- Email: [email protected]
Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults
Clinical Neurophysiology xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph
Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults S. Gollwitzer a,⇑, T. Groemer a,b, S. Rampp a, M. Hagge a, D. Olmes a, H.B. Huttner a, S. Schwab a, D. Madzˇar a, R. Hopfengaertner a, H.M. Hamer a a b
Department of Neurology, University Hospital Erlangen, Schwabachanlage 6, 91054 Erlangen, Germany Department of Psychiatry, University Hospital Erlangen, Schwabachanlage 6, 91054 Erlangen, Germany
a r t i c l e
i n f o
Article history: Accepted 31 October 2014 Available online xxxx Keywords: Quantitative EEG Subarachnoid hemorrhage Delayed cerebral ischemia Vasospasm
h i g h l i g h t s Our study introduces a new algorithm for the prediction of imminent ischemia after subarachnoid
hemorrhage based on quantitative EEG monitoring. We developed an automated technique that does not require the presence of an EEG expert to prese-
lect or analyze raw EEG data and is therefore suitable for everyday usage by ICU staff. Our method indicated imminent ischemia earlier than established diagnostics such as transcranial
Doppler sonography or imaging and could therefore trigger intervention early enough to prevent infarction.
a b s t r a c t Objectives: Delayed cerebral infarction (DCI) has a significant impact on mortality and morbidity of patients with subarachnoid hemorrhage (SAH). The aim of this study was to define quantitative EEG (qEEG) parameters for the early and reliable prediction of DCI and compare the validity and time course of qEEG to standard procedures. Methods: 12 consecutive unselected SAH patients (8 female, mean age 52 years, Hunt-and-Hess grade I– IV) were prospectively examined. Continuous six channel EEG monitoring was started within 48 h after admission (mean duration 5.2 days; range: 2–12 days). All raw and unselected EEG signal underwent automated artifact rejection, Short Time Fast Fourier Transformation and a detrending procedure in order to analyze regional spectral power changes in different frequency bands. According to clinical standards, transcranial Doppler sonography (TCD) was performed at least on alternate days and repeat cerebral computer tomography (CCT) as needed. Results: 6 patients (50%) developed vasospasm/DCI. Decrease of P40% in power persisting over P5 h in the alpha band and P6 h in the theta band marked the optimal cut-off to detect DCI (sensitivity 89%, specificity 77% for alpha). EEG changes preceded detection of vasospasm/DCI in standard procedures by 2.3 days. Changes in the beta and delta band as well as in the alpha/delta ratio demonstrated lower correlation with imminent DCI. Conclusions: Focal reduction in alpha power may represent a valid, observer independent, non-invasive and continuous marker for vasospasm/DCI in SAH patients. Significance: qEEG indicates imminent ischemia earlier than established diagnostic tools, such as TCD. Ó 2014 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
1. Introduction ⇑ Corresponding author at: University Hospital Erlangen, Department of Neurology/Epilepsy Center, Schwabachanlage 6, 91054 Erlangen, Germany. Tel.: +49 9131 8539116; fax: +49 9131 8536469. E-mail address: [email protected] (S. Gollwitzer).
Delayed cerebral infarction (DCI) has a decisive impact on mortality and functional outcome after aneurysmal subarachnoid hemorrhage (SAH) (Vergouwen et al., 2011). Vasospasm measurable by angiography or transcranial Doppler (TCD) sonography has been
http://dx.doi.org/10.1016/j.clinph.2014.10.215 1388-2457/Ó 2014 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
2
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
strongly associated with DCI (Crowley et al., 2011; Brown et al., 2013). However, there is evidence that DCI may also be a consequence of other mechanisms such as cortical spreading depression or microcirculatory dysfunction undetectable by TCD (Dreier, 2011; Brown et al., 2013). Approximately 20% of DCI after SAH are not clinically detected or identified too late mostly due to impaired clinical assessment in poor-grade, comatose SAH patients (Schmidt et al., 2008). Therefore, the early detection of imminent ischemia being prerequisite to prevent infarction remains challenging. The EEG sensitively mirrors metabolic deterioration and disturbed neuronal activity following reduced cerebral blood flow (Astrup et al., 1981; O’Gorman et al., 2013). EEG changes such as flattening and loss of higher frequencies are found to occur rapidly if cerebral blood flow decreases below 0.16–0.17 ml g1 min1 (Sundt et al., 1974; Diedler et al., 2009). Continuous EEG monitoring has been shown to be a promising approach for early detection of imminent DCI in SAH patients (Labar et al., 1991; Rivierez et al., 1991; Vespa et al., 1997; Claassen et al., 2004, 2013; Stuart et al., 2010; Rathakrishnan et al., 2011). Power spectral analyses of raw EEG data represent a fast and robust screening method for longterm EEG recordings (Hopfengartner et al., 2007). The aim of this study was to develop a semiautomatic method to predict and localize reliably imminent deterioration due to DCI at the earliest time point. 2. Methods 2.1. Patients Participation in this prospective study was offered to all adult patients who were admitted to the Neurological Intensive Care Unit (NICU) of the University Hospital Erlangen between November 2011 and February 2013 and suffered from nontraumatic, out-of-hospital SAH regardless of clinical severity. No patient in our cohort had a history of stroke or other cerebral lesions possibly causing preexisting focal EEG slowing. Moreover, the initial CCT of all patients did not reveal signs of previous brain insults. The following inclusion criteria had to be fulfilled: age P 18 years and symptom onset <24 h before hospital admission. In all cases, the diagnosis was confirmed by CCT, which was performed immediately after admission and within the first 24 h of symptom onset. Patients were excluded from the study in cases of angiographic or sonographic proof of vasospasm before initiation of EEG monitoring. The study was approved by the local ethics committee. All patients or their relatives provided written informed consent before entering the study. 2.2. Clinical management and multimodal monitoring All patients underwent diagnostic angiography on the day of admission. SAH was caused by aneurysmal bleeding in all patients. All aneurysms were treated by endovascular coiling. There were no procedure-related infarctions. All patients were treated with nimodipine (Kronvall et al., 2009). Our current standard operating procedure includes monitoring of intracranial pressure (ICP); systolic, diastolic, and mean arterial blood pressure; heart rate; temperature; and oxygen saturation. Neurological status was surveyed at least three times per day by NICU staff. Wake-up tests for neurological assessment are not performed routinely during the critical phase of the disease. All patients received TCD or transcranial color-coded Doppler sonography (TCCS) at least every other day and when clinically indicated. No TCD monitoring was administered. CCT or MRI scans were performed in the case of clinical deterioration of the patient or worsening of TCD measurements. The relevant vasospasms were diagnosed by TCD (defined as peak systolic Doppler frequencies
>4 kHz) or TCCS (defined by cerebral flow velocities >200 cm/s, or an increase of >50 cm/s in 24 h according to the guidelines of the German Neurological Society) or by angiography. DCI was defined as new ischemic lesions on imaging that were not present on initial CCT performed within 24 h of symptom onset. In awake patients, delayed ischemic neurological deficits (DINDs) were defined as new focal neurological signs not detectable at the time of symptom onset or immediately after aneurysm occlusion and not caused by other complications such as rebleeding or hydrocephalus (Drenckhahn et al., 2012). In the case of relevant vasospasm or DCI/DIND, triple-H therapy (hypervolemia, hypertension, and hemodilution) (Treggiari, 2011) was initiated. If neurological deficits persisted, patients underwent interventional angiography and they were treated by chemical vasospasmolysis or balloon angioplasty (Sayama et al., 2006; Aburto-Murrieta et al., 2012). In the case of cerebral edema, osmotherapy with mannitol was administered and mild-to-moderate hypothermia was induced using an invasive cooling device (CoolgardÒ) if ICP elevation persisted. 2.3. EEG monitoring EEG monitoring was started within the first 48 h after admission and was maintained for a maximum of 12 , but it was discontinued earlier in the case of discharge from the ICU or if requested by patients or relatives. We restricted the EEG to a number of 10 electrodes attached according to the International 10–20 system: C3/ C4, P3/P4, O2/O1, T3/T4, and F3/F4 (alternatively F5/6 instead of F3/F4, if necessary due to head bandages). Simplified EEG montages are widely and successfully used for EEG monitoring under ICU conditions, although general recommendations regarding the best montage and number of electrodes are still lacking (Labar et al., 1991; Vespa et al., 1997; Kull and Emerson, 2005). We aimed for a compromise between stable maintenance of the recording and a full coverage of all vascular territories. EEG data were digitized at a sampling rate of 256 Hz with a high-pass filter of 0.08 Hz and a low-pass filter of 86 Hz. We applied a bipolar montage (F4–C4, T4–P4, P4–O2; F3–C3, T3–P3, P3–O1), hypothesizing that F4–C4 (F3–C3) and T4–P4 (T3–P3) represent the territory of the anterior cerebral artery (ACA) and middle cerebral artery (MCA) while P4– O2 (P3–O1) captures changes in the vertebrobasilar circulation. 2.4. qEEG analysis In order to perform reliable semiautomatic qEEG analysis, automated artifact rejection is essential, and we used a standardized and efficient technique that has been described and evaluated in detail in recent work (Hopfengartner et al., 2007). The remaining raw EEG signal underwent short-time Fourier transformation of 2-s epochs. The frequency distribution in each epoch was identified, which allowed for the determination of power values in the following frequency bands: delta 0.5–3.5 Hz, theta 4–7.5 Hz, alpha 8–12 Hz, and beta 12.5–22 Hz. All EEG channels were subject to common fluctuations due to global effects such as changes of ICP, level of sedation, cooling and medication, or sleep–wake patterns. In order to eliminate such global or bilateral effects and unmask focal alterations, which we postulated to be the correlate of imminent DCI, several procedures were necessary. Because all channels correlated highly in their power (R > 0.92 and p < 0.001 for each combination of channels) over time while having different power levels, a detrending procedure was established that eliminated common fluctuations and uncovered significant regional drops in the EEG power. First, every median 1-min power measurement was divided by the median of all 1-min values of the same channel over the first
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
3
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
24 h of EEG recording for normalization. Second, a detrending procedure was performed. This was carried out by calculating the median of all normalized power values of the six channels at a given point in time generating the normalized common trend of all channels. Each single-channel power value was then divided by the common trend to eliminate general fluctuations and stress focal changes. In order to detect focal EEG changes over time from day 2 of monitoring, the median of every 60 normalized and detrended 1min EEG power measurements per hour of EEG recording in each channel was compared to the median normalized and detrended power value of the same channel over the baseline period. The baseline was defined as a minimum of 6 h and a maximum of 12 h of uninterrupted EEG recording within the first 24 h of EEG monitoring assuming that vasospasm or DCI were absent in this time period. Raw EEG of each monitoring day was also visually inspected by two trained electroencephalographers (SG and HMH) in order to confirm qEEG findings and exclude possible mimics of EEG power changes caused by artifacts, periodic complexes, or epileptic activity. 2.5. Statistics We correlated qEEG changes in each individual patient with the clinical end points, ‘‘delayed cerebral ischemia’’ or ‘‘increased risk of delayed cerebral ischemia,’’ which were marked by at least one of the following conditions: (1) proof of vasospasm in TCD/ TCCS measurements, (2) new cerebral infarction in CCT/MRI, and (3) new focal neurological deficit not otherwise explained. In addition, the time course of qEEG alterations was compared to the time point when the defined clinical end points were first noticed by standard diagnostics in order to assess the capability of qEEG to predict DCI. To calculate the strength of association between qEEG changes and DCI, we applied receiver operator characteristics (ROC) considering each hemisphere separately. In order to explore the informative value of the magnitude and duration of EEG power decrease, we assessed the sensitivity and specificity of different power decline scenarios ranging from 10% to 100% and persisting for a minimum of 1 h and a maximum of 8 h. To define the optimal threshold value on the ROC curve, we calculated the Youden index (J), which is a function of sensitivity and specificity (Youden, 1950). 3. Results Of 22 consecutive SAH patients admitted during the study period, 12 patients fulfilled the inclusion criteria and were included (Table 1), while 10 were admitted to hospital very late during
the course of the disease for baseline acquisition. Of the included patients, five were sedated and mechanically ventilated during the complete EEG monitoring, while seven were awake and breathing spontaneously. Sedation was achieved by continuous intravenous administration of opioid analgesics and either midazolam or propofol. In two patients, intravenous catecholamines were administered intermittently. No other medications that may have an influence on the EEG, such as neuroleptics or antidepressants, were administered to any of the patients. In four of the sedated patients, visual inspection of raw EEG revealed reduction of alpha activity, and the fifth sedated patient showed intermittent generalized slowing, probably due to the medication, which consisted of midazolam and sufentanil in this case (Herkes et al., 1992; Saletu et al., 2006). No triphasic waves, periodic complexes, or interictal or ictal activity was noted in the raw EEG including patients 3 and 8 who suffered from a single seizure prior to admission (Table 2). In all cases, visual inspection of raw EEG confirmed qEEG findings. Only one patient showed mildly increased transaminases; no patient developed renal, liver, or multiorgan failure. Detailed information on complications and clinical outcome is given in Table 2. Fifty percent (n = 6) of the patients developed vasospasm (Table 3). The territories of the MCA and ACA were most often affected. Isolated vasospasm of the posterior cerebral artery (PCA) and basilar artery (BA) did not occur. In two patients, vasospasm led to infarctions seen on CT/MRI. One of the two patients died due to complete infarction of the left MCA and ACA territory. DIND in the form of an incomplete paresis of the right arm was noted in one awake patient (patient 11) on day 10, 1 day after the first occurrence of left-hemispheric vasospasm. Decreases in the total alpha and theta power showed the strongest association with vasospasm/DCI (Youden index J = 0.66 Table 4, Fig. 1). In both frequency bands, a power decline of 40% marked the statistically optimal threshold. The duration of the power decrease providing the highest significance was 5 h for alpha and 6 h for theta. The optimal combined cut-off for power decrease and duration proved to be highly sensitive (89%) with a specificity of 77% for both frequency bands (Table 4). Changes in the beta frequency band demonstrated lower correlation with vasospasm/DCI; this was likely due to contamination by muscle artifacts as visual inspection of raw EEG revealed. The weakest correlation existed between DCI and fluctuation in delta power. Consequentially, the alpha–delta ratio showed only a moderate correlation with vasospasm/DCI (sensitivity of 55% and specificity of 85%). A significant alpha decrease was seen in six patients (Table 3). Three of these patients were awake, responding, and following commands. No sedative agents were given in these cases. The
Table 1 Patients’ characteristics. Patient
Age (years)/sex
Clinical grade (Hunt and Hess)
Fisher grade
Location of aneurysm
EEG recording (days)
EVD
Intubated
1 2 3 4 5 6 7 8 9 10 11 12
70/f 51/f 34/f 73/f 38/f 44/f 48/m 53/m 51/m 73/f 45/f 49/f
2 1 3 2 1 4 1 4 3 2 3 2
3 1 3 4 4 3 4 2 4 4 4 3
L PCOM BA ACOM ACOM R PCA R PCOM ACOM ACOM ACOM ACOM BA R ACI
2 6 12 4 3 4 5 6 7 8 7 10
No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
No No Yes No No Yes No Yes Yes No No Yes
Mean (SD)
52.4 (13.0)
5.2 (3.2)
L: left; R: right; f: female; m: male; ACA: anterior cerebral artery; ACOM: anterior communicating artery; BA: basilar artery; MCA: middle cerebral artery; PCA: posterior cerebral artery; PCOM: posterior communicating artery; SD: standard deviation; and EVD: external ventricular drain.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
4
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Table 2 Complications and clinical outcome of patients. Patient
Complication
Time of complication (day SAH)
Clinical outcome (mRS)
1
None
2
CSF circulatory dysfunction Infection, fever
1–4 2
0
3
Seizure NSTEMI Elevated ICP CSF circulatory dysfunction
Prior to admission 1 1–8 1–12
2
4
CSF circulatory dysfunction Infection (ventriculitis), fever
2–11 4
0
5
CSF circulatory dysfunction
2–10
0
6
CSF circulatory dysfunction Infection, fever Elevated ICP Elevated transaminases
1–16 2 2–7 10
5
7
Infection, fever CSF circulatory dysfunction
1 2–7
0
8
Seizure CSF circulatory dysfunction Infection (ventriculitis), fever
Prior to admission 1–4 4
2
9
CSF circulatory dysfunction Complete infarction of left MCA territory
1–10 9
6
10
Elevated ICP CSF circulatory dysfunction
1–7 1–15
5
11
Infection, fever
3
1
12
Delirium anemia
4 4–13
1
1
mRS: modified Rankin scale (Banks et al., 2007) assessed on day of discharge; CSF: cerebrospinal fluid; NSTEMI: non-ST-elevated myocardial infarction.
Table 3 Identification and localization of vasospasm/DCI by different modalities (qEEG, TCD, and CCT/MRI). Patient
EEG (reduced alpha power)a Channel
No DCI 1 4 5 7 8 100 DCI 20
3 6 9
110 12
Onsetb (day SAH)
TCD/TCCS (increased flow velocity)
CT/MRI (infarction)
Durationc (days)
Territory
Territory
– – – – – P3O1
3
4.5
– – – – – –
F4C4 T4P4
6 6
F4C4 F3C3 F4C4 T4P4 F3C3 T3P3 F3C3 T3P3 –
4 5 3 3 3 3 4 4 –
700 700 – 5–1200 6–1200 4,500 4,5,6 4,5,700 4–700 5–700 5–700 –
R ACA R MCA L MCA R ACA L ACA R MCA R MCA – – L ACA L MCA Ubiquitous
Days SAH
Days SAH
– – – – – – 2,7 2,5–9 5 9 9 6,7,10,13 6,7,8 – – 12,14 9,10–12,14 4,6–8,11
R ACA L ACA
12 12
L ACAd L MCAd
9 9
–
–
a
EEG: event defined as at least 40% decrease of alpha power persisting P5 h. b Onset: initial occurrence of EEG event. c Duration: following days during course of disease fulfilling criteria for EEG event. ‘‘Power decrease persisted until end of monitoring’’. d Patient died because of brain infarction. ‘Awake patients with decrease of alpha power’. Patient 2 and 10 did not reveal any neurological deficits, whereas patient 11 suffered from incomplete paresis of right arm from day 10 (DIND). L: left; R: right; ACA: anterior cerebral artery; MCA: middle cerebral artery; PCA: posterior cerebral artery.
remaining three patients were sedated, and they did not react to stimuli throughout the monitoring. However, brain-stem reflexes remained present. In four cases, EEG changes preceded vasospasm detected by TCD or DCI seen in CCT/MRI. On average, EEG power decreases
occurred 2.3 days (SD 3.3, range 2–5) earlier than proof of vasospasm in TCD/TCCS or DCI in CCT/MRI. Diminution of alpha power was persistent over several days in four cases. Recovery of alpha power was only seen over the right hemisphere of patient 9, concordant with the fact that no right-
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
5
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx Table 4 Association between decrease in qEEG power and vasospasm/DCI. Frequency band
Alpha Beta Delta Theta A/D ratio
Optimal cut-off Power decrease (%)
Duration (h)
40 50 70 40 70
5 3 2 6 2
Youden index (J)
Area under the curve (AUC)
Sensitivity (%)
Specificity (%)
0.66 0.47 0.24 0.66 0.40
0.66 0.66 0.52 0.72 0.64
89 78 55 89 55
77 69 69 77 85
Optimal cut-off values of power decrease and duration according to Youden index; AUC derived from receiver operator characteristics (ROC); and specificity and sensitivity weighted equally.
hemispheric infarction developed as shown by imaging. Details on the duration of power decreases are given in Table 3. In patient 2, blood flow velocity acceleration was seen in TCD first on day 2 in the right ACA and MCA, and exclusively in the left MCA on day 5. The EEG did not reveal any focal deterioration over the left hemisphere, and it captured alpha power decline over the right hemisphere not earlier than day 6. Inspection of raw EEG confirmed these findings. The patient was awake and breathing spontaneously during the entire monitoring period and did not develop neurological deficits or ischemic lesions, although no therapeutic countermeasures against vasospasm were taken. We, therefore, considered the EEG to be in line with imaging and the clinical status of the patient, but we had to regard the left-hemispheric EEG findings as false negative according to the study protocol. In patient 10, alpha power dropped significantly in the left posterior region for 3 days without proof of vasospasm/DCI in TCD or CCT. Visual inspection of the original EEG at the time revealed focal slowing and amplitude flattening without artifacts blurring the EEG. Patient 12 showed ubiquitous increased flow velocities identified by TCD/TCCS on day 4 (Table 2) which were not reflected by qEEG post processing. The patient, however, did not develop DCI and recovered without neurological deficits. Moreover, the patient suffered from anemia with hemoglobin levels dropping below 100 g/l from day 4 after symptom onset and only recovered from day 13 paralleling TCCS changes. Therefore, we regarded the anemia as the cause of the TCCS flow acceleration while the assumption of ubiquitous critical vasospasm not leading to any infarction or neurological deficit appeared less plausible. Patient 9 revealed marked bilateral alpha power reduction starting on day 3 (Fig. 2). TCD detected vasospasm in the right MCA on day 7. No vasospasm was seen on the left side in TCD throughout the monitoring. While alpha power in the right hemisphere recovered during the following days, it continued to decline in all channels on the left. CCT and MRI, triggered by clinical deterioration of the patient on day 9, showed complete infarction of the left ACA and MCA territory leading to the patient’s death. As alpha power on the left side dropped below the threshold long before clinical deterioration was noted and infarction was seen in CCT, we believe that the power decline marked the initial compromised perfusion in the MCA territory rather than the infarct. However, no CCT or MRI was performed at the time of EEG deterioration. Alternative explanations for the bilateral EEG changes, for example, changes in the sedation regime, could not be identified.
4. Discussion This study aims to introduce an automated, easy-to-use, reliable, and robust algorithm for the prediction of imminent ischemia in SAH patients early enough to allow for countermeasures possibly preventing infarction. In order to warrant clinical practicability, we applied a novel algorithm based on continuous EEG monitoring
without visual preselection of specific EEG sections. In this setting, >40% decrease in alpha or theta power for longer than 5 or 6 h, respectively, showed a strong association with vasospasm or DCI and was able to predict this complication 2.3 days on average before it was detected by other methods, such as TCD, in the majority of patients. EEG changes preceded identification of DCI in other diagnostics in four out of five patients, while only one patient exhibited early increased accelerated flow velocities in TCD not accompanied by EEG changes at that time. However, this patient neither deteriorated clinically nor developed cerebral infarction in the later course of the disease. Apart from the prediction of DCI, EEG was also able to lateralize the pathology correctly in all cases. Exact localization of the vascular territory was not always achieved by EEG. As therapeutic consequences are rather dependent on the detection of the presence of DCI than its exact localization, we aimed for a reliable method to predict the event but not necessarily to localize it precisely. However, this approach introduces its own set of challenges. First, maintenance of high-quality EEG monitoring over several days under ICU conditions is difficult and labor intensive (Young and Campbell, 1999). Second, various artifacts unavoidably contaminate EEG recordings on ICU. Third, EEG is sensitive but unspecific to various systemic changes, making it difficult to distinguish DCI from other factors such as metabolic alterations, ICP changes, or effects of sedation. We tried to address these shortcomings by using a reduced but multichannel montage that could be maintained over days with a reasonable effort and was still able to detect reliably focal EEG changes. In addition, an efficient automatic artifact rejection technique was implemented, which was evaluated extensively in previous studies (Hopfengartner et al., 2007). This allowed for unbiased analysis of changes in the alpha, theta, and delta frequency bands. It was, however, impossible to differentiate automatically between beta activity and muscle artifacts, which led to low usability of beta fluctuations. The key to the discrimination of EEG changes triggered by DCI from changes due to other causes was the perception of vasospasm as a predominantly focal event (Miller and Palestrant, 2012). Most other incidents with impact on the EEG of ICU patients suffering from SAH cause global changes. By detrending, general fluctuations were excluded from further analyses, thus making focal changes visible in qEEG. Especially changes caused by different states of wakefulness and medication effects were eliminated due to their non-focal nature. For example, increases in higher frequencies that are often induced by benzodiazepines were extinguished as they affected many channels equally on both sides. A theoretical disadvantage of this approach could be its lack of detection of global changes, which cannot be excluded to be also caused by SAH. However, in five out of six patients in our cohort, DCI was clearly restricted to one or two adjacent arterial territories; therefore, it was well detected by the algorithm. One patient exhibited ubiquitous increased flow velocities in TCD/TCCS that was more likely due to anemia (Brass et al., 1988; Ameriso et al.,
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
6
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Fig. 1. Statistical analysis: Youden index (J) for different frequency bands. Percentage of power decrease (y-axis) plotted against duration of power decrease in hours (x-axis). Youden index expressed by gray shades, corresponding to bar on the right. High J-values illustrated by light gray shades implicated strong correlation. Best results were achieved in the alpha band (J = 0.66; optimal threshold 40% decrease, duration of 5 h) and the theta band (J = 0.66; optimal threshold 40% decrease, duration of 6 h). Changes in beta, delta, and alpha–delta ratio were less sensitive and specific for DCI (beta: J = 0.47; delta: J = 0.24; and alpha–delta ratio: J = 0.40). ROC curve for varying power decrease values and fixed duration (5 h); for AUC values, see Table 3.
1990; Isikay et al., 2005) than due to generalized vasospasm. The team of neurointensivists treating the patient shared this interpretation and the patient did not receive any specific treatment for vasospasms. In contrast to the TCD/TCCS findings, the patient recovered without neurological deficit and TCD normalized with
rising hemoglobin levels. We, therefore, assume that EEG was not prone to this specific error source and may be superior to TCD in this situation. A strength of EEG is its ability to directly detect disturbed neuronal activity regardless of the underlying cause rather than to
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
7
Fig. 2. Correlation of alpha power, TCD, and imaging: patient 9, SAH III, aneurysm of anterior communicating artery; (A) TCD showing critical vasospasm of the right MCA on days 7–10; no increase of frequency in left MCA; (B) EEG alpha power: marked decrease of alpha power in all channels on days 3 and 4 (black arrows); recovery of alpha power in channels T4P4 and F8C4 on day 6/7 (gray arrow); no recovery in T3P3/F7C3; (C) CCT and ; (D) perfusion weighted MRI on day 10 showing complete infarction of the left ACA and MCA territory.
focus on vasoconstriction in larger vessels as an indirect marker of neuronal damage. For many years, vasospasm has been regarded as the main cause for ischemia in SAH patients. There is, however, emerging evidence that DCI has a multifactorial pathogenesis, as patients can develop cerebral infarction in vascular territories unaffected by angiographic vasoconstriction (Naidech et al., 2006). An alternative pathophysiological mechanism potentially contributing to the development of DCI may be cortical spreading depression (CSD). Animal models have shown that SAH can cause a primary cellular dysfunction, which can induce spreading
depolarizations accompanied by alterations in ion homeostasis (Hubschmann and Kornhauser, 1982). Dreier et al. found that hemoglobin in the subarachnoid space can cause spreading ischemia leading to widespread cortical necrosis (Dreier et al., 2000, 2006; Drenckhahn et al., 2012). It has also been shown in microdialysis studies that changes in brain metabolism such as an increase in lactate–pyruvate ratio and decreased brain tissue oxygenation preceding imaging are proof of ischemia. EEG could be a powerful tool to capture these changes at an early stage (Helbok et al., 2011).
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
8
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Fig. 3. Alpha and delta power during different stages of ischemia: patient 3, SAH III, aneurysm of anterior communicating artery. (A) Critical vasospasm in left and right ACA on day 9 as detected by TCD; (B) alpha power decrease on days 4 and 5 in F8C4 and F7C3 (black arrow), no relevant EEG changes in T4P4 and T3P3; (C) delta power increase in F8C4 and F7C3 on day 7/8 (gray arrow); and (D) moderate decrease of alpha/delta ratio from day 6 (black arrow).
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
To be a reliable marker for DCI, the alpha and theta power decreases had to be persistent over a certain period of time. The optimum cut-off duration was slightly shorter for alpha than for theta. Accordingly, alpha appears to be the preferred parameter in awake as well as in comatose patients. In our cohort, physiological changes such as sleep and wake patterns (including alpha background activity) were symmetrical in appearance and were at least in parts subject to detrending, and they remained below the power threshold, which was associated with vasospasm or DCI. Thus, the method seems applicable for less-affected awake patients as well as for poor-grade SAH patients. There was one patient in our cohort without any proof of vasospasm or DCI in TCD and neuroimaging who demonstrated a significant isolated left occipital alpha decrease, which had to be interpreted as false positive according to the study protocol. As theta power decreased equally in this patient, the alternative explanation is that the sensitive qEEG correctly identified reduced cortical perfusion in this area as a sign of imminent DCI, but clinical treatment or the natural course of the disease prevented the perfusion deficit from progressing into persistent and more severe DCI, which would also be identifiable by other diagnostic tools. While previous studies focused on changes in delta power to detect DCI, it is the novelty of our study to show the significance of alpha power decreases in this context. Although alpha may not be prominent especially anteriorly in awake patients on visual analysis of EEG, Fourier analysis reveals that alpha is a part of the spectrum in all brain regions and states of wakefulness and it is therefore eligible for the purpose of the study. Unexpectedly, delta power changes and the alpha–delta ratio showed only weak-to-moderate correlations with vasospasm in our cohort. At first glance, this result appears to be in contrast to previous studies. Claassen et al. found a high probability for patients to develop DCI when the alpha–delta ratio decreased by >10% in six consecutive recordings or by >50% in a single recording to develop DCI (Claassen et al., 2004). Stuart et al. also reported the alpha–delta ratio to be most predictive for DCI in intracortical EEG (Stuart et al., 2010). A possible explanation for the divergent findings may be the hypothesis corroborated by Machado et al. (Machado et al., 2004) who applied qEEG to patients with MCA ischemic strokes and found delta power increases to be related to the ischemic core while the tissue at risk, penumbra, and edema were characterized by alpha and theta alterations. Correspondingly, the alpha decrease in our study marked the early point where perfusion dropped below a critical level, generating tissue at risk of, but not yet persistent, ischemic infarction. This hypothesis is supported by patient 3 who developed ischemic infarction during the course of the disease. Delta power increased 2 days after alpha power declined, probably indicating irreversible ischemia (Fig. 3). In this context, the alpha/delta ratio could be a valuable parameter to estimate the prognosis and predict the patient’s functional outcome (Burghaus et al., 2007; Sheorajpanday et al., 2011a,b) while the alpha power may be the superior parameter to prevent infarction by initiating countermeasures. Although the diminution of alpha power as a marker of DCI has not yet been confirmed by other studies, our results give reasons to regard alpha power as a sensitive and specific predictor of DCI in the context of pathophysiological considerations. Claassen et al. found a strong correlation between a decrease of the alpha–delta power ratio and DCI (a sensitivity of 100% and a specificity of 76%) by analyzing artifact-free clips visually preselected by experienced electroencephalographers (Claassen et al., 2004). Our approach introduces an automated technique that does not require the presence of an EEG expert to preselect or analyze raw EEG data, and it is therefore suitable for everyday use by ICU staff. Vespa et al. also chose a continuous approach and identified the relative alpha (RA) variability as the most sensitive (100%)
9
indicator of imminent ischemia (Vespa et al., 1997). Although highly sensitive, RA variability not only corresponded exclusively to vasospasm or DCI but also dropped significantly as a consequence of a variety of other conditions, for example, increase of ICP or herniation, causing low specificity (50%). Labar et al. used a two-channel EEG montage to compare total power changes and plotted consistent EEG trends day to day and left to right (Labar et al., 1991). To be identified as persistent, the EEG trend had to persist over 24 h in this study. The algorithm used in our study found a focal alpha power decrease persistent over just 5 h to be highly sensitive and specific for vasospasm/DCI, thus allowing for earlier intervention. There are several limitations to our study. First, the small number of patients limited the statistical power of our results. Despite the small number of analyzed patients, the results of this proof-ofprinciple study were significant and supported the hypothesis that qEEG is a reliable predictor of DCI. However, verification of the results in larger cohorts is needed in order to be able to rely on the clinical utility of this approach. Another restriction of the algorithm is the necessity of a vasospasm-free baseline EEG. Therefore, patients had to be excluded when they were admitted to our hospital at later stages of their disease when vasospasms may already have been present. We are also aware that the restricted number of EEG electrodes lowers the spatial resolution of our findings but increases the practicability of the method in ICU settings. A possible problem with our approach is the detection of DCI in cases with preexisting focal slowing due to prior ischemic strokes or cerebral lesions. As no patient in our cohort had a history of stroke or structural cerebral lesion nor showed signs of it in the initial CCT, it is unclear if the power decline in comparison to baseline would be large enough in such cases to exceed the threshold established in this study. In conclusion, our study introduces an automated, easy-to-use, reliable, and robust algorithm for the prediction of imminent ischemia in SAH patients early enough to allow for countermeasures possibly preventing infarction. In contrast with established diagnostics such as TCD or imaging, the real-time character of this method creates the opportunity for an EEG-based alarm system that has the potential to substantially improve the functional outcome of SAH patients. Sources of funding The study was not funded. Acknowledgment We would like to thank Arndt Dörfler, MD, Department of Neuroradiology, for the provision of MRI and CT data. Conflict of interest: None of the authors have potential conflict of interest to be disclosed. References Aburto-Murrieta Y, Marquez-Romero JM, Bonifacio-Delgadillo D, Lopez I, Hernandez-Curiel B. Endovascular treatment: balloon angioplasty versus nimodipine intra-arterial for medically refractory cerebral vasospasm following aneurysmal subarachnoid hemorrhage. Vasc Endovascular Surg 2012;46:460–5. Ameriso SF, Paganini-Hill A, Meiselman HJ, Fisher M. Correlates of middle cerebral artery blood velocity in the elderly. Stroke 1990;21:1579–83. Astrup J, Siesjo BK, Symon L. Thresholds in cerebral ischemia – the ischemic penumbra. Stroke 1981;12:723–5. Brass LM, Pavlakis SG, DeVivo D, Piomelli S, Mohr JP. Transcranial Doppler measurements of the middle cerebral artery. Effect of hematocrit. Stroke 1988;19:1466–9. Brown RJ, Kumar A, Dhar R, Sampson TR, Diringer MN. The relationship between delayed infarcts and angiographic vasospasm after aneurysmal subarachnoid hemorrhage. Neurosurgery 2013;72:702–8.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
10
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Burghaus L, Hilker R, Dohmen C, Bosche B, Winhuisen L, Galldiks N, et al. Early electroencephalography in acute ischemic stroke: prediction of a malignant course? Clin Neurol Neurosurg 2007;109:45–9. Claassen J, Hirsch LJ, Kreiter KT, Du EY, Connolly ES, Emerson RG, et al. Quantitative continuous EEG for detecting delayed cerebral ischemia in patients with poorgrade subarachnoid hemorrhage. Clin Neurophysiol 2004;115:2699–710. Claassen J, Taccone FS, Horn P, Holtkamp M, Stocchetti N, Oddo M. Recommendations on the use of EEG monitoring in critically ill patients: consensus statement from the neurointensive care section of the ESICM. Intensive Care Med 2013;39:1337–51. Crowley RW, Medel R, Dumont AS, Ilodigwe D, Kassell NF, Mayer SA, et al. Angiographic vasospasm is strongly correlated with cerebral infarction after subarachnoid hemorrhage. Stroke 2011;42:919–23. Diedler J, Sykora M, Bast T, Poli S, Veltkamp R, Mellado P, et al. Quantitative EEG correlates of low cerebral perfusion in severe stroke. Neurocrit Care 2009;11: 210–6. Dreier JP. The role of spreading depression, spreading depolarization and spreading ischemia in neurological disease. Nat Med 2011;17:439–47. Dreier JP, Ebert N, Priller J, Megow D, Lindauer U, Klee R, et al. Products of hemolysis in the subarachnoid space inducing spreading ischemia in the cortex and focal necrosis in rats: a model for delayed ischemic neurological deficits after subarachnoid hemorrhage? J Neurosurg 2000;93:658–66. Dreier JP, Woitzik J, Fabricius M, Bhatia R, Major S, Drenckhahn C, et al. Delayed ischaemic neurological deficits after subarachnoid haemorrhage are associated with clusters of spreading depolarizations. Brain 2006;129:3224–37. Drenckhahn C, Winkler MK, Major S, Scheel M, Kang EJ, Pinczolits A, et al. Correlates of spreading depolarization in human scalp electroencephalography. Brain 2012;135:853–68. Helbok R, Madineni RC, Schmidt MJ, Kurtz P, Fernandez L, Ko SB, et al. Intracerebral monitoring of silent infarcts after subarachnoid hemorrhage. Neurocrit Care 2011;14:162–7. Herkes GK, Wszolek ZK, Westmoreland BF, Klass DW. Effects of midazolam on electroencephalograms of seriously ill patients. Mayo Clin Proc 1992;67:334–8. Hopfengartner R, Kerling F, Bauer V, Stefan H. An efficient, robust and fast method for the offline detection of epileptic seizures in long-term scalp EEG recordings. Clin Neurophysiol 2007;118:2332–43. Hubschmann OR, Kornhauser D. Effect of subarachnoid hemorrhage on the extracellular microenvironment. J Neurosurg 1982;56:216–21. Isikay CT, Uzuner N, Gucuyener D, Ozdemir G. The effects of hematocrit and age on transcranial Doppler measurements in patients with recent ischemic stroke. Neurol India 2005;53:51–4. discussion 4. Kronvall E, Undren P, Romner B, Saveland H, Cronqvist M, Nilsson OG. Nimodipine in aneurysmal subarachnoid hemorrhage: a randomized study of intravenous or peroral administration. J Neurosurg 2009;110:58–63. Kull LL, Emerson RG. Continuous EEG monitoring in the intensive care unit: technical and staffing considerations. J Clin Neurophysiol 2005;22:107–18. Labar DR, Fisch BJ, Pedley TA, Fink ME, Solomon RA. Quantitative EEG monitoring for patients with subarachnoid hemorrhage. Electroencephalogr Clin Neurophysiol 1991;78:325–32. Machado C, Cuspineda E, Valdes P, Virues T, Llopis F, Bosch J, et al. Assessing acute middle cerebral artery ischemic stroke by quantitative electric tomography. Clin EEG Neurosci 2004;35:116–24.
Miller CM, Palestrant D. Distribution of delayed ischemic neurological deficits after aneurysmal subarachnoid hemorrhage and implications for regional neuromonitoring. Clin Neurol Neurosurg 2012;114:545–9. Naidech AM, Drescher J, Tamul P, Shaibani A, Batjer HH, Alberts MJ. Acute physiological derangement is associated with early radiographic cerebral infarction after subarachnoid haemorrhage. J Neurol Neurosurg Psychiatry 2006;77:1340–4. O’Gorman RL, Poil SS, Brandeis D, Klaver P, Bollmann S, Ghisleni C, et al. Coupling between resting cerebral perfusion and EEG. Brain Topogr 2013;26:442–57. Rathakrishnan R, Gotman J, Dubeau F, Angle M. Using continuous electroencephalography in the management of delayed cerebral ischemia following subarachnoid hemorrhage. Neurocrit Care 2011;14:152–61. Rivierez M, Landau-Ferey J, Grob R, Grosskopf D, Philippon J. Value of electroencephalogram in prediction and diagnosis of vasospasm after intracranial aneurysm rupture. Acta Neurochir (Wien) 1991;110:17–23. Saletu B, Anderer P, Saletu-Zyhlarz GM. EEG topography and tomography (LORETA) in the classification and evaluation of the pharmacodynamics of psychotropic drugs. Clin EEG Neurosci 2006;37:66–80. Sayama CM, Liu JK, Couldwell WT. Update on endovascular therapies for cerebral vasospasm induced by aneurysmal subarachnoid hemorrhage. Neurosurg Focus 2006;21:E12. Schmidt JM, Wartenberg KE, Fernandez A, Claassen J, Rincon F, Ostapkovich ND, et al. Frequency and clinical impact of asymptomatic cerebral infarction due to vasospasm after subarachnoid hemorrhage. J Neurosurg 2008;109: 1052–9. Sheorajpanday RV, Nagels G, Weeren AJ, De Deyn PP. Quantitative EEG in ischemic stroke: correlation with infarct volume and functional status in posterior circulation and lacunar syndromes. Clin Neurophysiol 2011a;122:884–90. Sheorajpanday RV, Nagels G, Weeren AJ, van Putten MJ, De Deyn PP. Quantitative EEG in ischemic stroke: correlation with functional status after 6 months. Clin Neurophysiol 2011b;122:874–83. Stuart RM, Waziri A, Weintraub D, Schmidt MJ, Fernandez L, Helbok R, et al. Intracortical EEG for the detection of vasospasm in patients with poor-grade subarachnoid hemorrhage. Neurocrit Care 2010;13:355–8. Sundt Jr TM, Sharbrough FW, Anderson RE, Michenfelder JD. Cerebral blood flow measurements and electroencephalograms during carotid endarterectomy. J Neurosurg 1974;41:310–20. Treggiari MM. Hemodynamic management of subarachnoid hemorrhage. Neurocrit Care 2011;15:329–35. Vergouwen MD, Ilodigwe D, Macdonald RL. Cerebral infarction after subarachnoid hemorrhage contributes to poor outcome by vasospasm-dependent and independent effects. Stroke 2011;42:924–9. Vespa PM, Nuwer MR, Juhasz C, Alexander M, Nenov V, Martin N, et al. Early detection of vasospasm after acute subarachnoid hemorrhage using continuous EEG ICU monitoring. Electroencephalogr Clin Neurophysiol 1997;103:607–15. Youden WJ. Index for rating diagnostic tests. Cancer 1950;3:32–5. Young GB, Campbell VC. EEG monitoring in the intensive care unit: pitfalls and caveats. J Clin Neurophysiol 1999;16:40–5.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
Contents lists available at ScienceDirect
Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph
Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults S. Gollwitzer a,⇑, T. Groemer a,b, S. Rampp a, M. Hagge a, D. Olmes a, H.B. Huttner a, S. Schwab a, D. Madzˇar a, R. Hopfengaertner a, H.M. Hamer a a b
Department of Neurology, University Hospital Erlangen, Schwabachanlage 6, 91054 Erlangen, Germany Department of Psychiatry, University Hospital Erlangen, Schwabachanlage 6, 91054 Erlangen, Germany
a r t i c l e
i n f o
Article history: Accepted 31 October 2014 Available online xxxx Keywords: Quantitative EEG Subarachnoid hemorrhage Delayed cerebral ischemia Vasospasm
h i g h l i g h t s Our study introduces a new algorithm for the prediction of imminent ischemia after subarachnoid
hemorrhage based on quantitative EEG monitoring. We developed an automated technique that does not require the presence of an EEG expert to prese-
lect or analyze raw EEG data and is therefore suitable for everyday usage by ICU staff. Our method indicated imminent ischemia earlier than established diagnostics such as transcranial
Doppler sonography or imaging and could therefore trigger intervention early enough to prevent infarction.
a b s t r a c t Objectives: Delayed cerebral infarction (DCI) has a significant impact on mortality and morbidity of patients with subarachnoid hemorrhage (SAH). The aim of this study was to define quantitative EEG (qEEG) parameters for the early and reliable prediction of DCI and compare the validity and time course of qEEG to standard procedures. Methods: 12 consecutive unselected SAH patients (8 female, mean age 52 years, Hunt-and-Hess grade I– IV) were prospectively examined. Continuous six channel EEG monitoring was started within 48 h after admission (mean duration 5.2 days; range: 2–12 days). All raw and unselected EEG signal underwent automated artifact rejection, Short Time Fast Fourier Transformation and a detrending procedure in order to analyze regional spectral power changes in different frequency bands. According to clinical standards, transcranial Doppler sonography (TCD) was performed at least on alternate days and repeat cerebral computer tomography (CCT) as needed. Results: 6 patients (50%) developed vasospasm/DCI. Decrease of P40% in power persisting over P5 h in the alpha band and P6 h in the theta band marked the optimal cut-off to detect DCI (sensitivity 89%, specificity 77% for alpha). EEG changes preceded detection of vasospasm/DCI in standard procedures by 2.3 days. Changes in the beta and delta band as well as in the alpha/delta ratio demonstrated lower correlation with imminent DCI. Conclusions: Focal reduction in alpha power may represent a valid, observer independent, non-invasive and continuous marker for vasospasm/DCI in SAH patients. Significance: qEEG indicates imminent ischemia earlier than established diagnostic tools, such as TCD. Ó 2014 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
1. Introduction ⇑ Corresponding author at: University Hospital Erlangen, Department of Neurology/Epilepsy Center, Schwabachanlage 6, 91054 Erlangen, Germany. Tel.: +49 9131 8539116; fax: +49 9131 8536469. E-mail address: [email protected] (S. Gollwitzer).
Delayed cerebral infarction (DCI) has a decisive impact on mortality and functional outcome after aneurysmal subarachnoid hemorrhage (SAH) (Vergouwen et al., 2011). Vasospasm measurable by angiography or transcranial Doppler (TCD) sonography has been
http://dx.doi.org/10.1016/j.clinph.2014.10.215 1388-2457/Ó 2014 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
2
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
strongly associated with DCI (Crowley et al., 2011; Brown et al., 2013). However, there is evidence that DCI may also be a consequence of other mechanisms such as cortical spreading depression or microcirculatory dysfunction undetectable by TCD (Dreier, 2011; Brown et al., 2013). Approximately 20% of DCI after SAH are not clinically detected or identified too late mostly due to impaired clinical assessment in poor-grade, comatose SAH patients (Schmidt et al., 2008). Therefore, the early detection of imminent ischemia being prerequisite to prevent infarction remains challenging. The EEG sensitively mirrors metabolic deterioration and disturbed neuronal activity following reduced cerebral blood flow (Astrup et al., 1981; O’Gorman et al., 2013). EEG changes such as flattening and loss of higher frequencies are found to occur rapidly if cerebral blood flow decreases below 0.16–0.17 ml g1 min1 (Sundt et al., 1974; Diedler et al., 2009). Continuous EEG monitoring has been shown to be a promising approach for early detection of imminent DCI in SAH patients (Labar et al., 1991; Rivierez et al., 1991; Vespa et al., 1997; Claassen et al., 2004, 2013; Stuart et al., 2010; Rathakrishnan et al., 2011). Power spectral analyses of raw EEG data represent a fast and robust screening method for longterm EEG recordings (Hopfengartner et al., 2007). The aim of this study was to develop a semiautomatic method to predict and localize reliably imminent deterioration due to DCI at the earliest time point. 2. Methods 2.1. Patients Participation in this prospective study was offered to all adult patients who were admitted to the Neurological Intensive Care Unit (NICU) of the University Hospital Erlangen between November 2011 and February 2013 and suffered from nontraumatic, out-of-hospital SAH regardless of clinical severity. No patient in our cohort had a history of stroke or other cerebral lesions possibly causing preexisting focal EEG slowing. Moreover, the initial CCT of all patients did not reveal signs of previous brain insults. The following inclusion criteria had to be fulfilled: age P 18 years and symptom onset <24 h before hospital admission. In all cases, the diagnosis was confirmed by CCT, which was performed immediately after admission and within the first 24 h of symptom onset. Patients were excluded from the study in cases of angiographic or sonographic proof of vasospasm before initiation of EEG monitoring. The study was approved by the local ethics committee. All patients or their relatives provided written informed consent before entering the study. 2.2. Clinical management and multimodal monitoring All patients underwent diagnostic angiography on the day of admission. SAH was caused by aneurysmal bleeding in all patients. All aneurysms were treated by endovascular coiling. There were no procedure-related infarctions. All patients were treated with nimodipine (Kronvall et al., 2009). Our current standard operating procedure includes monitoring of intracranial pressure (ICP); systolic, diastolic, and mean arterial blood pressure; heart rate; temperature; and oxygen saturation. Neurological status was surveyed at least three times per day by NICU staff. Wake-up tests for neurological assessment are not performed routinely during the critical phase of the disease. All patients received TCD or transcranial color-coded Doppler sonography (TCCS) at least every other day and when clinically indicated. No TCD monitoring was administered. CCT or MRI scans were performed in the case of clinical deterioration of the patient or worsening of TCD measurements. The relevant vasospasms were diagnosed by TCD (defined as peak systolic Doppler frequencies
>4 kHz) or TCCS (defined by cerebral flow velocities >200 cm/s, or an increase of >50 cm/s in 24 h according to the guidelines of the German Neurological Society) or by angiography. DCI was defined as new ischemic lesions on imaging that were not present on initial CCT performed within 24 h of symptom onset. In awake patients, delayed ischemic neurological deficits (DINDs) were defined as new focal neurological signs not detectable at the time of symptom onset or immediately after aneurysm occlusion and not caused by other complications such as rebleeding or hydrocephalus (Drenckhahn et al., 2012). In the case of relevant vasospasm or DCI/DIND, triple-H therapy (hypervolemia, hypertension, and hemodilution) (Treggiari, 2011) was initiated. If neurological deficits persisted, patients underwent interventional angiography and they were treated by chemical vasospasmolysis or balloon angioplasty (Sayama et al., 2006; Aburto-Murrieta et al., 2012). In the case of cerebral edema, osmotherapy with mannitol was administered and mild-to-moderate hypothermia was induced using an invasive cooling device (CoolgardÒ) if ICP elevation persisted. 2.3. EEG monitoring EEG monitoring was started within the first 48 h after admission and was maintained for a maximum of 12 , but it was discontinued earlier in the case of discharge from the ICU or if requested by patients or relatives. We restricted the EEG to a number of 10 electrodes attached according to the International 10–20 system: C3/ C4, P3/P4, O2/O1, T3/T4, and F3/F4 (alternatively F5/6 instead of F3/F4, if necessary due to head bandages). Simplified EEG montages are widely and successfully used for EEG monitoring under ICU conditions, although general recommendations regarding the best montage and number of electrodes are still lacking (Labar et al., 1991; Vespa et al., 1997; Kull and Emerson, 2005). We aimed for a compromise between stable maintenance of the recording and a full coverage of all vascular territories. EEG data were digitized at a sampling rate of 256 Hz with a high-pass filter of 0.08 Hz and a low-pass filter of 86 Hz. We applied a bipolar montage (F4–C4, T4–P4, P4–O2; F3–C3, T3–P3, P3–O1), hypothesizing that F4–C4 (F3–C3) and T4–P4 (T3–P3) represent the territory of the anterior cerebral artery (ACA) and middle cerebral artery (MCA) while P4– O2 (P3–O1) captures changes in the vertebrobasilar circulation. 2.4. qEEG analysis In order to perform reliable semiautomatic qEEG analysis, automated artifact rejection is essential, and we used a standardized and efficient technique that has been described and evaluated in detail in recent work (Hopfengartner et al., 2007). The remaining raw EEG signal underwent short-time Fourier transformation of 2-s epochs. The frequency distribution in each epoch was identified, which allowed for the determination of power values in the following frequency bands: delta 0.5–3.5 Hz, theta 4–7.5 Hz, alpha 8–12 Hz, and beta 12.5–22 Hz. All EEG channels were subject to common fluctuations due to global effects such as changes of ICP, level of sedation, cooling and medication, or sleep–wake patterns. In order to eliminate such global or bilateral effects and unmask focal alterations, which we postulated to be the correlate of imminent DCI, several procedures were necessary. Because all channels correlated highly in their power (R > 0.92 and p < 0.001 for each combination of channels) over time while having different power levels, a detrending procedure was established that eliminated common fluctuations and uncovered significant regional drops in the EEG power. First, every median 1-min power measurement was divided by the median of all 1-min values of the same channel over the first
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
3
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
24 h of EEG recording for normalization. Second, a detrending procedure was performed. This was carried out by calculating the median of all normalized power values of the six channels at a given point in time generating the normalized common trend of all channels. Each single-channel power value was then divided by the common trend to eliminate general fluctuations and stress focal changes. In order to detect focal EEG changes over time from day 2 of monitoring, the median of every 60 normalized and detrended 1min EEG power measurements per hour of EEG recording in each channel was compared to the median normalized and detrended power value of the same channel over the baseline period. The baseline was defined as a minimum of 6 h and a maximum of 12 h of uninterrupted EEG recording within the first 24 h of EEG monitoring assuming that vasospasm or DCI were absent in this time period. Raw EEG of each monitoring day was also visually inspected by two trained electroencephalographers (SG and HMH) in order to confirm qEEG findings and exclude possible mimics of EEG power changes caused by artifacts, periodic complexes, or epileptic activity. 2.5. Statistics We correlated qEEG changes in each individual patient with the clinical end points, ‘‘delayed cerebral ischemia’’ or ‘‘increased risk of delayed cerebral ischemia,’’ which were marked by at least one of the following conditions: (1) proof of vasospasm in TCD/ TCCS measurements, (2) new cerebral infarction in CCT/MRI, and (3) new focal neurological deficit not otherwise explained. In addition, the time course of qEEG alterations was compared to the time point when the defined clinical end points were first noticed by standard diagnostics in order to assess the capability of qEEG to predict DCI. To calculate the strength of association between qEEG changes and DCI, we applied receiver operator characteristics (ROC) considering each hemisphere separately. In order to explore the informative value of the magnitude and duration of EEG power decrease, we assessed the sensitivity and specificity of different power decline scenarios ranging from 10% to 100% and persisting for a minimum of 1 h and a maximum of 8 h. To define the optimal threshold value on the ROC curve, we calculated the Youden index (J), which is a function of sensitivity and specificity (Youden, 1950). 3. Results Of 22 consecutive SAH patients admitted during the study period, 12 patients fulfilled the inclusion criteria and were included (Table 1), while 10 were admitted to hospital very late during
the course of the disease for baseline acquisition. Of the included patients, five were sedated and mechanically ventilated during the complete EEG monitoring, while seven were awake and breathing spontaneously. Sedation was achieved by continuous intravenous administration of opioid analgesics and either midazolam or propofol. In two patients, intravenous catecholamines were administered intermittently. No other medications that may have an influence on the EEG, such as neuroleptics or antidepressants, were administered to any of the patients. In four of the sedated patients, visual inspection of raw EEG revealed reduction of alpha activity, and the fifth sedated patient showed intermittent generalized slowing, probably due to the medication, which consisted of midazolam and sufentanil in this case (Herkes et al., 1992; Saletu et al., 2006). No triphasic waves, periodic complexes, or interictal or ictal activity was noted in the raw EEG including patients 3 and 8 who suffered from a single seizure prior to admission (Table 2). In all cases, visual inspection of raw EEG confirmed qEEG findings. Only one patient showed mildly increased transaminases; no patient developed renal, liver, or multiorgan failure. Detailed information on complications and clinical outcome is given in Table 2. Fifty percent (n = 6) of the patients developed vasospasm (Table 3). The territories of the MCA and ACA were most often affected. Isolated vasospasm of the posterior cerebral artery (PCA) and basilar artery (BA) did not occur. In two patients, vasospasm led to infarctions seen on CT/MRI. One of the two patients died due to complete infarction of the left MCA and ACA territory. DIND in the form of an incomplete paresis of the right arm was noted in one awake patient (patient 11) on day 10, 1 day after the first occurrence of left-hemispheric vasospasm. Decreases in the total alpha and theta power showed the strongest association with vasospasm/DCI (Youden index J = 0.66 Table 4, Fig. 1). In both frequency bands, a power decline of 40% marked the statistically optimal threshold. The duration of the power decrease providing the highest significance was 5 h for alpha and 6 h for theta. The optimal combined cut-off for power decrease and duration proved to be highly sensitive (89%) with a specificity of 77% for both frequency bands (Table 4). Changes in the beta frequency band demonstrated lower correlation with vasospasm/DCI; this was likely due to contamination by muscle artifacts as visual inspection of raw EEG revealed. The weakest correlation existed between DCI and fluctuation in delta power. Consequentially, the alpha–delta ratio showed only a moderate correlation with vasospasm/DCI (sensitivity of 55% and specificity of 85%). A significant alpha decrease was seen in six patients (Table 3). Three of these patients were awake, responding, and following commands. No sedative agents were given in these cases. The
Table 1 Patients’ characteristics. Patient
Age (years)/sex
Clinical grade (Hunt and Hess)
Fisher grade
Location of aneurysm
EEG recording (days)
EVD
Intubated
1 2 3 4 5 6 7 8 9 10 11 12
70/f 51/f 34/f 73/f 38/f 44/f 48/m 53/m 51/m 73/f 45/f 49/f
2 1 3 2 1 4 1 4 3 2 3 2
3 1 3 4 4 3 4 2 4 4 4 3
L PCOM BA ACOM ACOM R PCA R PCOM ACOM ACOM ACOM ACOM BA R ACI
2 6 12 4 3 4 5 6 7 8 7 10
No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
No No Yes No No Yes No Yes Yes No No Yes
Mean (SD)
52.4 (13.0)
5.2 (3.2)
L: left; R: right; f: female; m: male; ACA: anterior cerebral artery; ACOM: anterior communicating artery; BA: basilar artery; MCA: middle cerebral artery; PCA: posterior cerebral artery; PCOM: posterior communicating artery; SD: standard deviation; and EVD: external ventricular drain.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
4
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Table 2 Complications and clinical outcome of patients. Patient
Complication
Time of complication (day SAH)
Clinical outcome (mRS)
1
None
2
CSF circulatory dysfunction Infection, fever
1–4 2
0
3
Seizure NSTEMI Elevated ICP CSF circulatory dysfunction
Prior to admission 1 1–8 1–12
2
4
CSF circulatory dysfunction Infection (ventriculitis), fever
2–11 4
0
5
CSF circulatory dysfunction
2–10
0
6
CSF circulatory dysfunction Infection, fever Elevated ICP Elevated transaminases
1–16 2 2–7 10
5
7
Infection, fever CSF circulatory dysfunction
1 2–7
0
8
Seizure CSF circulatory dysfunction Infection (ventriculitis), fever
Prior to admission 1–4 4
2
9
CSF circulatory dysfunction Complete infarction of left MCA territory
1–10 9
6
10
Elevated ICP CSF circulatory dysfunction
1–7 1–15
5
11
Infection, fever
3
1
12
Delirium anemia
4 4–13
1
1
mRS: modified Rankin scale (Banks et al., 2007) assessed on day of discharge; CSF: cerebrospinal fluid; NSTEMI: non-ST-elevated myocardial infarction.
Table 3 Identification and localization of vasospasm/DCI by different modalities (qEEG, TCD, and CCT/MRI). Patient
EEG (reduced alpha power)a Channel
No DCI 1 4 5 7 8 100 DCI 20
3 6 9
110 12
Onsetb (day SAH)
TCD/TCCS (increased flow velocity)
CT/MRI (infarction)
Durationc (days)
Territory
Territory
– – – – – P3O1
3
4.5
– – – – – –
F4C4 T4P4
6 6
F4C4 F3C3 F4C4 T4P4 F3C3 T3P3 F3C3 T3P3 –
4 5 3 3 3 3 4 4 –
700 700 – 5–1200 6–1200 4,500 4,5,6 4,5,700 4–700 5–700 5–700 –
R ACA R MCA L MCA R ACA L ACA R MCA R MCA – – L ACA L MCA Ubiquitous
Days SAH
Days SAH
– – – – – – 2,7 2,5–9 5 9 9 6,7,10,13 6,7,8 – – 12,14 9,10–12,14 4,6–8,11
R ACA L ACA
12 12
L ACAd L MCAd
9 9
–
–
a
EEG: event defined as at least 40% decrease of alpha power persisting P5 h. b Onset: initial occurrence of EEG event. c Duration: following days during course of disease fulfilling criteria for EEG event. ‘‘Power decrease persisted until end of monitoring’’. d Patient died because of brain infarction. ‘Awake patients with decrease of alpha power’. Patient 2 and 10 did not reveal any neurological deficits, whereas patient 11 suffered from incomplete paresis of right arm from day 10 (DIND). L: left; R: right; ACA: anterior cerebral artery; MCA: middle cerebral artery; PCA: posterior cerebral artery.
remaining three patients were sedated, and they did not react to stimuli throughout the monitoring. However, brain-stem reflexes remained present. In four cases, EEG changes preceded vasospasm detected by TCD or DCI seen in CCT/MRI. On average, EEG power decreases
occurred 2.3 days (SD 3.3, range 2–5) earlier than proof of vasospasm in TCD/TCCS or DCI in CCT/MRI. Diminution of alpha power was persistent over several days in four cases. Recovery of alpha power was only seen over the right hemisphere of patient 9, concordant with the fact that no right-
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
5
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx Table 4 Association between decrease in qEEG power and vasospasm/DCI. Frequency band
Alpha Beta Delta Theta A/D ratio
Optimal cut-off Power decrease (%)
Duration (h)
40 50 70 40 70
5 3 2 6 2
Youden index (J)
Area under the curve (AUC)
Sensitivity (%)
Specificity (%)
0.66 0.47 0.24 0.66 0.40
0.66 0.66 0.52 0.72 0.64
89 78 55 89 55
77 69 69 77 85
Optimal cut-off values of power decrease and duration according to Youden index; AUC derived from receiver operator characteristics (ROC); and specificity and sensitivity weighted equally.
hemispheric infarction developed as shown by imaging. Details on the duration of power decreases are given in Table 3. In patient 2, blood flow velocity acceleration was seen in TCD first on day 2 in the right ACA and MCA, and exclusively in the left MCA on day 5. The EEG did not reveal any focal deterioration over the left hemisphere, and it captured alpha power decline over the right hemisphere not earlier than day 6. Inspection of raw EEG confirmed these findings. The patient was awake and breathing spontaneously during the entire monitoring period and did not develop neurological deficits or ischemic lesions, although no therapeutic countermeasures against vasospasm were taken. We, therefore, considered the EEG to be in line with imaging and the clinical status of the patient, but we had to regard the left-hemispheric EEG findings as false negative according to the study protocol. In patient 10, alpha power dropped significantly in the left posterior region for 3 days without proof of vasospasm/DCI in TCD or CCT. Visual inspection of the original EEG at the time revealed focal slowing and amplitude flattening without artifacts blurring the EEG. Patient 12 showed ubiquitous increased flow velocities identified by TCD/TCCS on day 4 (Table 2) which were not reflected by qEEG post processing. The patient, however, did not develop DCI and recovered without neurological deficits. Moreover, the patient suffered from anemia with hemoglobin levels dropping below 100 g/l from day 4 after symptom onset and only recovered from day 13 paralleling TCCS changes. Therefore, we regarded the anemia as the cause of the TCCS flow acceleration while the assumption of ubiquitous critical vasospasm not leading to any infarction or neurological deficit appeared less plausible. Patient 9 revealed marked bilateral alpha power reduction starting on day 3 (Fig. 2). TCD detected vasospasm in the right MCA on day 7. No vasospasm was seen on the left side in TCD throughout the monitoring. While alpha power in the right hemisphere recovered during the following days, it continued to decline in all channels on the left. CCT and MRI, triggered by clinical deterioration of the patient on day 9, showed complete infarction of the left ACA and MCA territory leading to the patient’s death. As alpha power on the left side dropped below the threshold long before clinical deterioration was noted and infarction was seen in CCT, we believe that the power decline marked the initial compromised perfusion in the MCA territory rather than the infarct. However, no CCT or MRI was performed at the time of EEG deterioration. Alternative explanations for the bilateral EEG changes, for example, changes in the sedation regime, could not be identified.
4. Discussion This study aims to introduce an automated, easy-to-use, reliable, and robust algorithm for the prediction of imminent ischemia in SAH patients early enough to allow for countermeasures possibly preventing infarction. In order to warrant clinical practicability, we applied a novel algorithm based on continuous EEG monitoring
without visual preselection of specific EEG sections. In this setting, >40% decrease in alpha or theta power for longer than 5 or 6 h, respectively, showed a strong association with vasospasm or DCI and was able to predict this complication 2.3 days on average before it was detected by other methods, such as TCD, in the majority of patients. EEG changes preceded identification of DCI in other diagnostics in four out of five patients, while only one patient exhibited early increased accelerated flow velocities in TCD not accompanied by EEG changes at that time. However, this patient neither deteriorated clinically nor developed cerebral infarction in the later course of the disease. Apart from the prediction of DCI, EEG was also able to lateralize the pathology correctly in all cases. Exact localization of the vascular territory was not always achieved by EEG. As therapeutic consequences are rather dependent on the detection of the presence of DCI than its exact localization, we aimed for a reliable method to predict the event but not necessarily to localize it precisely. However, this approach introduces its own set of challenges. First, maintenance of high-quality EEG monitoring over several days under ICU conditions is difficult and labor intensive (Young and Campbell, 1999). Second, various artifacts unavoidably contaminate EEG recordings on ICU. Third, EEG is sensitive but unspecific to various systemic changes, making it difficult to distinguish DCI from other factors such as metabolic alterations, ICP changes, or effects of sedation. We tried to address these shortcomings by using a reduced but multichannel montage that could be maintained over days with a reasonable effort and was still able to detect reliably focal EEG changes. In addition, an efficient automatic artifact rejection technique was implemented, which was evaluated extensively in previous studies (Hopfengartner et al., 2007). This allowed for unbiased analysis of changes in the alpha, theta, and delta frequency bands. It was, however, impossible to differentiate automatically between beta activity and muscle artifacts, which led to low usability of beta fluctuations. The key to the discrimination of EEG changes triggered by DCI from changes due to other causes was the perception of vasospasm as a predominantly focal event (Miller and Palestrant, 2012). Most other incidents with impact on the EEG of ICU patients suffering from SAH cause global changes. By detrending, general fluctuations were excluded from further analyses, thus making focal changes visible in qEEG. Especially changes caused by different states of wakefulness and medication effects were eliminated due to their non-focal nature. For example, increases in higher frequencies that are often induced by benzodiazepines were extinguished as they affected many channels equally on both sides. A theoretical disadvantage of this approach could be its lack of detection of global changes, which cannot be excluded to be also caused by SAH. However, in five out of six patients in our cohort, DCI was clearly restricted to one or two adjacent arterial territories; therefore, it was well detected by the algorithm. One patient exhibited ubiquitous increased flow velocities in TCD/TCCS that was more likely due to anemia (Brass et al., 1988; Ameriso et al.,
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
6
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Fig. 1. Statistical analysis: Youden index (J) for different frequency bands. Percentage of power decrease (y-axis) plotted against duration of power decrease in hours (x-axis). Youden index expressed by gray shades, corresponding to bar on the right. High J-values illustrated by light gray shades implicated strong correlation. Best results were achieved in the alpha band (J = 0.66; optimal threshold 40% decrease, duration of 5 h) and the theta band (J = 0.66; optimal threshold 40% decrease, duration of 6 h). Changes in beta, delta, and alpha–delta ratio were less sensitive and specific for DCI (beta: J = 0.47; delta: J = 0.24; and alpha–delta ratio: J = 0.40). ROC curve for varying power decrease values and fixed duration (5 h); for AUC values, see Table 3.
1990; Isikay et al., 2005) than due to generalized vasospasm. The team of neurointensivists treating the patient shared this interpretation and the patient did not receive any specific treatment for vasospasms. In contrast to the TCD/TCCS findings, the patient recovered without neurological deficit and TCD normalized with
rising hemoglobin levels. We, therefore, assume that EEG was not prone to this specific error source and may be superior to TCD in this situation. A strength of EEG is its ability to directly detect disturbed neuronal activity regardless of the underlying cause rather than to
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
7
Fig. 2. Correlation of alpha power, TCD, and imaging: patient 9, SAH III, aneurysm of anterior communicating artery; (A) TCD showing critical vasospasm of the right MCA on days 7–10; no increase of frequency in left MCA; (B) EEG alpha power: marked decrease of alpha power in all channels on days 3 and 4 (black arrows); recovery of alpha power in channels T4P4 and F8C4 on day 6/7 (gray arrow); no recovery in T3P3/F7C3; (C) CCT and ; (D) perfusion weighted MRI on day 10 showing complete infarction of the left ACA and MCA territory.
focus on vasoconstriction in larger vessels as an indirect marker of neuronal damage. For many years, vasospasm has been regarded as the main cause for ischemia in SAH patients. There is, however, emerging evidence that DCI has a multifactorial pathogenesis, as patients can develop cerebral infarction in vascular territories unaffected by angiographic vasoconstriction (Naidech et al., 2006). An alternative pathophysiological mechanism potentially contributing to the development of DCI may be cortical spreading depression (CSD). Animal models have shown that SAH can cause a primary cellular dysfunction, which can induce spreading
depolarizations accompanied by alterations in ion homeostasis (Hubschmann and Kornhauser, 1982). Dreier et al. found that hemoglobin in the subarachnoid space can cause spreading ischemia leading to widespread cortical necrosis (Dreier et al., 2000, 2006; Drenckhahn et al., 2012). It has also been shown in microdialysis studies that changes in brain metabolism such as an increase in lactate–pyruvate ratio and decreased brain tissue oxygenation preceding imaging are proof of ischemia. EEG could be a powerful tool to capture these changes at an early stage (Helbok et al., 2011).
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
8
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Fig. 3. Alpha and delta power during different stages of ischemia: patient 3, SAH III, aneurysm of anterior communicating artery. (A) Critical vasospasm in left and right ACA on day 9 as detected by TCD; (B) alpha power decrease on days 4 and 5 in F8C4 and F7C3 (black arrow), no relevant EEG changes in T4P4 and T3P3; (C) delta power increase in F8C4 and F7C3 on day 7/8 (gray arrow); and (D) moderate decrease of alpha/delta ratio from day 6 (black arrow).
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
To be a reliable marker for DCI, the alpha and theta power decreases had to be persistent over a certain period of time. The optimum cut-off duration was slightly shorter for alpha than for theta. Accordingly, alpha appears to be the preferred parameter in awake as well as in comatose patients. In our cohort, physiological changes such as sleep and wake patterns (including alpha background activity) were symmetrical in appearance and were at least in parts subject to detrending, and they remained below the power threshold, which was associated with vasospasm or DCI. Thus, the method seems applicable for less-affected awake patients as well as for poor-grade SAH patients. There was one patient in our cohort without any proof of vasospasm or DCI in TCD and neuroimaging who demonstrated a significant isolated left occipital alpha decrease, which had to be interpreted as false positive according to the study protocol. As theta power decreased equally in this patient, the alternative explanation is that the sensitive qEEG correctly identified reduced cortical perfusion in this area as a sign of imminent DCI, but clinical treatment or the natural course of the disease prevented the perfusion deficit from progressing into persistent and more severe DCI, which would also be identifiable by other diagnostic tools. While previous studies focused on changes in delta power to detect DCI, it is the novelty of our study to show the significance of alpha power decreases in this context. Although alpha may not be prominent especially anteriorly in awake patients on visual analysis of EEG, Fourier analysis reveals that alpha is a part of the spectrum in all brain regions and states of wakefulness and it is therefore eligible for the purpose of the study. Unexpectedly, delta power changes and the alpha–delta ratio showed only weak-to-moderate correlations with vasospasm in our cohort. At first glance, this result appears to be in contrast to previous studies. Claassen et al. found a high probability for patients to develop DCI when the alpha–delta ratio decreased by >10% in six consecutive recordings or by >50% in a single recording to develop DCI (Claassen et al., 2004). Stuart et al. also reported the alpha–delta ratio to be most predictive for DCI in intracortical EEG (Stuart et al., 2010). A possible explanation for the divergent findings may be the hypothesis corroborated by Machado et al. (Machado et al., 2004) who applied qEEG to patients with MCA ischemic strokes and found delta power increases to be related to the ischemic core while the tissue at risk, penumbra, and edema were characterized by alpha and theta alterations. Correspondingly, the alpha decrease in our study marked the early point where perfusion dropped below a critical level, generating tissue at risk of, but not yet persistent, ischemic infarction. This hypothesis is supported by patient 3 who developed ischemic infarction during the course of the disease. Delta power increased 2 days after alpha power declined, probably indicating irreversible ischemia (Fig. 3). In this context, the alpha/delta ratio could be a valuable parameter to estimate the prognosis and predict the patient’s functional outcome (Burghaus et al., 2007; Sheorajpanday et al., 2011a,b) while the alpha power may be the superior parameter to prevent infarction by initiating countermeasures. Although the diminution of alpha power as a marker of DCI has not yet been confirmed by other studies, our results give reasons to regard alpha power as a sensitive and specific predictor of DCI in the context of pathophysiological considerations. Claassen et al. found a strong correlation between a decrease of the alpha–delta power ratio and DCI (a sensitivity of 100% and a specificity of 76%) by analyzing artifact-free clips visually preselected by experienced electroencephalographers (Claassen et al., 2004). Our approach introduces an automated technique that does not require the presence of an EEG expert to preselect or analyze raw EEG data, and it is therefore suitable for everyday use by ICU staff. Vespa et al. also chose a continuous approach and identified the relative alpha (RA) variability as the most sensitive (100%)
9
indicator of imminent ischemia (Vespa et al., 1997). Although highly sensitive, RA variability not only corresponded exclusively to vasospasm or DCI but also dropped significantly as a consequence of a variety of other conditions, for example, increase of ICP or herniation, causing low specificity (50%). Labar et al. used a two-channel EEG montage to compare total power changes and plotted consistent EEG trends day to day and left to right (Labar et al., 1991). To be identified as persistent, the EEG trend had to persist over 24 h in this study. The algorithm used in our study found a focal alpha power decrease persistent over just 5 h to be highly sensitive and specific for vasospasm/DCI, thus allowing for earlier intervention. There are several limitations to our study. First, the small number of patients limited the statistical power of our results. Despite the small number of analyzed patients, the results of this proof-ofprinciple study were significant and supported the hypothesis that qEEG is a reliable predictor of DCI. However, verification of the results in larger cohorts is needed in order to be able to rely on the clinical utility of this approach. Another restriction of the algorithm is the necessity of a vasospasm-free baseline EEG. Therefore, patients had to be excluded when they were admitted to our hospital at later stages of their disease when vasospasms may already have been present. We are also aware that the restricted number of EEG electrodes lowers the spatial resolution of our findings but increases the practicability of the method in ICU settings. A possible problem with our approach is the detection of DCI in cases with preexisting focal slowing due to prior ischemic strokes or cerebral lesions. As no patient in our cohort had a history of stroke or structural cerebral lesion nor showed signs of it in the initial CCT, it is unclear if the power decline in comparison to baseline would be large enough in such cases to exceed the threshold established in this study. In conclusion, our study introduces an automated, easy-to-use, reliable, and robust algorithm for the prediction of imminent ischemia in SAH patients early enough to allow for countermeasures possibly preventing infarction. In contrast with established diagnostics such as TCD or imaging, the real-time character of this method creates the opportunity for an EEG-based alarm system that has the potential to substantially improve the functional outcome of SAH patients. Sources of funding The study was not funded. Acknowledgment We would like to thank Arndt Dörfler, MD, Department of Neuroradiology, for the provision of MRI and CT data. Conflict of interest: None of the authors have potential conflict of interest to be disclosed. References Aburto-Murrieta Y, Marquez-Romero JM, Bonifacio-Delgadillo D, Lopez I, Hernandez-Curiel B. Endovascular treatment: balloon angioplasty versus nimodipine intra-arterial for medically refractory cerebral vasospasm following aneurysmal subarachnoid hemorrhage. Vasc Endovascular Surg 2012;46:460–5. Ameriso SF, Paganini-Hill A, Meiselman HJ, Fisher M. Correlates of middle cerebral artery blood velocity in the elderly. Stroke 1990;21:1579–83. Astrup J, Siesjo BK, Symon L. Thresholds in cerebral ischemia – the ischemic penumbra. Stroke 1981;12:723–5. Brass LM, Pavlakis SG, DeVivo D, Piomelli S, Mohr JP. Transcranial Doppler measurements of the middle cerebral artery. Effect of hematocrit. Stroke 1988;19:1466–9. Brown RJ, Kumar A, Dhar R, Sampson TR, Diringer MN. The relationship between delayed infarcts and angiographic vasospasm after aneurysmal subarachnoid hemorrhage. Neurosurgery 2013;72:702–8.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215
10
S. Gollwitzer et al. / Clinical Neurophysiology xxx (2015) xxx–xxx
Burghaus L, Hilker R, Dohmen C, Bosche B, Winhuisen L, Galldiks N, et al. Early electroencephalography in acute ischemic stroke: prediction of a malignant course? Clin Neurol Neurosurg 2007;109:45–9. Claassen J, Hirsch LJ, Kreiter KT, Du EY, Connolly ES, Emerson RG, et al. Quantitative continuous EEG for detecting delayed cerebral ischemia in patients with poorgrade subarachnoid hemorrhage. Clin Neurophysiol 2004;115:2699–710. Claassen J, Taccone FS, Horn P, Holtkamp M, Stocchetti N, Oddo M. Recommendations on the use of EEG monitoring in critically ill patients: consensus statement from the neurointensive care section of the ESICM. Intensive Care Med 2013;39:1337–51. Crowley RW, Medel R, Dumont AS, Ilodigwe D, Kassell NF, Mayer SA, et al. Angiographic vasospasm is strongly correlated with cerebral infarction after subarachnoid hemorrhage. Stroke 2011;42:919–23. Diedler J, Sykora M, Bast T, Poli S, Veltkamp R, Mellado P, et al. Quantitative EEG correlates of low cerebral perfusion in severe stroke. Neurocrit Care 2009;11: 210–6. Dreier JP. The role of spreading depression, spreading depolarization and spreading ischemia in neurological disease. Nat Med 2011;17:439–47. Dreier JP, Ebert N, Priller J, Megow D, Lindauer U, Klee R, et al. Products of hemolysis in the subarachnoid space inducing spreading ischemia in the cortex and focal necrosis in rats: a model for delayed ischemic neurological deficits after subarachnoid hemorrhage? J Neurosurg 2000;93:658–66. Dreier JP, Woitzik J, Fabricius M, Bhatia R, Major S, Drenckhahn C, et al. Delayed ischaemic neurological deficits after subarachnoid haemorrhage are associated with clusters of spreading depolarizations. Brain 2006;129:3224–37. Drenckhahn C, Winkler MK, Major S, Scheel M, Kang EJ, Pinczolits A, et al. Correlates of spreading depolarization in human scalp electroencephalography. Brain 2012;135:853–68. Helbok R, Madineni RC, Schmidt MJ, Kurtz P, Fernandez L, Ko SB, et al. Intracerebral monitoring of silent infarcts after subarachnoid hemorrhage. Neurocrit Care 2011;14:162–7. Herkes GK, Wszolek ZK, Westmoreland BF, Klass DW. Effects of midazolam on electroencephalograms of seriously ill patients. Mayo Clin Proc 1992;67:334–8. Hopfengartner R, Kerling F, Bauer V, Stefan H. An efficient, robust and fast method for the offline detection of epileptic seizures in long-term scalp EEG recordings. Clin Neurophysiol 2007;118:2332–43. Hubschmann OR, Kornhauser D. Effect of subarachnoid hemorrhage on the extracellular microenvironment. J Neurosurg 1982;56:216–21. Isikay CT, Uzuner N, Gucuyener D, Ozdemir G. The effects of hematocrit and age on transcranial Doppler measurements in patients with recent ischemic stroke. Neurol India 2005;53:51–4. discussion 4. Kronvall E, Undren P, Romner B, Saveland H, Cronqvist M, Nilsson OG. Nimodipine in aneurysmal subarachnoid hemorrhage: a randomized study of intravenous or peroral administration. J Neurosurg 2009;110:58–63. Kull LL, Emerson RG. Continuous EEG monitoring in the intensive care unit: technical and staffing considerations. J Clin Neurophysiol 2005;22:107–18. Labar DR, Fisch BJ, Pedley TA, Fink ME, Solomon RA. Quantitative EEG monitoring for patients with subarachnoid hemorrhage. Electroencephalogr Clin Neurophysiol 1991;78:325–32. Machado C, Cuspineda E, Valdes P, Virues T, Llopis F, Bosch J, et al. Assessing acute middle cerebral artery ischemic stroke by quantitative electric tomography. Clin EEG Neurosci 2004;35:116–24.
Miller CM, Palestrant D. Distribution of delayed ischemic neurological deficits after aneurysmal subarachnoid hemorrhage and implications for regional neuromonitoring. Clin Neurol Neurosurg 2012;114:545–9. Naidech AM, Drescher J, Tamul P, Shaibani A, Batjer HH, Alberts MJ. Acute physiological derangement is associated with early radiographic cerebral infarction after subarachnoid haemorrhage. J Neurol Neurosurg Psychiatry 2006;77:1340–4. O’Gorman RL, Poil SS, Brandeis D, Klaver P, Bollmann S, Ghisleni C, et al. Coupling between resting cerebral perfusion and EEG. Brain Topogr 2013;26:442–57. Rathakrishnan R, Gotman J, Dubeau F, Angle M. Using continuous electroencephalography in the management of delayed cerebral ischemia following subarachnoid hemorrhage. Neurocrit Care 2011;14:152–61. Rivierez M, Landau-Ferey J, Grob R, Grosskopf D, Philippon J. Value of electroencephalogram in prediction and diagnosis of vasospasm after intracranial aneurysm rupture. Acta Neurochir (Wien) 1991;110:17–23. Saletu B, Anderer P, Saletu-Zyhlarz GM. EEG topography and tomography (LORETA) in the classification and evaluation of the pharmacodynamics of psychotropic drugs. Clin EEG Neurosci 2006;37:66–80. Sayama CM, Liu JK, Couldwell WT. Update on endovascular therapies for cerebral vasospasm induced by aneurysmal subarachnoid hemorrhage. Neurosurg Focus 2006;21:E12. Schmidt JM, Wartenberg KE, Fernandez A, Claassen J, Rincon F, Ostapkovich ND, et al. Frequency and clinical impact of asymptomatic cerebral infarction due to vasospasm after subarachnoid hemorrhage. J Neurosurg 2008;109: 1052–9. Sheorajpanday RV, Nagels G, Weeren AJ, De Deyn PP. Quantitative EEG in ischemic stroke: correlation with infarct volume and functional status in posterior circulation and lacunar syndromes. Clin Neurophysiol 2011a;122:884–90. Sheorajpanday RV, Nagels G, Weeren AJ, van Putten MJ, De Deyn PP. Quantitative EEG in ischemic stroke: correlation with functional status after 6 months. Clin Neurophysiol 2011b;122:874–83. Stuart RM, Waziri A, Weintraub D, Schmidt MJ, Fernandez L, Helbok R, et al. Intracortical EEG for the detection of vasospasm in patients with poor-grade subarachnoid hemorrhage. Neurocrit Care 2010;13:355–8. Sundt Jr TM, Sharbrough FW, Anderson RE, Michenfelder JD. Cerebral blood flow measurements and electroencephalograms during carotid endarterectomy. J Neurosurg 1974;41:310–20. Treggiari MM. Hemodynamic management of subarachnoid hemorrhage. Neurocrit Care 2011;15:329–35. Vergouwen MD, Ilodigwe D, Macdonald RL. Cerebral infarction after subarachnoid hemorrhage contributes to poor outcome by vasospasm-dependent and independent effects. Stroke 2011;42:924–9. Vespa PM, Nuwer MR, Juhasz C, Alexander M, Nenov V, Martin N, et al. Early detection of vasospasm after acute subarachnoid hemorrhage using continuous EEG ICU monitoring. Electroencephalogr Clin Neurophysiol 1997;103:607–15. Youden WJ. Index for rating diagnostic tests. Cancer 1950;3:32–5. Young GB, Campbell VC. EEG monitoring in the intensive care unit: pitfalls and caveats. J Clin Neurophysiol 1999;16:40–5.
Please cite this article in press as: Gollwitzer S et al. Early prediction of delayed cerebral ischemia in subarachnoid hemorrhage based on quantitative EEG: A prospective study in adults. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.10.215