroencephalograph (EEG) pattern of burst suppression with identical bursts (BSIB), with or without myoclonus, occurs often after resuscitation from cardiac arrest. These patterns are associated with severe brain injury but their neuropathological basis is unknown. Using EEG source localization, we tested whether post-cardiac arrest myoclonus was associated with specific anatomical distribution of BSIB.

Methods:

We performed a single center, case-control study of EEG-monitored post-cardiac arrest patients with BSIB. We determined the presence of myoclonus from clinical notes and video recordings. We generated normalized source density maps (sLORETA) for the first 0.5 seconds of each burst projected onto a standard anatomic model, and compared proportion of EEG power in the precentral gyrus (motor cortex) to the rest of the brain.

Results:

We included 20 patients, 10 with and 10 without myoclonus. Patients with myoclonus had greater electrical activation localized to the precentral gyrus compared to those without (median 3.25 [IQR 2.74–3.59] vs 2.68 [IQR 2.66–2.71], P = 0.04). There was no difference between groups in region of burst origin.

Conclusion:

Among patients with BSIB after cardiac arrest, those with clinical myoclonus have more electrocortical activation in the precentral gyrus.

Introduction

Cardiac arrest is common and often results in brain injury. One manifestation of severe anoxic brain injury is myoclonus, which occurs in 16–20% of patients resuscitated from cardiac arrest(1–3). While many patients with post-anoxic myoclonus have irrecoverable brain injury, myoclonus is a non-specific marker of poor outcome (4, 5). A related but distinct electroencephalographic (EEG) manifestation of severe anoxic brain injury is burst-suppression with identical bursts (BSIB). Patients with BSIB have severe and widespread damage to the cerebral cortex, cerebellum, hippocampus, and thalamus (6). Some patients with BSIB exhibit myoclonus in lock-step with their EEG discharges, while some have qualitatively similar EEG abnormalities without associated myoclonus (7, 8).

The neuroanatomic site of initiation and propagation of BSIB is unknown. We hypothesized BSIB cooccurs with myoclonus when bursts have predominant activation of the primary motor cortex. Therefore, we tested whether post-cardiac arrest patients with BSIB and myoclonus compared to patients with BSIB and no myoclonus demonstrated more prominent electrocortical activation in the precentral gyrus.

Methods

Patients and setting

The University of Pittsburgh Human Research Protection Office approved all aspects of this study. We examined a convenience sample of 20 post-cardiac arrest patients treated at a single academic medical center between 2014 – 2018 whose EEGs showed BSIB. We screened EEGs for BSIB with qualitative interpretation by two study authors, then confirmed BSIB by calculating between-burst correlation coefficients as previously described (8). We selected a sample of 10 consecutive patients with myoclonus and 10 patients without myoclonus from our prospective registry. We excluded subjects with less than 5 minutes of artifact-free EEG within 60 minutes of initiation of monitoring. We determined presence of myoclonus by reviewing clinical notes and video recordings, and verified patients underwent structured neurological examination including assessment for myoclonus without concomitant use of sedation or neuromuscular blockade. We defined myoclonus as stereotyped jerking movements time locked with burst activity on EEG. For included patients, we extracted clinical covariates from our registry including age, gender, etiology and location of cardiac arrest (in-hospital vs out-of-hospital), first monitored arrest rhythm (ventricular tachycardia or fibrillation vs pulseless electrical activity or asystole), targeted temperature management strategy, anticonvulsant drug use, and patient outcomes including proximate cause of death.

During the study period, many providers considered BSIB to represent potentially treatable epileptiform activity and administered anticonvulsants. We used propofol infusions for sedation, control of shivering, or to suppress motor myoclonus that interferes with care. We added continuous infusions of fentanyl if needed. We managed temperature in comatose patient to target 33°C or 36°C for 24 hours regardless of initial rhythm or arrest location. After 24 hours, we rewarmed patients at 0.25°C/h and maintain active normothermia until 72 hours post-arrest.

In-house EEG technologists routinely initiated continuous monitoring for all comatose post-arrest patients on arrival to the intensive care unit using gold-plated cup electrodes applied to the scalp in the standard 10–20 International System of electrode placement. We record data using XLTech Natus Neuroworks digital video/EEG systems (Natus Medical, Pleasanton, CA).

EEG modeling and analysis

We identified bursts on EEG by detecting voltages >6 standard deviations from average background electrical activation, and manually verified the accuracy of this event detection algorithm. We calculated the inter-burst interval, defined as the time between initiation of consecutive bursts. We aligned individual bursts within patients by optimizing cross-correlations, then calculated correlation coefficients between all pairwise combinations of in-phase bursts. We determined time from arrest to BSIB pattern from clinical notes and inspection of EEG. Next, we performed forward modeling using a three-layer symmetric boundary element model (BEM) generated with OpenMEEG and estimated source intensities using a standardized low resolution brain electromagnetic tomography (sLORETA) approach (9). sLORETA is commonly used mathematical solution to the so-called “inverse problem” of localizing a large number of electrical dipoles (neurons) using a small number of sensors (EEG electrodes). To allow source localization, sLORETA makes simplifying assumptions constraining dipole orientations and candidate locations to create a solvable set of equations. We projected normalized source density maps onto a standard anatomic head model, the Montreal Neurological Institute head model, which is a non-linear average of 152 subjects processed with FreeSurfer 5.3 ICBM152 (10). Because we used a composite head model instead of individual MRIs for patients, we allowed dipole orientations to be unconstrained to minimize biasing source estimation. We inspected the sLORETA for the first 0.5 seconds of each burst to identify its neuroanatomical location of origin using a Destrieux atlas (11). We calculated regional electrical amplitude over time by integrating the amplitude of electrical activation vs time for each brain region then normalized the precentral electrical activation to that of the rest of the brain over the duration of the EEG clipping. We performed all waveform analyses with MATLAB using the Brainstorm package with additional custom coding. We compared patients with and without myoclonus using Rank Sum tests.

Results

Clinical features of the 10 subjects with and 10 subjects without myoclonus were similar (Table 1). Providers administered levetiracetam to 40% of subjects with myoclonus but never to subjects with no motor correlate. Arrest etiologies were respiratory (n = 7), cardiac (n = 4), and other/unknown causes (n = 9). No patient survived to hospital discharge.

Table 1:

Baseline clinical characteristics and outcomes of included patients

Characteristic	Myoclonus n = 10	No myoclonus n = 10	Overall cohort n = 20
Age, years	61 [49 – 74]	58 [38 – 68]	60 [46 – 71]
Female	3 (30)	5 (50)	8 (40)
Out-of-hospital arrest	10 (100)	8 (80)	18 (90)
Arrest etiology
Respiratory	3 (30)	4 (40)	7 (35)
Cardiac	3 (30)	1 (10)	4 (20)
Other/unknown	4 (40)	5 (50)	9 (45)
Shockable rhythm	1 (10)	1 (10)	2 (10)
Survived	0 (0)	0 (0)	0 (0)
Median length of stay, days	3.5 [2 – 4]	4.5 [3 – 11]	4 [3 – 5.5]
Proximate cause of death
Withdrawal for perceived poor neurological prognosis	6 (60)	7 (70)	13 (65)
Withdrawal for non-neurological reasons	1 (10)	0 (0)	1 (5)
Rearrested, intractable shock, multisystem organ failure	3 (30)	1 (10)	4 (20)
Brain Death	0 (0)	2 (20)	2 (20)
Temp at time of EEG (median iqr)	35.5 [35.3 – 36.3]	35.1 [34.1 – 36.2]	35.4 [34.8 – 36.3
Anticonvulsant drug use, n (%)
Propofol	5 (50)	7 (70)	12 (60)
Midazolam	1 (10)	1 (10)	2 (10)
Levetiracetam	4 (40)	0 (0)	4 (20)
Total number of agents	1 [0 – 2]	1 [0 – 2]	1 [0 – 2]

Data are presented as medians with interquartile range or raw number with corresponding percentage

Inter-burst intervals were shorter in patients with myoclonus versus those without myoclonus (median 50 seconds [IQR 27 – 61] versus 71 seconds [IQR 25 – 185]). Between-burst correlation coefficients were similar for the two groups: median 0.77 [IQR 0.65 – 0.85] and 0.76 [IQR 0.61 – 0.86], respectively (Table 2, Figure 1). BISB was present on EEG initiation in all patients, and time from collapse to detection of BSIB was similar in patients with myoclonus versus those without myoclonus (median 702 [IQR 393 – 816] minutes versus 661 [IQR 548 – 821] minutes).

Table 2:

Electroencephalographic characteristics

Parameter	Myoclonus	No myoclonus
Inter-burst intervals, sec	50 [27 – 61]	71 [25 – 185]
Between-burst correlations	0.77 [0.65 – 0.85]	0.76 [0.61 – 0.86]
Time from arrest to BSIB, min	702 [393 – 816]	661 [548 – 821]
Normalized precentral activation	3.25 [2.74 – 3.59]	2.68 [2.66 – 2.71]
Region of burst origin (lobe)
Frontal	6 (60)	5 (50)
Occipital, parietal, temporal	4 (40)	5 (50)

Data are presented as medians with interquartile range or raw number with corresponding percentage

Figure 1:

Representative electroencephalographic (EEG) findings of two patients with burst suppression with identical bursts with (a) myoclonus or (b) no myoclonus observed. Inter-burst correlation coefficients are shown in (c) and (d), respectively.

EEG source localization suggested that bursts began bilaterally synchronously (Table 2). Most common motor manifestations were axial muscle jerks with or without associated eye opening and upward gaze deviation. There was no difference between groups in region of burst origin: amongst the 10 patients with myoclonus 6 patients had bursts initiated in the frontal cortex and 4 had bursts initiated in other parts of the brain, while in the no myoclonus group 5 patients had bursts initiated in the frontal cortex and 5 had bursts initiated in other parts of the brain. Calculated sources for bursts propagated across the brain over 100–200 ms. In patients with myoclonus, the proportion of calculated source power for each burst that localized to the precentral gyrus was 3.25 [IQR 2.74–3.59] compared to 2.68 [IQR 2.66–2.71] among patients without myoclonus (P = 0.04) (Figure 2).

Figure 2:

(A) Source localization from the patient with clinically apparent myoclonus shown from a superior view, with brain activation in subject with myoclonus over the first 0.105 seconds of a burst, progressing sequentially in panels from left to right. The burst begins bilaterally in the frontal lobe, progressing posteriorly and increasing in intensity over the motor cortex. (B) By contrast, source localization for patients without myoclonus (superior view (top two rows) and inferior view (bottom two rows)), showing brain activation focused primarily in the occipital lobe with comparatively little activation in the motor cortex

Discussion

We find that among patients who develop burst-suppression with identical bursts after resuscitation from cardiac arrest, those with myoclonus have more prominent activation in the motor cortex. Other clinical and electroencephalographic characteristics were comparable. This supports our hypothesis that myoclonus in this population may not have particular prognostic significance in and of itself. Historically, post-anoxic myoclonus was viewed as a sign of severe global anoxic injury and therefore a strong predictor of poor outcome (12). More recently, different electrographic phenotypes of myoclonus have been described, at least one of which is compatible with favorable recovery (1–3, 13, 14). There is growing recognition that EEG patterns most commonly seen in patients who do not recover meet criteria for BSIB (3, 8).

Recent work has better characterized BSIB, but much remains unknown. In seminal work, Hofmeijer et al. identified a cohort of twenty patients with BSIB, 100% of whom had poor outcomes while only 36% of patients with variable burst suppression had poor outcomes (8). Brain pathology of non-survivors with BSIB show structural damage in the cortex, cerebellum, hippocampus, and thalamus suggesting widespread severe injury (6). Nevertheless, the underlying substrate that results in BSIB is uncertain, and it is unknown why motor activation is observed in some patients but not others. We did not find a difference between groups in burst origin; thus, origin does not seem to explain the differences in motor activation and clinical manifestations between groups. Larger studies utilizing source localization and network analysis may provide further insight.

If myoclonus is an epiphenomenon that develops in a subset of patients with BSIB and non-specific for poor outcome, it is not a useful prognostic tool. Anticonvulsant drugs do not increase the probability that BSIB will evolve to a more favorable EEG pattern (15), suggesting it is unlikely that treatment with these classes of medications will improve outcomes. Thus, until results from randomized trials become available (16), there seems little evidence to support treating with the intent of suppressing myoclonus or BSIB in this population.

We were able to identify BSIB on average within twelve hours of arrest. Awakening with this EEG pattern has not been described in the literature with current conventional therapies. Few animal models of post anoxic myoclonus have been described, in which mechanistic hypotheses have been explored (17–19). Few novel therapies specific to this phenotype of post anoxic brain injury, which account for a high percentage of total neurological demise in adult cardiac arrest patients (20), have been trialed in animal models (21–23). Techniques such as point of care EEG may identify these patients even earlier on in their course for trials of novel cytoprotective therapies specific to this phenotype of post anoxic injury (24). Conversely, as conventional therapies have failed to yield awake survivors, trials without novel mechanisms specific to BSIB should obtain EEG and control for these patients separately in analyses.

Our study had important limitations. The single center design limits generalizability. We chose a small sample of consecutive patients with early, artifact free EEG data and clear documentation of myoclonus, and this sampling strategy is imperfect. Although sLORETA is commonly used and well-validated for EEG source localization, we do not have criterion standard data (e.g. invasive neuromonitoring) against which to validate our findings. Our work is an observational clinical study rather than laboratory-based neuroscience. As such, we were unable to account for the myriad factors present in the ICU setting that might affect EEG including hemodynamics, temperature, ventilation and pharmacotherapy. Further, we used a standardized anatomic map to define precentral activation, rather than patient-specific volumetric imaging. This may have introduced random variation into our results and biased our findings towards the null. Our findings should be viewed as hypothesis generating, but not guide clinical care.

In conclusion, we found that among patients with BSIB after cardiac arrest, those with myoclonus have greater activation in the precentral (motor) cortex. Further exploration of both BSIB and post-anoxic myoclonus will further elucidate the clinical and prognostic significance of these phenomena.

Figure 3:

Patients with myoclonus had a higher localization of electrical activity to the precentral gyrus (motor cortex) than patients without myoclonus (p = 0.04).