High-dose barbiturate therapy is considered in traumatic brain-injured (TBI) patients with high intracranial pressure (ICP) refractory to other treatments.1 Their main ICP-decreasing action is a decrease in metabolic demand, which allows a decrease in cerebral blood flow and cerebral blood volume. However, barbiturates induce serious hemodynamic disturbances2 and can lead to immunosuppression or leucopenia.3 Thus, barbiturate overdose must be avoided,4 and monitoring of barbiturate efficiency is necessary to establish the lowest useful dose in real time. Because serum concentrations are not well correlated with cerebral activity,4 the reference method until now has been electroencephalography (EEG). Barbiturates induce a specific EEG pattern, i.e., burst suppression5 interspersed between electrical silences (suppression) and bursting activity. In experimental studies, cerebral metabolic depression reached a plateau when barbiturate infusion produced EEG burst suppression. Administration of additional barbiturates significantly decreased systemic arterial blood pressure and cardiac output, but produced no further decrement in the cerebral metabolic rate of oxygen or the cerebral blood flow.6,7 The goal of sedation is to obtain a “burst-suppression” pattern and to avoid a continuous electrical silence signal, the hallmark of overdose.
However, multiple-lead EEG monitoring needs specific cumbersome equipment and may be complex to interpret, requiring a trained neurologist. Portable EEG-based monitors like the Bispectral Index (BIS) BIS-XP™ monitor (Medical Aspect System, Newtown, MA) are now used in critical care. They digitize EEG signals obtained from surface electrodes placed on the forehead using an algorithm based on healthy brain EEG data and provide a BIS (a number from 0 to 100, 100 for unsedated patients and 0 for those without any EEG activity) and a suppression ratio, i.e., the percentage of the previous minute analyzed by the monitor as electrical silence.8–11
We hypothesized that a BIS range could predict a specific burst-suppression pattern. We had first to confirm that the relationship between BIS and suppression ratios is as strong in TBI patients treated with barbiturates as in healthy patients sedated by propofol.12 In 12 TBI patients treated by barbiturates with a suppression pattern, Riker et al.13 found a correlation between BIS number and the number of bursts/min. However, as a significant statistical correlation does not help the clinician to adapt the level of barbiturate infusion, the main objective of this study was to determine whether a BIS range could help in predicting a burst-suppression pattern in these patients.
After IRB approval, TBI patients treated with barbiturates in our 24-bed intensive care unit (ICU) were prospectively included from May to October 2003. Because this was an observational study in which barbiturate therapy was still guided by the raw EEG, and because no additional blood samples were taken, the IRB waived the requirement for informed consent.
The sedation protocol consisted of midazolam (0.1 mg · kg−1 · h−1) and fentanyl (7 μg · kg−1 · h−1), adjusted on sedation (Ramsay score 6 in case of intracranial hypertension) and ICP levels. No patient needed myorelaxants. Focal mass lesions were evacuated. Increases in ICP >20 mm Hg were treated with a stair-step approach: increase in sedation, cerebrospinal fluid drainage if available, and osmotic drugs. The goal was to maintain both ICP <20 mm Hg (or <15 mm Hg in case of large craniectomy) and cerebral perfusion pressure around 70 mm Hg, the mean arterial blood pressure being optimized by vascular filling or the use of norepinephrine. Barbiturates were started according to the criteria of Eisenberg et al.14 (ICP around 25 mm Hg for 30 min, around 30 mm Hg for 15 min, or as high as 40 mm Hg for 1 min). After an initial dose of 5–7 mg/kg of thiopental, the infusion was adjusted from 2.3 mg · kg−1 · h−1 guided by the level of ICP. If ICP remained uncontrolled, the managing physician increased the infusion rate or added a bolus of barbiturate (3–5 mg/kg). When a barbiturate infusion was prescribed, the midazolam infusion was stopped. At the time of this study, a daily 8-lead EEG was our usual way to check that barbiturate infusion was well adjusted to obtain a burst-suppression pattern. The EEG was analyzed by a neurophysiologist. Our suppression goal was 2–5 bursts/min. The serum thiopental concentration was also measured once a day.
For the purpose of the study, the raw daily EEG was recorded on a computer (Deltamed, Paris, France) for at least 1 h by 8 scalp needle electrodes positioned after skin disinfection according to the international 10–20 system, referenced to ground over a frontal needle. Standard positions were Fp1, Fp2, T3, T4, C3, C4, O1, and O2. Impedances were kept less than 5 kΩ. The EEG was filtered at 70 Hz. The time constant was 0.3 s and sampling frequency was 256 Hz. The EEG signal was digitized and stored to be analyzed off-line by a senior neurophysiologist unaware of BIS data. Every 5 min during recording, a period of 1 min was analyzed during which the duration (seconds) of the suppression periods, i.e., the sum of the EEG silences, and the number of bursts/min were reviewed. Suppression ratio derived from EEG (SREEG) was counted as the percentage of silence during the previous minute. Thus, each EEG resulted in 12 pairs of data points (number of bursts/min and SREEG). When present, the burst-suppression pattern may be observed in all areas of the brain. As a general rule, frontal leads were chosen to measure bursts because waveform amplitudes are higher in these leads. In this study, the fact that frontal leads were also used by BIS-XP™ was another reason for this choice. However, the senior neurophysiologist was free to use other leads if he thought there might be some interference in frontal leads.
The BIS was measured with a BIS-XP™ monitor (Aspect Medical System, Newton, MA; software version 3.11). A BIS-Quatro electrode was placed on the left side of the forehead. BIS and suppression ratio, i.e., the percentage of the previous minute analyzed by the BIS-XP™ as electrical silence (suppression ratio from BIS [SRBIS]), was monitored continuously and data were acquired on a computer. EEG and BIS-XP™ clocks were synchronized. At the end of each minute studied on the EEG graph, BIS and SRBIS values were recorded and the data were collected on an Excel spreadsheet. Medical care was carefully noted because of the risk of artifacts. If they occurred, the corresponding data were disregarded. Electromyographic activity (EMG) was monitored continuously after the tenth examination by way of the BIS-XP™. We did not use additional filters. The spectral power in the 70–110 Hz frequency band is considered to represent EMG and is displayed in decibels (dB).
Data obtained from the BIS monitor were kept for analysis only if the signal quality index was more than 95%. Normally distributed data are presented as mean ± sd or 95% confidence interval (95% CI). In case of ordinal values or nonnormal distribution, data are shown as median with their interquartile distribution (25th–75th interquartile [IQ]) or their range. Relationships between data obtained by the two methods were first assessed graphically. Agreement between SRBIS and SREEG was analyzed by the intraclass correlation coefficient estimated by a linear mixed effect regression model considering the correlation between patients and between examinations.15 The same kind of model was used to analyze the association between BIS and SREEG.
All data were studied to estimate the best range of BIS to predict optimal sedation. Optimal sedation was defined in our study as an EEG with a pattern of 2–5 bursts/min. This range was applied to each examination to determine the rate of accuracy, defined as the number of well-classified data (optimal sedation with BIS in the defined range [true positive data] and nonoptimal sedation with BIS out of the defined range [true-negative data]) divided by the total number of data for each examination. A sample size was not calculated for this observational study. At the end of the 11th patient’s monitoring, we became fully aware that one EEG once a day was not enough to avoid barbiturate overdose, so we changed our management strategy.
Eleven TBI patients were included. Eight were men (73%) and the median age was 30 yrs (range, 14–53). Cause of trauma was a traffic accident in eight cases and a fall in three cases. Intracerebral lesions, associated injuries, and neurosurgical management are shown in Table 1. All patients needed norepinephrine but no myorelaxants. Pneumopathy was the most frequent complication. Two patients had acute respiratory distress syndrome but only one of these began during barbiturate infusion. All patients survived.
Daily average serum barbiturate values varied widely, median 40 μg/mL (range, 3–120) depending on difficulties in reducing ICP. EMG activity was low: 26.6 ± 1.9 dB (Table 2).
Thirty EEGs were performed (1–4 per patient) generating 347 measurements. The depth of sedation fluctuated throughout the examination in 14 of 30 EEGs. There was a strong statistical agreement between SRBIS and SREEG (Fig. 1) with an intraclass correlation coefficient of 0.94 (95% CI, 0.90–0.96). The median difference between SRBIS and SREEG in the 347 pairs of points was −1 (IQ: [−6] − [+2]). However, when data were studied per EEG (Fig. 2), six examinations of 30 showed a difference ≥10 in half of their 12 data. These discrepancies did not depend on sedation state. The shift was positive in four examinations and negative in two.
A scatter graph of BIS number versus SRBIS showed that BIS number depended especially on the percentage of suppression assessed by the BIS-XP™ when SRBIS was more than 40% (Fig. 3). The relationship between BIS and SREEG was significant (regression slope: −0.29; P < 0.0001) (Fig. 4).
BIS ranged from 1 to 45 (median: 7 [IQ: 4–13]) in the 106 measurements with optimal sedation (2–5 bursts/min), from 4 to 50 (median 24 [IQ: 20–32]) in the 122 measurements if number of bursts/min was more than 5, and from 0 to 11 (median: 1 [IQ: 0–3]) in the 119 measurements with overdose (0–1 bursts/min) (Table 3). The accuracy of a BIS <6 to predict overdose was 85%. The accuracy of a BIS >15 to predict insufficient suppression was 93%. The accuracy of a BIS between 6 and 15 to predict a correct state of suppression was 81%. The global accuracy was more than 80% in 18/30 examinations (Table 3). In two other examinations, the patient’s level of sedation decreased and BIS and number of bursts/min increased, but at different rates over time. In the other cases, the lower level of accuracy may be explained in most cases by borderline measurements (number of bursts/min around 1 or 5 and BIS around 4 or 16).
All our data were read again retrospectively to determine whether the proposed rule (to decrease infusion rate when BIS is <6 and to increase infusion rate when BIS is >15) would have been efficient. To decrease the infusion rate owing to consecutive measures of BIS <6 would have been a good choice 11 times and a mistake in 1 case (during examination 17) where the number of bursts (5–7/min) was in the high borderline of suppression. To maintain the rate of barbiturate infusion when BIS was between 5 and 16 would have been a suitable decision five times and a mistake once (examination 9), where the patient was overdosed throughout the examination. To increase the infusion rate in cases of consecutive values of BIS >15 would have been a good choice 12 times and a mistake once (examination 24), with an optimal number of bursts when BIS was around 40.
The present study shows that the BIS number may be used to help clinicians to assess the level of EEG suppression in TBI patients with high ICP treated with barbiturates. Continuous EEG monitoring is not available in all ICUs. Moreover, the ability of caregivers in ICU to interpret EEG is low when confronted with epileptiform discharges,16 and it may be difficult for caregivers to differentiate artifacts from bursts. Thus, a continuous real-time assessment method would be an improvement in monitoring the depth of barbiturate coma. Riker et al.13 demonstrated a correlation between number of bursts/min and BIS in 12 TBI patients in 45 data sets with a burst-suppression pattern. However, there were no data on BIS when number of bursts/min was outside this range. In our study in 11 patients, we considered BIS and studied the suppression pattern in each BIS range. By studying examinations one by one, a BIS number <6 seemed to be a good indication of barbiturate infusion overdose. On the other hand, a BIS >15 could be a good indication of insufficient suppression. Using the target of suppressing 2–5 bursts, we found that BIS values ranged from 1 to 45 when sedation was optimal (median, 7). In the study of Riker et al.,13 the mean BIS was 15 (95% CI, 10–20), but his goal was 3–5 bursts/min. There is an absence of consensus regarding the appropriate number of bursts/min as a “gold standard” end point for the therapy: 3–5,13 1–3,17 1–5,4 2–6,18 and 4–6.3 There is also no consensus about a strict definition of a burst. In this study, a burst was defined as a period of electric activity interspersed with at least 1 s of suppression. The definition of a burst might be different if one considers the length of electric activity. For example, Riker et al.13 stated that three bursts lasting 5 s may represent a higher degree of suppression than three bursts lasting 15 s. The BIS range we propose may be considered as being indicative of a burst-suppression state. It should be remembered that BIS and number of bursts/min are quantitative criteria. To determine a cutoff, we transformed quantitative data into qualitative assessment. It is arbitrary to decide that 1 burst/min indicates an overdose and that 2 bursts/min is a suitable goal for brain protection. Borderline measurements accounted for most of the lack of accuracy we found in our analysis. A range should be established for each patient, considering their hemodynamic tolerance to high barbiturate levels and the need to increase barbiturate administration to control persistent high ICP.
The large margin of error we sometimes observed deserves some attention. Data from our 11 patients showed a good concordance between SRBIS and SREEG. with a statistical method considering variability among patients and among examinations in the same patient. Despite good statistical agreement, the difference between SRBIS and SREEG might be significant for clinicians in some examinations. This implies that BIS values obtained from the suppression state shown on the monitor may be discrepant from the suppression ratio calculated from the raw EEG. The localization of the brain injury may play a role in these differences because of unilateral lesion13 or brainstem involvement.19 However, discrepancies between SRBIS and SREEG may differ from 1 day to another in the same patient (Fig. 2). Muscular activity may disturb BIS calculation20 but our EMG data had a pattern similar to that expected in patients treated by neuromuscular blocking drugs.21 Moreover, interference from electrical devices may falsely increase BIS. Depending on the source of artifact, EEG tracing may be disturbed (endoscopic shaver22,23) or not, as was the case with forced-air warming blankets.24 Signal acquisition may also be disturbed by changes in skin-electrode contact.25 This signal quality index given by the BIS monitor is supposed to be able to detect most artifacts. As a precaution, the measurements retained for analysis in this study had a signal quality index >95%. However, some BIS values remained perturbed by unexpected artifacts. In such cases, trained neurophysiologists may be more efficient. Besides artifacts, methodological problems may also account for some of the discrepancies occurring here. First, the neurophysiologist did not consider periods of silence lasting less than 1 s, whereas the BIS-XP™ uses periods longer than 0.5 s.26 This was a pitfall in our methodology but it was detected too late. Second, in some examinations there was a time lapse between the change in BIS and the change in EEG. Our smoothing window was 30 s because we thought that the suppression ratio had to be calculated over a long period because of its nonstationary nature. However, this implies that there was a delay before the BIS considered a change in sedation level.
Owing to these unpredictable discrepancies between BIS and EEG, the correspondence between BIS and burst-suppression pattern has to be checked regularly by a trained observer using the raw EEG. For example, in our retrospective study of all examinations, to estimate the rule we propose, we found consecutive BIS numbers in the range of 6–15 in one examination, despite an overdose. In this case, a careful observation of the EEG pattern on the monitor would have shown a flat line and would have prompted us to decrease the infusion rate. Currently, in our unit, we monitor the BIS number all day long. However, a clinician checks several times a day that the BIS number is coherent with the analogical EEG signal provided by the BIS-XP™ monitor screen. In some cases, where the clinician has a doubt about the concordance of BIS number with the EEG pattern, a full EEG is performed and read by a neurophysiologist who looks at the full brain pattern.
In conclusion, suppression rates obtained by a BIS-XP™ (SRBIS) are well correlated with those obtained by an 8-scalp needle electrode EEG read by a trained neurophysiologist. BIS is closely related to SRBIS in TBI patients with an SRBIS more than 50. The continuous data obtained from the monitor may be useful for helping clinicians or nurses to adjust barbiturate infusion.27 We suggest a decrease in the infusion rate if the BIS is <6 and an increase for BIS >15. However, interference between BIS monitoring and other electric or electromagnetic devices in an ICU is still not fully understood and the concordance between BIS and EEG tracing must be checked regularly.
The authors thank Pr Pierre-Marie Preux (Department of Biostatistics, Dupuytren University Hospital, Limoges, France) for expert statistical evaluation of the manuscript, and Ray Cooke (University of Bordeaux 2) for revising the English.
1. The Brain Trauma Foundation. The American Association of Neurological Surgeons. The Joint Section on Neurotrauma and Critical Care. Use of barbiturates in the control of intracranial hypertension. J Neurotrauma 2000;17:527–30
2. Todd MM, Drummond JC, U HS. The hemodynamic consequences of high-dose thiopental anesthesia. Anesth Analg 1985;64:681–7
3. Stover JF, Stocker R. Barbiturate coma may promote reversible bone marrow suppression in patients with severe isolated traumatic brain injury. Eur J Clin Pharmacol 1998;54: 529–34
4. Winer JW, Rosenwasser RH, Jimenez F. Electroencephalographic activity and serum and cerebrospinal fluid pentobarbital levels in determining the therapeutic end point during barbiturate coma. Neurosurgery 1991;29:739–41
5. Kiersey D, Bickford R, Faulconer A Jr. Electro-encephalographic patterns produced by thiopental sodium during surgical operations: description and classification. Br J Anaesth 1951;23:141–52
6. Kassell NF, Hitchon PW, Gerk MK, Sokoll MD, Hill TR. Alterations in cerebral blood flow, oxygen metabolism, and electrical activity produced by high dose sodium thiopental. Neurosurgery 1980;7:598–603
7. Turcant A, Delhumeau A, Premel-Cabic A, Granry JC, Cottineau C, Six P, Allain P. Thiopental pharmacokinetics under conditions of long-term infusion. Anesthesiology 1985;63:50–4
8. Rosow C, Manberg PJ. Bispectral index monitoring. Anesthesiol Clin North America 2001;19:947–66
9. Glass PS, Bloom M, Kearse L, Rosow C, Sebel P, Manberg P. Bispectral analysis measures sedation and memory effects of propofol, midazolam, isoflurane, and alfentanil in healthy volunteers. Anesthesiology 1997;86:836–47
10. Kearse LA Jr, Rosow C, Zaslavsky A, Connors P, Dershwitz M, Denman W. Bispectral analysis of the electroencephalogram predicts conscious processing of information during propofol sedation and hypnosis. Anesthesiology 1998;88:25–34
11. Katoh T, Suzuki A, Ikeda K. Electroencephalographic derivatives as a tool for predicting the depth of sedation and anesthesia induced by sevoflurane. Anesthesiology 1998;88:642–50
12. Bruhn J, Bouillon TW, Shafer SL. Bispectral index (BIS) and burst suppression: revealing a part of the BIS algorithm. J Clin Monit Comput 2000;16:593–6
13. Riker RR, Fraser GL, Wilkins ML. Comparing the bispectral index and suppression ratio with burst suppression of the electroencephalogram during pentobarbital infusions in adult intensive care patients. Pharmacotherapy 2003;23:1087–93
14. Eisenberg HM, Frankowski RF, Contant CF, Marshall LF, Walker MD. High-dose barbiturate control of elevated intracranial pressure in patients with severe head injury. J Neurosurg 1988;69:15–23
15. Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics 1982;38:963–74
16. Leira EC, Bertrand ME, Hogan RE, Cruz-Flores S, Wyrwich KW, Albaker OJ, Holzemer EM. Continuous or emergent EEG: can bedside caregivers recognize epileptiform discharges? Intensive Care Med 2004;30:207–12
17. Todd MM, Drummond JC, Hoi SU. Hemodynamic effects of high dose pentobarbital: studies in elective neurosurgical patients. Neurosurgery 1987;20:559–63
18. Cordato DJ, Herkes GK, Mather LE, Gross AS, Finfer S, Morgan MK. Prolonged thiopentone infusion for neurosurgical emergencies: usefulness of therapeutic drug monitoring. Anaesth Intensive Care 2001;29:339–48
19. Lazar LM, Milrod LM, Solomon GE, Labar DR. Asynchronous pentobarbital-induced burst suppression with corpus callosum hemorrhage. Clin Neurophysiol 1999;110:1036–40
20. Riker RR, Fraser GL, Simmons LE, Wilkins ML. Validating the Sedation-Agitation Scale with the Bispectral Index and Visual Analog Scale in adult ICU patients after cardiac surgery. Intensive Care Med 2001;27:853–8
21. Vivien B, Di Maria S, Ouattara A, Langeron O, Coriat P, Riou B. Overestimation of bispectral index in sedated intensive care unit patients revealed by administration of muscle relaxant. Anesthesiology 2003;99:9–17
22. Hemmerling TM, Migneault B. Falsely increased bispectral index during endoscopic shoulder surgery attributed to interferences with the endoscopic shaver device. Anesth Analg 2002;95:1678–9
23. Hemmerling TM, Fortier JD. Falsely increased bispectral index values in a series of patients undergoing cardiac surgery using forced-air-warming therapy of the head. Anesth Analg 2002;95: 322–3
24. Dahaba AA. Different conditions that could result in the bispectral index indicating an incorrect hypnotic state. Anesth Analg 2005;101:765–73
25. Walsh TS, Ramsay P, Kinnunen R. Monitoring sedation in the intensive care unit: can “black boxes” help us? Intensive Care Med 2004;30:1511–3
26. Rampil IJ. A primer for EEG signal processing in anesthesia. Anesthesiology 1998;89:980–1002
© 2008 International Anesthesia Research Society
27. Bader M, Arbour R, Palmer S. Refractory increased intracranial pressure in severe traumatic brain injury: barbiturate coma and bispectral index monitoring. AACN Clin Issues 2005;16:526–41