Electroencephalography and clinical Neurophysiology
90 (1994) 194-200 © 1994 Elsevier Science Ireland Ltd. 0013-4694/94/ 07.00 EEG93076
ispectral analysis of the electroencephalogram during induction of anesthesia may predict hemodynamic responses to laryngoscopy and intubation
L.A. Kearse Jr. a . p. Manberg c F. DeBros b N. Chamoun c and V. Sinai c
Departments of a Anesthesia and Neurology, and b Anesthesia, Massachusetts General Hospital, Harcard Medical School, 32 Fruit Street, Boston, MA 02114 USA), and C Aspect Medical System, Framingham, MA USA)
(Accepted for publication: 26 October 1993)
Summary The use of electroencephalography as a measure of adequacy of anesthesia has achieved limited success. Our purpose was to determine whether the non-linear properties of the electroencephalogram (EEG) as defined by the bispectral index was a better predictor of autonomic responses to endotracheal intubation during opioid-based anesthesia than the linear statistical properties of the EEG formulated by power spectral analysis. Thirty-nine adults scheduled for elective non-cranial surgery had a continuous EEG recorded during induction of anesthesia and endotracheal intubation. Anesthesia consisted of thiopental and nitrous oxide in oxygen, followed by 1 of 5 randomized opioid dose regimens. The EEG was continuously recorded and blood pressure was measured every minute. All electroencephalographic parameters were derived for the 3 min before and after intubation and were compared to the blood pressure and heart rate responses. Responders were defined by 2 analyses: patients who had a 20% or greater increase (1) in blood pressure or (2) in heart rate to laryngoscopy. Responders and non-responders were compared using Student's unpaired t test, and differences due to dose regimens were examined with logistic regression. Based on the criterion for blood pressure change, there were 27 responders and 12 non-responders. Heart rate changes did not differentiate between the two groups. There was a significant difference between response groups as measured by the bispectral index which distinguished responders from non-responders independently of the amount of drug given. None of the variables of power spectral analysis accurately distinguished responder from non-responder. We conclude that the bispectral index of the EEG may accurately predict blood pressure responses in patients undergoing endotracheal intubation after induction of general anesthesia. Key words: Bispectral analysis; Power spectrum; Electroencephalography; Anesthesia; Opioid; Laryngoscopy
Although changes in the electroencephalogram (EEG) during induction and maintenance of anesthesia have been reported to reflect planes of anesthetic depth (Brazier and Finesinger 1945; Martin et al. 1959; Kato et al. 1964; Smith et al. 1984), variability in drug effects on the EEG have made generalized conclusions concerning the correlation between observed EEG changes and anesthetic levels questionable (French et al. 1953; Arduini and Arduini 1954; Faulconer and Bickford 1960; Darbinjan et al. 1971; Clark and Rosner 1973a,b). Since each anesthetic produces its own dis- tinctive analog EEG pattern (Faulconer and Bickford 1960), Fourier transforms of the EEG signal as defined by spectral edge and power spectrum have generated no single profile outlining a scale of anesthetic levels applicable for all anesthetics. These narrow limitations of the EEG statistical properties (frequency and power) include little infor-
* Corresponding author. Tel.: (617) 726-1596; Fax.: (617) 726-7536.
mation pertinent to the rhythmicity, synchrony, mor- phology, or harmonics of the wave forms. Previous studies explaining the mechanisms for generating reac- tive alpha activity in awake animals and barbiturate-in- duced alpha frequency spindles in anesthetized ani- mals demonstrated coherence among electrical activi- ties observed in subcortical and cortical structures (Andersen and Andersson 1967; Lopes da Silva et al. 1973, 1980). EEG coherence spectra from scalp record- ings have also been analyzed by bispectral analysis. This analysis determines both EEG linear (frequency and power) and non-linear (phase and harmonic) com- ponents and quantitates the interfrequency phase coupling of EEG signals through an examination of the fundamental frequency components and their associ- ated higher-order harmonic relationships. Significant phase coupling of EEG frequency bands of the awake human EEG demonstrated sharp peaks in the EEG bispectra within narrow frequency bands (Barnett et al. 1971; Dumermuth et al. 1971) and though the magni- tude of the bispectra may have varied among individu-
EEG BISPECTRUM AND RESPONSES TO INTUBATION 195 als and lead placement, the dynamics of the bispectra were consistent: quadratic coupling (or phase locking) of frequencies occurred among component waves. Given the lack of a suitable EEG method capable of correlating the analog EEG or its statistical properties with anesthetic levels, and given the strong coher- ence relationships among certain EEG rhythmic activ- ity within various subcortical structures and the cortex, we investigated whether bispectral analysis of EEG contained information previously left unexamined which could predict responses to surgical or anesthetic manipulations mediated principally by subcortical gen- erators. In this study, we used a common model for assessing anesthetic depth (Rampil and Matteo 1987; White and Boyle 1989) and evaluated bispectral analy- sis relative to power spectral analysis as a predictor of increases in blood pressure and heart rate during laryngoscopy and intubation in patients undergoing opioid-based general anesthesia.
ethods and materials
The studies were approved by the Subcommittee on Human Studies. Informed consent was obtained from 43 patients who were scheduled for elective non-cranial surgery. There were 23 men and 20 women, between 19 and 62 years, mean of 44. Patients were excluded from study if they had a history of hypertension, seizure disorder, or other chronic neurologic or psychiatric illness, or if they were taking antihypertensive medica- tions.
Electroencephalograph ,
Five Grass E5 gold cup electrodes were placed according to the international 10-20 system in a fronto-parietal (FP1, FP2, P3, P4) montage referred to the Cz. Electrodes were secured with collodion and electrode impedances were kept at less than 2000 12. The time constant was set at 0.3 sec and high fre- quency filter at 100 Hz. The analog EEG was recorded throughout the operation using a prototype portable computer system (Aspect Medical Systems Model B500 Spectral EEG Monitor) comprised of a modified Toshiba 386/20 computer with an analog-to-digital converter. The sampling rate for data acquisition was 1250/sec. Each 4 sec epoch was analyzed, with each successive epoch overlapping the one before by 3 sec. A 2 min moving window of the analog EEG was divided into 4 sec epochs with an overlap of 3 sec between successive epochs. The resultant 120 epochs were then used to compute averaged bispectral and power spectral parameters for that period. A 5 min baseline recording was performed before the adminis- tration of anesthesia. Physiologic data, including blood pressure and heart rate were acquired and stored in time-synchronized files in the computer for subsequent off-line processing.
Anesthesia management
Approximately 90 min before induction of anesthe- sia, patients were given diazepam orally (0.05-0.15 mg/kg). Patients were continuously monitored with an electrocardiogram, pulse oximeter, capnograph, and brachial blood pressure cuff device (Dinamap Model) which provided minute by minute blood pressure read- ings. Induction of anesthesia was performed with thiopental (4.0-6.0 mg/kg) and 60% nitrous oxide in oxygen, followed by a muscle relaxant (vecuronium 0.1 mg/kg) to facilitate intubation. Fifteen seconds after the muscle relaxant was administered, each patient was given an intravenous injection of 1 of 4 opioid dose regimens or normal saline according to a randomiza- tion schedule: alfentanil 15 ~g/kg (group 1A), alfen- tanil 30/~g/kg (group 1B), sufentanil 0.5 ~g/kg (group 2A), sufentanil 1.5 p~g/kg (group 2B), or normal saline (group 3). Patients were then ventilated by mask for 3 min and laryngoscopy was performed. The anesthetic management after intubation was determined by the attending anesthetist.
Definitions of responders and non-responders
In order to derive the EEG power spectrum and the bispectral variables, the off-line analysis of EEG records included the 3 min between the administration of the opioid (prestimulus), and start of laryngoscopy (stimulus). Responders were defined as patients who demonstrated an increase of 20% or greater in mean arterial blood pressure within 3 rain of laryngoscopy. In a separate analysis, another category of responders was defined as those patients whose heart rate in- creased by 20% or greater within 3 rain after laryn- goscopy independent of any blood pressure changes. The lowest recorded value of mean arterial pressure and heart rate within the prestimulus period and the highest recorded value in the 3 min post stimulus were used to assess any change in response to laryngoscopy.
EEG parameters analyzed
Five EEG parameters were derived, each separately for frontal, parietal, and frontal/parietal montages. (1) Discriminant bispectrum (Dscrm BIS) is a uni- variate index generated by linear combination of bi- spectral variables for the preintubation period. The bispectral index is calculated using a linear combina- tion of measured bispectral parameters with corre- sponding weighting coefficients. The coefficients were derived using backwards-stepwise linear regression and discriminant analysis of a 165 patient database which included patients from this study. A multiple linear regression model with backward selection (the RSTEP
196 L.A. KEARSE, Jr. ET AL.
routine from IMSL, Houston, TX) was used to elimi- nate bispectral parameters which did not significantly contribute toward determining anesthetic state (re- sponse category). In this model, parameters with P values of greater than or equal to 0.005 were rejected and data from the eliminated variables were removed from the data matrix for the next step in the analysis. A discriminant analysis (the DSCRM routine from IMSL, Houston, TX) was then performed on the resul- tant data matrix. This discriminant analysis produced two sets of coefficients (and intercepts), one for each of the responder and non-responder descriptors. The differences between the coefficients for each remaining bispectral variable [(responder coefficient 1)-(non- responder coefficient 1, etc.)] and the intercepts were determined and were then used to produce a combined linear index that would classify a patient as a re- sponder if the index was above zero. In order to limit the dynamic range (0.0-1.0) and display the linear index in a more simplified form, it is scaled and expo- nentiated using the following function: exponential in- dex= 1/(1
q-e( -linear
index .
This results in an index with values which range between 0 and 1, with values above 0.5 corresponding to levels indicating an increas- ing likelihood of response. (2) Discriminant power spectrum (Dscrm PS) is a similar univariate index generated by a linear combina- tion of 7 spectral variables for the preintubation pe- riod. A series of parameters were calculated from the power spectrum of each epoch, and the individual power spectral parameters from the last 10 epochs were averaged together. The variables studied were relative delta (0.5-3.75 Hz), relative theta (4.0-7.75 Hz), relative alpha (8.0-13.5 Hz), and relative beta (13.75-30.0 Hz). (3) Delta power is the relative delta power in dB. (4) Spectral edge 95% is that value below which 95% of the EEG power exist. (5) Spectral edge 50% is the median frequency. The power spectral parameters and bispectral index computed from each channel were averaged. Thresh- olds providing the maximum accuracy for the power spectrum and the bispectral index were selected by using receiver operating curve characteristics. rectly identified as such, specificity was the percent of non-responders correctly identified, and accuracy was defined as the percent of total patients correctly identified as either responders or non-responders by each of the processed EEG variables. Differences that may be attributable to (dose) treatment regimens were examined by logistic regression. Statistical significance was defined as P < 0.05.
Complete data, including hemodynamic results and EEG parameters, were obtained in 39 patients. Twenty-seven of these patients had a 20% increase in mean arterial blood pressure at laryngoscopy and met the criteria for responders. . Table I outlines the distribution of responders and non-responders among the 5 drug groups. Patients in group 1A (low dose alfentanil) and group 3 (normal saline) had a higher response rate and significantly greater mean increases in both blood pressure and heart rate than the other treatment groups. As shown in Table II, there were no significant demographic differences in age or weight between re- sponders and non-responders, and treatment groups could not be distinguished reliably on the basis of any preintubation demographic or hemodynamic parame- ter. There were also no significant differences in pre- stimulus heart rate and blood pressure among the 5 drug treatment groups or between the responders and non-responders. The increase in heart rate among sub- jects did not differentiate responders from non-re- sponders, thus indicating a lack of correlation between blood pressure and heart rate increases in response to laryngoscopy.
TABLE I Distribution of responders and non-responders following intubation in each assigned treatment group. Treatment Responders Non-responders Total group 1A
tatistical analysis
Data were analyzed using 1-way analysis of variance
and Student's t tests. Data are presented as the mean 3 and the standard error of the mean. Diagnostic sensi-
tivity was defined as the percent of responders cor-
8 1 9 6 3 9 6 2 8 0 5 5 7 1 8 27 12 39 Fig. 1. A plot of the distribution of pre-intubation bispectral index and 95 spectral edge frequency values for responders (defined as having an increase in mean arterial blood pressure of > 20 ) and non-responders. Each point represents an individual patient. The triangles represent patients in group 2B who received 1.5/~g/kg of sufentanil. Both EEG parameters were measured using a bifrontal montage referenced to Cz (average of FP1 to Cz and FP2 to Cz).
RESPONDER .............................................................................. ........ o...o~.ooo-o--mlm,--.-.o--.~ .................. ~ .............. NON-RESPONDER ............................................... ~ ...................... ~ ................... A~ .................. o-.-..--o .................... .............................................
o4~o m e qmo eme Qo
..................................... • ................................ ~ .................. ~,o--.-~ ................. ~ ...... o~,---..--o.~ .................................. o .........
I0 15 20 25 95 SPECTRAL EDGE (HZ)
198 TABLE II Demographic and hemodynamic comparisons and non-responders to laryngoscopy. between responders Responders Non-responders (n = 27) (n = 12) Age (years) 40.22 ± 1.86 42.5 ± 3.03 Weight (kg) 72.85 ± 2.73 76.25 ± 5.22 Pre-intub. BP (mm Hg) 84.71 ± 3.26 82.08 ± 5.35 Pre-intub. HR (beats/min) 75.71 ± 2.38 69.75 ± 4.17 Change in BP 42.01 ± 2.79 8.92 ± 1.94 * * Change in HR 22.11± 4.39 10.33±4.67
P <
0.001. Data given as mean _+ S.E.M. TABLE III EEG parameters prior to laryngoscopy (frontal montage). Responders Non-responders (n = 27) (n = 12) Bispectrum 0.67 _+ 0.02 0.45 ± 0.04 * Power spectrum 0.53 ± 0.02 0.50 ± 0.03 Delta power (dB) - 5.86 ± 0.59 - 4.81 ± 0.62 95 Spectral edge frequency (Hz) 22.1 ± 0.8 21.9 ± 0.9 50 Spectral edge frequency (Hz) 9.7 ± 1.0 7.8 _+ 1.1 * P < 0.001. Data given as mean _+ S.E.M.
A comparison of mean EEG parameters before laryngoscopy is presented in Table III. Based on the receiver operating curve characteristics, the diagnostic thresholds of 0.41 for the bispectral index and 14 for the 95 spectral edge were used to calculate the percentages shown in Table IV. Although the overall BIS accuracy in predicting a response to laryngoscopy was consistently higher than power spectrum and spec- tral edge for all montages, a statistically significant difference between response groups was observed for the bispectrum only when using the frontal montage. The identification of responders and non-responders
TABLE IV Accuracy tests for bispectral and power spectral variables. Sens Spec T. acc
rontal montage
Bispectrum 100 50 85 Power spectrum 74 50 67 Delta power 48 83 59 95 Spectral edge frequency (Hz) 96 0 67 50 Spectral edge frequency (Hz) 96 0 67
Parietal montage
Bispectrum 100 8 72 Power spectrum 85 33 69 Delta power 93 17 69 95 Spectral edge frequency (Hz) 100 8 72 50 Spectral edge frequency (Hz) 100 17 74 L.A. KEARSE, Jr. ET AL.
by BIS and SEF 95 parameters are presented in Fig. 1. Among the 5 treatment groups, patients who re- ceived high doses of sufentanil had no responders. In this group, the amount of administered drug alone predicted a lack of response to laryngoscopy. There- fore, in order to determine whether the EEG parame- ters could distinguish between responders and non-re- sponders independent of drug amount, two additional analyses were completed: logistic regression analysis was performed on all results and on a subset of the two treatment groups, group 1B (alfentanil 30 p,g/kg) and 2A (sufentanil 0.5/xg/kg), in which there were similar numbers of responders and non-responders. Analysis of the complete data set showed that the bispectral index was a good predictor of response (P < 0.002) and knowing the dose group did not improve the predictive results. In a similar analysis of subsets 1B and 2A, the relationship between the bispectral index and blood pressure response remained significant (P<0.026). These analyses indicate that the bispectral index distin- guished responders from non-responders indepen- dently of the amount of drug given.
Bispectral analysis of the EEG is a potentially useful indicator of anesthetic adequacy, defined as the ability to prevent a 20 or greater increase in blood pressure to laryngoscopy and intubation. Using a diagnostic threshold of 0.41, the bispectrum derived from frontal leads yielded a sensitivity of 100 and a specificity of 50 with an accuracy of 85 . The sensitivity of the bispectrum in detecting responders was independent of the amount of drug delivered. This finding was of particular interest because in previous reports EEG manifestations of drug effects made difficult any visual identification of specific EEG pattern changes capable of discriminating among levels or gradations of altered consciousness (Faulconer and Bickford 1960; Clark and Rosner 1973a). Other studies have also demonstrated a dissociation between serum drug concentrations, the EEG patterns, and the ability to predict responses to different noxious stimulation (White and Boyle 1989; Hung et al. 1992). In other words, even after knowing the amount of drug delivered, the measured serum drug concentration, and the changes in the EEG due to the drugs, patients' responses to laryngoscopy and intubation could not be reliably predicted. Despite our use of a variety of drugs, including oral (diazepam), intravenous (thiopental and narcotic) and inhalational (nitrous oxide) agents, the bispectrum maintained a high accuracy in predicting responders and non-re- sponders.
of 7