Quantitative continuous EEG for detecting delayed cerebral ischemia in patients with poor-grade subarachnoid hemorrhage

Transcription

1 Clinical Neurophysiology 115 (2004) Quantitative continuous EEG for detecting delayed cerebral ischemia in patients with poor-grade subarachnoid hemorrhage Jan Claassen a,b, *, Lawrence J. Hirsch b, Kurt T. Kreiter a, Evelyn Y. Du d, E. Sander Connolly c, Ronald G. Emerson b, Stephan A. Mayer a,c a The Division of Stroke and Critical Care Neurology, Department of Neurology, Neurological Institute, Columbia University, 710 West 168th Street, Unit 1, New York, NY 10032, USA b The Comprehensive Epilepsy Center, Department of Neurology, Neurological Institute, Columbia University, New York, NY 10032, USA c Department of Neurosurgery, College of Physicians and Surgeons, Columbia University, New York, NY, USA d Department of Biostatistics, School of Public Health, Columbia University, New York, NY, USA Accepted 24 June 2004 Available online 3 August 2004 Abstract Objective: Delayed cerebral ischemia (DCI) due to vasospasm is often undetected by clinical exam in patients with poor-grade subarachnoid hemorrhage (SAH). The purpose of this study was to identify quantitative EEG (qeeg) parameters that are most sensitive and specific for the detection of DCI in stuporous or comatose SAH patients. Methods: Of 78 consecutive Hunt Hess grade 4 or 5 SAH patients admitted to our Neuro-ICU over a 2-year period, 48 were eligible for participation and 34 were enrolled. Continuous EEG monitoring was performed from post-operative day 2 to post-sah day 14. In each patient, 20 artifact-free, 1 min EEG-clips following an alerting stimulus were analyzed: 10 clips were obtained on monitoring day 1 (baseline), and 10 on days 4 6 (follow-up). In DCI patients, follow-up clips were obtained after the onset of deterioration and before infarction had occurred. Twelve qeeg parameters were calculated using fast Fourier transformation; generalized estimating equations were used to compare ratios of change in qeeg parameters in patients with and without DCI. Results: Nine of 34 patients (26%) developed DCI. The alpha/delta ratio (alpha power/delta power; ADR) demonstrated the strongest association with DCI. The median decrease of ADR for patients with DCI was 24%, compared to an increase of 3% for patients without DCI (ZZ4.0, P!0.0001). Clinically useful cut-offs included 6 consecutive recordings with a O10% decrease in ADR from baseline (sensitivity 100%, specificity 76%) and any single measurement with a O50% decrease (sensitivity 89%, specificity 84%). Conclusions: A decrease in the ADR may be a sensitive method of detecting DCI, with reasonable specificity. This post-stimulation qeeg parameter may supplement the clinical exam in poor-grade SAH patients and may prove useful for the detection of DCI. Significance: Following ADRs may allow earlier detection of DCI and initiation of interventions at a reversible stage, thus preventing infarction and neurological morbidity. q 2004 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. Keywords: Delayed cerebral ischemia; Subarachnoid hemorrhage; Cerebral aneurysm; Vasospasm; Continuous EEG monitoring 1. Introduction Delayed cerebral ischemia (DCI) from vasospasm is an important cause of morbidity and mortality after * Corresponding author. Tel.: C ; fax: C address: jc1439@columbia.edu (J. Claassen). subarachnoid hemorrhage (SAH) (Kassell et al., 1982; Mayberg, 1998). Angiographic vasospasm is detected in 50 70% of patients with SAH (Weir et al., 1978) and DCI occurs in 19 46% of SAH patients (Charpentier et al., 1999; Claassen et al., 2001; Hijdra et al., 1988; Hop et al., 1999; Murayama et al., 1997; Qureshi et al., 2000) and 54% of poor-grade SAH patients (Kirmani et al., 2002). One of the greatest challenges for the acute management of poor-grade /$30.00 q 2004 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. doi: /j.clinph

2 2700 J. Claassen et al. / Clinical Neurophysiology 115 (2004) SAH patients remains the early detection of clinically relevant DCI due to vasospasm. In these poor-grade patients, deterioration is often missed and silent infarction occurs in 21% (Kirmani et al., 2002). Silent infarction accounts for approximately one fourth of patients with DCI (Claassen et al., 2001; Shimoda et al., 2001) and is associated with poor clinical grade and poor outcome (Claassen et al., 2001; Kirmani et al., 2002). EEG monitoring provides continuous information about cerebral function and may allow early detection of focal ischemia (Labar et al., 1991; Vespa et al., 1997, 1999). EEG changes due to energy and ion pump failure occur when cerebral blood flow (CBF) falls below ml/100 g per min, at a time when therapeutic interventions might prevent permanent brain damage as infarction does not occur until CBF falls below 18 ml/100 g per min (Astrup et al., 1981; Baron, 2001; Sundt et al., 1981). Cerebral infarction may result in polymorphic delta, and attenuation of fast activity, including sleep spindles (Cohn et al., 1948; Niedermeyer, 1999; Nuwer et al., 1987). These EEG findings reflect abnormal CBF and disrupted metabolism (cerebral metabolic rate of oxygen) as seen with positron emission tomography (Nagata et al., 1989) and Xenon-CT-CBF studies (Ingvar et al., 1976; Tolonen and Sulg, 1981). The sensitivity of EEG to detect reversible ischemia was clinically first applied to intraoperative monitoring, specifically to monitor CBF during carotid endarterectomy (Sharbrough et al., 1973; Zampella et al., 1991). Digital EEG can be transformed into power spectra by fast Fourier transformation (FFT), compressing long periods of raw EEG into quantitative EEG (qeeg) parameters. These can be displayed as graphs and may reveal subtle changes in the EEG earlier than any other monitoring technique. These graphs can be used to monitor depth of sedation, intraoperative brain function (effects of sedation or brain injury), detect seizures, and possibly detect DCI due to vasospasm after SAH (Labar et al., 1991; Vespa et al., 1997). Two prior studies have related quantitative analysis of continuous EEG (ceeg) data with vasospasm in SAH patients. They found the trend analysis of total power (1 30 Hz) (Labar et al., 1991), and the variability of relative alpha (6 14/1 20 Hz) (Vespa et al., 1997) to be most predictive. Both these studies were almost entirely limited to non-comatose, good-grade patients (Hunt Hess grades 1 3). However, awake patients are easily monitored with serial clinical exams. In this context, the utility of ceeg may still exist, as one prior report demonstrated qeeg changes preceding the clinical deterioration by a mean of 2.9 days (Vespa et al., 1997). It is unclear if this technique may also be useful in poor-grade SAH patients, who often have severely abnormal EEG findings, which are more pronounced with increasing impairment of consciousness (Daly and Markand, 1990). The purpose of this study was to identify EEG parameters with a high sensitivity for DCI in poor-grade patients in order to alert the treating neurologist so that more definitive diagnostic tests and therapeutic interventions may be initiated. We also intended to determine the feasibility of this monitoring technique in these often medically unstable patients that require frequent in-hospital transportation and procedures. 2. Methods 2.1. Subjects All SAH patients admitted to the Neurological Intensive Care Unit (NICU) of Columbia-Presbyterian Medical Center between January 2000 and January 2002 were offered enrollment in the study. Eligibility criteria included: (1) Hunt Hess grade 4 or 5 (stupor or coma) on admission or within 24 h of NICU admission but prior to aneurysm treatment; (2) aneurysm treated with surgery or Guglielmi Detachable Coils; (3) alive on post-operative day 1; (4) EEG hook-up possible within 7 days of SAH onset; and (5) no documented symptomatic or asymptomatic angiographic vasospasm prior to the start of EEG monitoring. The study was approved by the hospital s Institutional Review Board. The diagnosis of SAH was established by the admission CT scan, or by xanthochromia of the cerebrospinal fluid if the CT was non-diagnostic. Patients with spontaneous nonaneurysmal SAH, or SAH due to trauma, arteriovenous malformation rupture, vasculitis, and other structural lesions were excluded Definition of DCI DCI was defined as otherwise unexplained (1) clinical deterioration (i.e. a new focal deficit, decrease in level of consciousness, or both), and/or (2) a new infarct on CT that was not visible on the admission or immediate postoperative scan (Claassen et al., 2001). Other potential causes of clinical deterioration, such as hydrocephalus, rebleeding, or seizures, were rigorously excluded. DCI was diagnosed by the treating study neurologist, and confirmed in a retrospective review of each subject s clinical course by two additional study physicians. In each patient with DCI, symptomatic territories were prospectively identified by the study neurologists, and clinical deficits, angiographic, and CT findings were recorded. Evidence of arterial spasm by transcranial Doppler (TCD) ultrasonography (O140 cm/s) or angiography was used to support the diagnosis, but was not mandatory Clinical and radiographic variables We recorded baseline demographic data (age, gender), location and size of the aneurysm, TCD findings, and mode of aneurysm treatment (clipping versus coiling). A study neurologist performed a neurological and general medical

3 J. Claassen et al. / Clinical Neurophysiology 115 (2004) evaluation on hospital admission. Neurological status on admission was assessed with the Hunt Hess scale (Hunt and Hess, 1968) and the Glasgow Coma Scale (GCS; Teasdale and Jennett, 1974). Admission and follow-up CT scans were independently evaluated by a study neurologist for the amount and location of subarachnoid (Hijdra et al., 1988) and intraventricular blood (Brouwers et al., 1993), the presence and degree of hydrocephalus (van Gijn et al., 1985), the presence of cerebral edema (Claassen et al., 2002), and cerebral infarction Clinical management External ventricular drainage (EVD) was placed in all patients with symptomatic hydrocephalus or intraventricular hemorrhage with reduced level of consciousness. All patients were followed with daily or every-other-day TCD, and received oral nimodipine. Clinical deterioration from DCI was treated with hypertensive, hypervolemic therapy (HHT) to maintain systolic blood pressure greater than 200 mmhg. All patients with clinical deterioration underwent CT or MRI scanning to identify the etiology of the changed clinical exam whenever clinically feasible. When clinical evidence of DCI persisted for more than 2 h despite HHT, cerebral angiography was used to identify vasospasm and balloon angioplasty was performed whenever feasible. Details about the clinical management of the study patients have been described previously (Claassen et al., 2001) Continuous EEG monitoring protocol EEG monitoring was started on post-operative day 2 after aneurysm surgery, and on day 1 after endovascular treatment. Monitoring was continued through SAH day 14 with digital EEG machines using a digital video bedside monitoring system (Nicolet-Bravo machines, Madison, WI; and XLTEK, Oakville, ON) with a 200 Hz digital sampling rate. The XLTEK acquisition system includes an antialiasing filter (4th order Butterworth) that depends on the sampling rate. At the rate of 200 Hz per channel that we utilize, the low pass filter is set at 53 Hz (the 3 db point on the frequency response curve), and the high pass filter is set at 0.1 Hz. Monitoring was stopped prior to day 14 if patients died, or at the request of the attending neurointensivist, neurosurgeon, or family members. Monitoring was continued beyond day 14 when clinically indicated (e.g. seizures). A board-certified electroencephalographer, who was blinded to the clinical course of the patient except to the diagnosis of poor-grade SAH, read the EEG recordings twice daily. For the purposes of this study, we recorded if the electroencephalographer identified seizures (electrographic seizures were defined as seizures with no detectable clinical correlate), non-convulsive status epilepticus, other repetitive epileptiform discharges, new onset slowing, or attenuation of fast activity in the EEG reading. Twenty-three disposable Ag/AgCl-Hydrospote-electrodes (Physiometrix, Inc.) were attached to the patient s head with collodion. A reduced number of electrodes (minimum 10) was used if operative wounds, EVDs, intracerebral pressure (ICP) cables, and other obstacles made this necessary. Patients were frequently (at least hourly) aroused or stimulated by physicians, ICU nurses, family members, and EEG technicians as part of routine clinical care. For the post hoc EEG analysis, these episodes were identified by notes entered into the EEG record at the patient s bedside or by correlating the EEG and simultaneous digital video recordings of the patients. Stimuli consisted of auditory alerting, clinical examinations, cleaning/feeding the patients, chest percussion, suctioning, and painful stimuli in the course of examination. Artifacts were identified by reviewing the recorded raw EEG EEG clips For each patient, 20 artifact-free, 1 min clips of continuous digital EEG (10 clips at baseline, and 10 at follow-up), directly following stimulation were analyzed. EEG clips obscured by artifact such as that related to movement, respirators, and chest physical therapy were disregarded. If artifact persisted beyond 10 s after stimulation episodes were disregarded. For baseline, 10 post-stimulation clips were taken on day 1 of the ceeg recording. For patients with DCI, care was taken to ensure that these clips were taken preceding the onset of clinical symptoms or CT infarction by at least 1 day. Patients were excluded from the analysis if there was no pre DCI clip available; we specifically excluded any patient with clinical evidence of possible DCI preceding the start of ceeg. For follow-up, 10 post-stimulation clips were taken on days 4 6 of the ceeg recording (preferably all on day 4). For patients with DCI, clips were obtained as soon as possible after the onset of clinical deterioration in symptomatic patients and after TCD or angiography first indicated vasospasm in asymptomatic patients who later developed infarction. All EEG clips from DCI patients were obtained prior to documented infarction on CT or MRI scanning. Patients were excluded if imaging was not available to prove that infarction had not been present at the time of collecting EEG clips qeeg Analysis Using fast Fourier transform, 12 qeeg parameters (Table 1) were calculated (Magic Marker Insight, Persyst Inc., Prescott, AZ) for 4 recording sites in each patient. These qeeg parameters included absolute power variables (the amplitude power in the respective frequency ranges is given as mv) (Labar et al., 1991), relative power expressed as a ratio (e.g. relative alpha power:alpha power [mv]/total power [mv]) (Labar et al., 1991; Vespa et al., 1997), coherence (a measure of cross-correlation in the frequency

4 2702 J. Claassen et al. / Clinical Neurophysiology 115 (2004) Table 1 Quantitative EEG parameters used for this study Parameter Abbreviation Analyzed frequency Absolute power Total power TP 1 30 Hz Alpha power AP 8 13 Hz Delta power DP 1 4 Hz Beta power BP Hz Fast power FP 8 30 Hz Relative power Alpha/total power RA 8 13/1 30 Hz Delta/total power RD 1 4/1 30 Hz Alpha/delta power AD 8 13/1 4 Hz Fast/delta power FD 8 30/1 4 Hz Coherence Alpha power CA 8 13 Hz left vs Hz right Delta power CD 1 4 Hz left vs. 1 4 Hz right Average frequency Total AF Average FFT frequency 1 30 Hz FFT, fast Fourier transformation. domain suggesting a relationship between two signals, such as one driving the other, e.g. coherence between left and right anterior recording sites for the 8 13 Hz frequency range) (Davey et al., 2000), and average frequency (mean frequency of the EEG signal) (Ingvar et al., 1976; Tolonen and Sulg, 1981). Recording sites included the left and right anterior circulation (F3 C3 and F4 C4, or if these were not available or artifact contaminated, Fp1 F3 and Fp2 F4, or F7 T3 and F8 T4), and the left and right posterior circulation (P3 O1 and P4 O2, or if these were not available, T5 O1 and T6 O2). In each patient, electrode selections were kept constant for all 20 EEG epochs; qeeg analysis was performed on the unfiltered raw EEG Statistical analysis Statistical analyses were performed with commercially available statistical software (SPSS version 9.0, SPSS Inc., Chicago, IL; and SAS, Cary, NC). qeeg values were averaged for individual patients over times 1 baseline and 2 follow-up separately, and a ratio of change was calculated (follow-up divided by time 1). For each individual recording site, we analyzed qeeg changes in patients with and without DCI using the Mann Whitney U test. In a second set of analyses, to incorporate 4 recording sites into one model, we used generalized estimating equations (GEE; with an exchangeable working correlation matrix accounting for site as an independent variable) to analyze the association between qeeg parameters and DCI (Liang et al., 1986). GEE models extend the general linear model to accommodate correlated data by quantifying and separating the influence of random effects on the statistical model. We calculated receiver operator curves (ROC) between each qeeg parameter and DCI (Hanley and McNeil, 1982). The area under the curve reflects the strength of association between qeeg parameters and DCI (values close to 1 or close to 0 indicating strong associations and values close to 0.5 representing the null hypothesis; values smaller than 0.5 are given as 1-x ). The area under the ROC curve is an estimate of the probability that for a randomly selected pair of normal and abnormal subjects, the clinician will correctly identify the subjects at risk for developing DCI based on the qeeg parameters. We then calculated sensitivity, specificity, positive predictive values (PPV), and negative predictive values (NPV) for changes of the EEG parameter with the closest correlation with DCI. Single and 2, 4, 6, 8, and 10 consecutive recordings with changes greater than 10, 20, and 50% were analyzed. For each patient, this analysis was based on qeeg values derived from 10 individual Fig. 1. Screening and enrollment.

5 Table 2 Clinical features of patients with DCI Age/sex Aneurysm location Raw EEG change on day of worsening Clinical exam findings New infarct detected on CT Angiographic vasospasm (moderate to severe) SAH day Slowing Attenuation SAH day Symptomatic territory TCD evidence of vasospasm (O140) SAH day Territory SAH day Territory SAH day Territory 56/F L 6 L 7 L 17 L 7 L NA No window 66/F L 8 Diffuse 7 L NA NA NA NA 4 L 57/F R PCoA 7 R R 7 R 7 R IC/hypothalamus 7 R, L VA/PICA, BA 6 R 42/F L PCoA 6 L front L front 3 L 7 L ACA/ PCA/thalamus/BG 59/F R ICA 4 L NA NA 6 L frontal lobe 80/F L VA/PICA NA NA 10 L cerebellum 30/F R PCoA 8 L and R ICA 11 Anterior choroidal 19/F L 10 R 10 R ACA/ 14 R ACA/ /ICA 37/F L ICA 8 Diffuse 7 R ACA/, L 11 R ACA, L ACA 4 L ACA/ /PCA/ ICA, R ACA, BA 8 R ACA/, L NA NA 4 L and R 7 L PCA, L NA No window supraclinoid 7 L ICA 7 LCR ACA/ 7 R ICA/ 2 R ACA/, L PCA 9 BA, R ICA/, L ACA/ 5 L, R ACA, anterior cerebral artery; BA, basilar artery; BG, basal ganglia; DCI, delayed cerebral ischemia; F, female; IC, internal capsule; ICA, internal carotid artery; L, left; M, male;, middle cerebral artery; NA, not available; PCA, posterior cerebral artery; PCoA, posterior communicating artery; PICA, posterior inferior cerebellar artery; R, right; TCD, transcranial Doppler sonography; VA, vertebral artery. J. Claassen et al. / Clinical Neurophysiology 115 (2004)

6 2704 J. Claassen et al. / Clinical Neurophysiology 115 (2004) exams and not the averaged value used above. Due to the large number of comparisons, significance was judged at P% Results 3.1. Study cohort Of 78 consecutive Hunt Hess grade 4 or 5, SAH patients admitted to the Columbia Neuro-ICU between 1/2000 and 1/2002, 48 were eligible for enrollment in the study protocol, and ceeg was performed in 34 (Fig. 1). Six eligible patients were not included because they were medically unstable, and 8 because of technical or logistical difficulties. Among the 30 non-eligible Hunt Hess grade 4 or 5 SAH patients, 17 died prior to surgery, 11 had documented vasospasm on their initial angiogram, and 2 were admitted more than 7 days after SAH. Mean age of the study patients was 55 years (range 18 82), and 77% (26 of 34) were women CEEG monitoring Monitoring was started between SAH days 2 and 7 (mean 4G1), and continued for 2 17 days (mean 6G3). Nonconvulsive status epilepticus was recorded in 6% (NZ2), electrographic seizures in 9% (NZ3), and periodic lateralized epileptiform discharges in 15% (NZ5) DCI patients Twenty-seven percent (9 of 34) of the patients studied developed DCI. Clinical symptoms and CT infarction attributable to DCI were seen in 6, infarction without detectable clinical symptoms occurred in 2, and clinical deterioration without CT infarction occurred in 1 patient (Table 2). Symptomatic territories involved the middle cerebral artery () distribution in 7 of 9 patients (3 of these involved only the, and in 4 the anterior cerebral artery (ACA) and posterior cerebral artery (PCA) territories were also affected). In the other two patients with DCI, single territories were affected (ACA and PCA). Angiography (O50% narrowing; NZ7) or TCD (O140 cm/s; NZ7) confirmed vasospasm in all patients. Raw, nonquantitative ceeg demonstrated new onset slowing or focal attenuation in 87% (7 of 9) of patients with DCI (Table 2). Among patients with DCI, 3 had an EVD placed for hydrocephalus prior to ceeg monitoring; no patient developed hydrocephalus during ceeg monitoring. Rebleeding did not occur in any patient during the monitoring period. Only 3 of these patients received sedation. In all 3 cases, propofol was used and rates were kept constant while patients were monitored qeeg parameters There were significant associations between DCI and changes in two relative qeeg parameters recorded from Table 3 Association between quantitative EEG parameters and delayed cerebral ischemia Relative change of qeeg Receiver operator curve DCI (NZ36) No DCI (NZ100) Z P Area under the curve 95% CI Absolute power Total power 0.12 K0.06 NS Alpha power 0 K0.03 NS Delta power 0.29 K Beta power NS Fast power 0.05 K0.02 NS Relative power Alpha/total K K Delta/total ! Alpha/delta K K4.1! Fast/delta K K3.5! Coherence Alpha power K0.12 K0.04 NS Delta power K0.21 K0.01 NS Average frequency Total K NS Relative changes of qeeg parameters are expressed as percent change of follow-up/baseline; Z-scores and P values are based on generalized estimating equations method assuming all territories as separate observations. The strength of association between quantitative EEG parameters and DCI is expressed by the area under the receiver operator curve (for details see Section 2). The 36 observations in the DCI group relate to averaged changes in 4 separate recording sites (right and left frontal, and right and left posterior), the 100 observation in the non-dci group relate to 25 patients, respectively. DCI, delayed cerebral ischemia; qeeg, quantitative EEG.

7 J. Claassen et al. / Clinical Neurophysiology 115 (2004) anterior left and right hemispheric sites: an increase in the delta/total power ratio (left 22% with DCI vs. K6% without, P!0.001; right 22% with DCI vs. 0% without, PZ0.001), and a decrease in the alpha/delta power ratio (left K23% with DCI vs. 6% without, P!0.001; right K24% with DCI vs. 3% without, PZ0.001). None of absolute power, coherence, or average frequency variables and none of the posterior recording sites reached significance. In the GEE models combining different recording sites (Table 3), most of the models analyzing the relationship between DCI and relative qeeg variables were significant. Stronger associations were observed for relative than for absolute power variables. When incorporating all territories or just the affected territories in the patients with DCI, the ADR demonstrated the strongest association with DCI (decrease of 24% in patients with vs. increase of 3% in patients without, P!0.0001). Figs. 2 4 illustrate an example of a patient where the change in the ADR precedes the clinical and radiological documentation of DCI. Strong associations were also seen for delta/total, fast/delta and alpha/total power. In a repeat analysis, comparing only qeeg parameters from clinically affected territories in patients with DCI to qeeg parameters from Fig. 2. Alpha/delta ratio (ADR) calculated every 15 min and Glasgow Coma Score (GCS), shown for days 6 8 of continuous EEG (ceeg) monitoring. Fifty seven-year-old woman admitted for acute subarachnoid hemorrhage (admission Hunt Hess grade 4) from a right posterior communicating aneurysm. Admission angiography did not show vasospasm. The aneurysm was clipped on SAH day 2. No infarcts were seen on post-operative CT (Fig. 3, day 2). Postoperatively she had a GCS of 14. ceeg monitoring was performed from SAH days 3 to 8. The ADR progressively decreased after day 6, particularly in the right anterior region (gray arrow), to settle into a steady trough level later that night, reflecting loss of fast frequencies and increased slowing over the right hemisphere in the raw ceeg (Fig. 4, EEG 2). On SAH day 6 flow velocities in the right were marginally elevated (144 cm/s), but the patient remained clinically stable with hypertensive, hypervolemic therapy (systolic blood pressure O180 mmhg). On day 7, the GCS dropped from 14 to 12 and a CT scan showed a right internal capsule and hypothalamic infarction (Fig. 3, Day 7). Angiography demonstrated severe distal right and left vertebral artery spasm; however, due to the marked tortuosity of the parent vessels and the location of vasospasm, a decision was made not to perform angioplasty, but to infuse verapamil and papaverine. This resulted in a marked, but transient increase of the right anterior and posterior alpha/delta ratios (gray shaded area). Later that day the patient further deteriorated clinically to a GCS of 7, with a new onset left hemiparesis, and died on SAH day 9 from widespread infarction due to vasospasm (Fig. 3, day 8; notice streak artifact from EEG electrodes).

8 2706 J. Claassen et al. / Clinical Neurophysiology 115 (2004) Fig. 3. CT scans obtained on SAH days 2, 7, and 8 (for details please refer to Fig. 2). patients that did not develop DCI, almost identical results were obtained (data not shown). ROC curves showed the largest areas under the curve for the alpha/delta and delta/ total power variables (both variables reached an area under the curve of 0.83; Table 3). Analyzing sensitivity, specificity, positive, and NPV for changes of the alpha/delta ratio (ADR) a number of constellations were found to be closely associated with DCI (Table 4). Clinically useful cut-offs include 6 consecutive post-stimulation recordings with a O10% decrease in ADR from baseline (sensitivity 100%, specificity 76%, PPV 60%, NPV 100%) and any single measurement with a O50% decrease of the ADR (sensitivity 89%, specificity 84%, PPV 67%, NPV 96%). 4. Discussion We studied the ability of qeeg parameters to detect DCI from vasospasm in poor-grade SAH patients. The poststimulation ADR (PSADR) detected patients with DCI with good sensitivity and specificity. In poor-grade stuporous or comatose patients, the earliest stages of DCI often go undetected. Alternative methods of continuous brain monitoring are limited by lack of sensitivity (e.g. clinical exam, CT, TCD), high false positive rates (e.g. TCD), and no standardized criteria for abnormality (e.g. CT-perfusion, DWI-MRI). Another major disadvantage of all of the above techniques is that they are usually only applied once or twice per day. EEG on the other hand can be recorded

9 J. Claassen et al. / Clinical Neurophysiology 115 (2004) Fig. 4. Sample of raw ceeg prior (SAH day 6) and during (SAH day 7) change in the alpha/delta ratio. Increase in delta and decrease in faster activity, more pronounced on the right (bottom 3 channels) in EEG 2 compared to EEG 1 (for details refer to Fig. 2). Table 4 Change in the alpha/delta ratio (ADR) for patients with and without DCI Anterior or posterior a Sensitivity Specificity PPV NPV P O10% change Any recording NS 2 consecutive recordings NS 4 consecutive recordings consecutive recordings ! consecutive recordings ! consecutive recordings !0.001 O20% change Any recording NS 2 consecutive recordings NS 4 consecutive recordings consecutive recordings consecutive recordings consecutive recordings O50% change Any recording ! consecutive recordings consecutive recordings consecutive recordings NS 8 consecutive recordings NS 10 consecutive recordings NS Sensitivity, specificity, positive predictive value, and negative predictive value are given in percent. Changes represent percent changes from baseline recordings. P values are based on c 2. PPV, positive predictive value; NPV, negative predictive value. a If consecutive recordings fulfilled criteria on either the left or right side.

10 2708 J. Claassen et al. / Clinical Neurophysiology 115 (2004) continuously, and qeeg parameters can be calculated at any time. The PSADR can be determined as often as the patient is stimulated and may supplement the clinical exam in these patients. We were able to perform ceeg monitoring in 71% (34 of 48) of eligible SAH patients. Although labor intensive, ceeg monitoring seems feasible in this population of critically ill patients. We restricted our inclusion criteria to poor-grade SAH patients since silent infarction occurs most frequently in this patient group (Claassen et al., 2001; Kirmani et al., 2002; Shimoda et al., 2001), and since neuromonitoring techniques are most useful for these patients. In our study, we were able to detect new onset focal slowing or focal attenuation of fast frequencies in the raw EEG in 78% of patients with DCI. Interpreting the raw EEG, however, is time-consuming, needs a considerable amount of expertise, and lacks concrete cut-offs for abnormality (i.e. a possible slight increase of slowing and new attenuation ) (Vespa et al., 1997). To make this technique applicable for the ICU setting, strict quantitative criteria categorizing patients EEGs into normal or abnormal are desirable. Within subject state changes, variations in physiological parameters (e.g. ICP, cerebral perfusion pressure), medications (i.e. sedatives), and artifacts heavily influence qeeg parameters (Adams et al., 1995). Attempts to minimize these effects include the use of relative instead of absolute qeeg parameters (Vespa et al., 1997), and limiting the analysis to artifact-free clips of EEG (Vespa et al., 1997). In addition to utilizing these techniques, we restricted our analysis to periods immediately following alerting stimuli to minimize physiological variability in the EEG, allowing greater sensitivity for changes related to DCI. To our knowledge, no prior study has analyzed post-stimulation qeeg or accounted for state changes in another fashion. We also carefully recorded physiological parameters (e.g. hypotensive episodes), clinical developments (e.g. hydrocephalus), and sedation in all patients during the EEG monitoring period. There is a large body of evidence showing an increase in delta and a decrease in alpha frequencies in patients during the acute phase of suffering a stroke (Niedermeyer, 1999; Nuwer et al., 1987). It is less clear if this holds true for reversible ischemia that has not yet progressed to an infarct, but diffuse attenuation of faster frequencies and increases in delta activity have been observed within seconds of intraoperative clamping of the carotid artery during carotid endarterectomy (Blume and Sharbrough, 1999; Sharbrough et al., 1973). It is not clear which qeeg parameters are most sensitive to detect ischemia in patients with severe diffuse brain injury leading to coma, such as poor-grade SAH. Patients with the ADR falling O10% below baseline in 6 consecutive recordings, or a single recording O50% below baseline, were likely to have developed DCI in our cohort. Prior studies had limited their analysis to awake patients (Labar et al., 1991; Vespa et al., 1997), and used angiographic vasospasm as the main outcome variable (Vespa et al., 1997). However, in stuporous or comatose patients, alpha frequencies might be invariable or may even be severely minimized with or without cerebral ischemia (Young, 2000). Labar et al. (1991) found the trend analysis of total power (1 30 Hz) to be predictive of DCI; however, because total power variables tend to be very state-sensitive, relative power variables are preferred for long-term monitoring (Vespa et al., 1997). Vespa et al. (1997) found that relative alpha variability calculated over an 8 12 h time window as [(peakktrough)/(peakctrough)] decreased in all patients with angiographic vasospasm (NZ19). Whether alpha variability can be calculated over shorter time periods to allow for the timely detection of ischemia in the ICU setting is unknown. The ADR reflects both attenuation of faster alpha frequencies and increase of slower delta frequencies, both of which occur with cerebral ischemia (Cohn et al., 1948; Nagata et al., 1989; Tolonen and Sulg, 1981; Vespa et al., 1997). In our analysis, single ADR changes greater than 50% were highly predictive of DCI. Confirming prior observations (Vespa et al., 1997), we found that DCI causes widespread EEG changes not limited to the vascular territory of vasospastic vessels. Importantly, qeeg parameters should never be interpreted without access to the raw EEG. A board-certified electroencephalographer, preferably one with extensive experience with critically ill patients, should be involved in the interpretation of EEG findings, as EEG can be quite complex and easily misinterpreted in these patients (Hirsch and Claassen, 2002). Conclusions concerning the etiology of qeeg changes should only be drawn with the knowledge of the clinical situation and review of the corresponding raw EEG. Our study has several limitations. This was a post hoc analysis and the EEG clips we analyzed were carefully selected to avoid contamination from artifact. Future studies will have to determine the best strategy to incorporate ceeg monitoring into clinical practice. Although the actual calculation of ADR values could easily be automated, the recording and selection of an artifact-free epoch of raw EEG to be analyzed in these critically ill patients requires skilled EEG technicians and experienced electroencephalographers. Since 24 h coverage by electroencephalographers is unrealistic for most ICUs, some of the steps in obtaining ADR values might best be delegated. Patients are routinely stimulated by ICU staff, physicians, and family members in most ICUs; the ICU staff could be trained to recognize artifact-free EEG after stimulation and generate PSADR values at this time by just pushing a button. These could then be validated by trained electroencephalographers with a web-based EEG monitoring systems (Scheuer, 2002) if significant changes are detected. The patients who were included in this study were carefully selected based on predetermined inclusion criteria. Therefore, our findings may only apply to this carefully selected group of severely ill ICU patients with SAH.

11 J. Claassen et al. / Clinical Neurophysiology 115 (2004) If confirmed in an independent dataset and a larger sample size, our results may have promising clinical implications. By rigorously limiting our analysis to EEG clips that were taken at time points that preceded CT documented infarction, we feel confident that the observed EEG changes reflect the early stages of ischemia. Interventions such as HHT or angioplasty initiated as early as possible after qeeg parameters first change may have the potential to prevent infarction. To determine the clinical impact of PSADR monitoring, future studies will have to assess if interventions triggered by pre-specified criteria can result in the prevention of infarction or improve outcome. These studies should also compare clinical exam, raw EEG interpretation, and TCD with qeeg parameters to determine which test shows evidence of DCI first and most accurately, or if these tests are complementary. Future studies are needed to systematically study the effects of sedation, increased ICP, reduced cerebral perfusion pressure, medications, metabolic changes, hydrocephalus, and rebleeding on the ADR. Acknowledgements We would like to thank the ICU fellows, residents, and nurses of New York Presbyterian Hospital at Columbia University for their overall support of this project. This paper was supported by a Grant-in-Aid to Stephan A. Mayer, MD from the American Heart Association (# N). References Adams DC, Heyer EJ, Emerson RG, Moeller JR, Spotnitz HM, Smith DH, Delphin E, Turner C. The reliability of quantitative electroencephalography as an indicator of cerebral ischemia. Anesth Analg 1995;81: Astrup J, Siesjo BK, Symon L. Thresholds in cerebral ischemia the ischemic penumbra. Stroke 1981;12: Baron JC. Perfusion thresholds in human cerebral ischemia: historical perspective and therapeutic implications. Cerebrovasc Dis 2001; 11(1):2 8. Blume WT, Sharbrough FW. EEG monitoring during carotid endarterectomy and open heart surgery. In: Niedermeyer E, Lopes da Silva F, editors. Electroencephalography. Baltimore, MD: Urban & Schwarzenberg; 1999, p Brouwers PJ, Dippel DW, Vermeulen M, Lindsay KW, Hasan D, van Gijn J. Amount of blood on computed tomography as an independent predictor after aneurysm rupture. Stroke 1993;24: Charpentier C, Audibert G, Guillemin F, Civit T, Ducrocq X, Bracard S, Hepner H, Picard L, Laxenaire MC. Multivariate analysis of predictors of cerebral vasospasm occurrence after aneurysmal subarachnoid hemorrhage. Stroke 1999;30: Claassen J, Bernardini GL, Kreiter KT, Bates JE, Du YE, Copeland D, Connolly ES, Mayer SA. Effect of cisternal and ventricular blood on risk of delayed cerebral ischemia after subarachnoid hemorrhage: the Fisher Scale revisited. Stroke 2001;32: Claassen J, Carhuapoma JR, Kreiter KT, Du EY, Connolly ES, Mayer SA. Global cerebral edema in patients with subarachnoid hemorrhage: frequency, predictors, and impact on outcome. Stroke 2002;33: Cohn HR, Raines RG, Mulder DW, Neumann MH. Cerebral vascular lesions: electroencephalographic and neuropathologic correlations. Arch Neurol 1948;60: Daly DD, Markand ON. Focal brain lesions. In: Pedley TA, editor. Current practice of clinical electroencephalography, 2nd ed.. New York: Raven Press; 1990, p Davey MP, Victor JD, Schiff ND, Davey MP, Victor JD, Schiff ND. Power spectra and coherence in the EEG of a vegetative patient with severe asymmetric brain damage. Clin Neurophysiol 2000;111: Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143: Hijdra A, van Gijn J, Nagelkerke NJ, Vermeulen M, van Crevel H. Prediction of delayed cerebral ischemia, rebleeding, and outcome after aneurysmal subarachnoid hemorrhage. Stroke 1988;19: Hirsch LJ, Claassen J. The current state of treatment of status epilepticus. Curr Neurol Neurosci Rep 2002;2: Hop JW, Rinkel GJ, Algra A, van Gijn J. Initial loss of consciousness and risk of delayed cerebral ischemia after aneurysmal subarachnoid hemorrhage. Stroke 1999;30: Hunt WE, Hess RM. Surgical risk as related to time of intervention in the repair of intracranial aneurysms. J Neurosurg 1968;28: Ingvar DH, Sjolund B, Ardo A. Correlation between dominant EEG frequency, cerebral oxygen uptake and blood flow. Electroencephalogr Clin Neurophysiol 1976;41: Kassell NF, Peerless SJ, Durward QJ, Beck DW, Drake CG, Adams HP. Treatment of ischemic deficits from vasospasm with intravascular volume expansion and induced arterial hypertension. Neurosurgery 1982;11: Kirmani JF, Qureshi AI, Hanel RA, Siddiqui AM, Safdar A, Yahia AM, et al. Silent cerebral infarction in poor-grade patients with subarachnoid hemorrhage. Neurology 2002;7(3):159 [Abstract]. Labar DR, Fisch BJ, Pedley TA, Fink ME, Solomon RA. Quantitative EEG monitoring for patients with subarachnoid hemorrhage. Electroencephalogr Clin Neurophysiol 1991;5: Liang KY, Zeger SL, Scott L. Longitudinal data analysis for discrete and continuous outcomes. Biometrics 1986;42: Mayberg MR. Cerebral vasospasm. Neurosurg Clin N Am 1998;3: Murayama Y, Malisch T, Guglielmi G, Mawad ME, Vinuela F, Duckwiler GR, Gobin YP, Klucznick RP, Martin NA, Frazee J. Incidence of cerebral vasospasm after endovascular treatment of acutely ruptured aneurysms: report on 69 cases. J Neurosurg 1997;87: Nagata K, Tagawa K, Hiroi S, Shishido F, Uemura K. Electroencephalographic correlates of blood flow and oxygen metabolism provided by positron emission tomography in patients with cerebral infarction. Electroencephalogr Clin Neurophysiol 1989;72: Niedermeyer E. Cerebrovascular disorders and EEG. In: Niedermeyer E, Lopes da Silva F, editors. Electroencephalography. Baltimore, MD: Urban & Schwarzenberg; 1999, p Nuwer MR, Jordan SE, Ahn SS. Evaluation of stroke using EEG frequency analysis and topographic mapping. Neurology 1987;37: Qureshi AI, Sung GY, Razumovsky AY, Lane K, Straw RN, Ulatowski JA. Early identification of patients at risk for symptomatic vasospasm after aneurysmal subarachnoid hemorrhage. Crit Care Med 2000;28: Scheuer ML. Portable remote wireless EEG review using a cellular CDMA network. Epilepsia 2002;42:27 8 [Abstract]. Sharbrough FW, Messick Jr JM, Sundt Jr TM. Correlation of continuous electroencephalograms with cerebral blood flow measurements during carotid endarterectomy. Stroke 1973;4: Shimoda M, Takeuchi M, Tominaga J, Oda S, Kumasaka A, Tsugane R. Asymptomatic versus symptomatic infarcts from vasospasm in patients with subarachnoid hemorrhage: serial magnetic resonance imaging. Neurosurgery 2001;49:

12 2710 J. Claassen et al. / Clinical Neurophysiology 115 (2004) Sundt Jr TM, Sharbrough FW, Piepgras DG, Kearns TP, Messick Jr JM, O Fallon WM. Correlation of cerebral blood flow and electroencephalographic changes during carotid endarterectomy: with results of surgery and hemodynamics of cerebral ischemia. Mayo Clin Proc 1981; 56: Teasdale G, Jennett B. Assessment of coma and impaired consciousness: a practical scale. Lancet 1974;2:81 3. Tolonen U, Sulg IA. Comparison of quantitative EEG parameters from four different analysis techniques in evaluation of relationships between EEG and CBF in brain infarction. Electroencephalogr Clin Neurophysiol 1981;51: van Gijn J, Hijdra A, Wijdicks EF, Vermeulen M, van Crevel H. Acute hydrocephalus after aneurysmal subarachnoid hemorrhage. J Neurosurg 1985;63: Vespa PM, Nuwer MR, Juhasz C, Alexander M, Nenov V, Martin N, Becker DP. Early detection of vasospasm after acute subarachnoid hemorrhage using continuous EEG ICU monitoring. Electroencephalogr Clin Neurophysiol 1997;6: Vespa PM, Nenov V, Nuwer MR. Continuous EEG monitoring in the intensive care unit: early findings and clinical efficacy. J Clin Neurophysiol 1999;16:1 13. Weir B, Grace M, Hansen J, Rothberg C. Time course of vasospasm in man. J Neurosurg 1978;48: Young GB. The EEG in coma. J Clin Neurophysiol. 2000;17: Zampella E, Morawetz RB, McDowell HA, Zeiger HE, Varner PD, McKay RD, Halsey Jr JH. The importance of cerebral ischemia during carotid endarterectomy. Neurosurgery 1991;29: