The electroencephalogram (EEG) can be used to measure the effects of anesthetics on the brain [1,2]. Recently, research has been directed toward using computer-processed EEG variables for intraoperative monitoring of anesthetic adequacy. Many of these derivatives (such as zero crossing frequency, density spectral array, aperiodic analysis) have not been useful for monitoring anesthetic adequacy due to the different effects produced by the various anesthetics, and large interpatient variability [3,4]. Most work to date has focused on utilizing power spectral analysis to derive various univariate descriptors of EEG activity. One commonly used variable, the Spectral Edge Frequency (SEF), reduces the power spectrum to a single number that represents the highest frequency present in the EEG . It has long been thought to correlate with anesthetic adequacy . SEF correlates well with plasma concentrations of opioids  and thiopental [8,9], and may be used to predict the hemodynamic response to laryngoscopy and endotracheal intubation . It may also be used to guide opioid administration in order to reduce drug dose and the magnitude of hemodynamic responses . White and Boyle  have been unable to correlate SEF with hemodynamic responses. Dutton et al.  found it a good predictor of movement to trocar insertion with a variety of anesthetic techniques, but in another study during one minimum alveolar anesthetic concentration (MAC) isoflurane anesthesia, SEF did not predict movement at skin incision .
The median frequency of the EEG power spectrum has also been studied during anesthesia. A median frequency of less than 5 Hz is associated with unconsciousness during etomidate  or methohexital  anesthesia. The median frequency has also been used in an adaptive feedback control algorithm for the closed-loop control of propofol anesthesia . Nevertheless, none of these methods has been shown to be sufficiently reliable for general use for assessing anesthetic efficacy during routine procedures .
Conventional power spectral analysis assumes a Gaussian, stationary, and first order (linear) model of the frequencies within the EEG; i.e., the amplitudes of the EEG are normally distributed, its statistical properties do not change over time, and its frequency constituents are uncorrelated. Under this assumption, the EEG is considered to be made up by a linear superimposition of statistically independent sinusoidal wave components (fundamentals). Only frequency and power estimates are considered, while phase information is generally ignored.
In reality, biologic systems exhibit significant nonlinear complexities which do not conform to the assumptions of conventional power spectral analysis. To overcome this limitation, bispectral analysis (BIS) provides a means by which quadratic (second order) interactions can be quantified . It does so by quantifying the phase coupling between any two frequencies and a third frequency (harmonic) at their sum (or difference). The extent of the coupling between two frequencies (bicoherence) can vary from 0% if no harmonic is being generated, to 100% if a harmonic is generated for the duration of the period analyzed. BIS provides a more comprehensive description of the information available from Fourier analysis than the power spectrum, thus allowing the construction of a better profile of changes that may take place in the EEG under a variety of clinical conditions. A detailed description of BIS has been published . In a previous retrospective study, we demonstrated that BIS can be used to predict patient movement in response to skin incision during isoflurane anesthesia with a sensitivity of 96%, a specificity of 63%, and an accuracy rate of 83% . That study was limited in that only responses to volatile anesthetics were assessed.
The present study was designed to investigate whether BIS predicted movement at incision when an opioid was used in combination with isoflurane and to compare these results with an intravenous (IV) anesthetic technique.
After institutional review board approval, 50 patients (ASA grade I-III) ranging in age from 20 to 65 yr (38.6 +/- 10.8 yr), scheduled for elective noncranial surgical procedures, were studied. Written, informed consent was obtained from each patient. Patients with known neurologic disorders, including patients requiring anticonvulsant therapy and patients in whom the anesthetics were contraindicated, were excluded from the study.
Patients were randomly allocated to two groups: propofol/alfentanil (31 patients) or isoflurane/alfentanil (19 patients). After the first 20 patients in the propofol/alfentanil group had been studied, there was a low incidence of movement, and additional patients were recruited to this group (see Results). No sedative premedication was administered prior to surgery. After breathing oxygen, anesthesia was induced and maintained in the first group with a combination of IV propofol and alfentanil. In the second group, anesthesia was induced with propofol and maintained with isoflurane/alfentanil. Endotracheal intubation was performed after neuromuscular block with succinylcholine 1 mg/kg IV, and ventilation was with air/O2. To achieve stable concentrations of IV anesthetics during induction and maintenance of anesthesia, a computer-controlled infusion device (Ohmeda 9000 infusion pump with a Psion controller; Ohmeda, Madison, WI)  was used. In the propofol/alfentanil group, the initial infusion target plasma concentrations were propofol 4 micro gram/mL and alfentanil 125 ng/mL. In the isoflurane/alfentanil group, after induction of anesthesia with propofol 1.5-2 mg/kg, the initial end-tidal concentration of isoflurane was adjusted to 0.5% and the alfentanil target plasma concentration was 125 ng/mL.
The initial surgical incision was made after return of neuromuscular function as determined by a peripheral nerve stimulator, and when the target concentrations had been maintained for at least 10 min. A determination of any movement in the succeeding 2 min was recorded. Any movement (except coughing and bucking on the tube) was considered a positive response.
In the alfentanil/propofol group, when movement occurred at incision, the propofol target concentration for the next patient was increased by 0.5 micro gram/mL; when movement was not detected it was decreased by 0.5 micro gram/mL. In the isoflurane/alfentanil group, if movement was detected at incision, the isoflurane end-tidal concentration was increased by 0.1% for the next patient or decreased by 0.1% if movement was not detected. In this manner, a modification of the up-down method of Dixon , roughly equal numbers of patients were expected to fall into the move and no-move categories.
Before induction of anesthesia silver/silver chloride gel-filled electrodes were secured using collodion to the left and right parietal (P3 and P4) and left and right frontal (Fp1 and Fp2) regions and referred to a vertex electrode (Cz). Electrode impedance was maintained below 5 kohms. EEG data were band-pass filtered to 0.5-300 Hz, amplified using a gain of 5000 and then digitized at 2000 samples/s using a laptop microcomputer-based EEG system (Model B500 Spectral EEG Monitor; Aspect Medical Systems, Inc., Framingham, MA). Arterial blood pressure and heart rate were measured every minute noninvasively using a Dinamap (Critikon) and acquired simultaneously with the EEG using a serial interface between the Dinamap and the EEG monitor. All acquired data were stored on disk and transferred to tape for subsequent analysis.
EEG data were analyzed off-line. For every patient, the 1-min period of EEG data prior to incision was divided into 4-s epochs and used to compute the bispectrum [18,23,24] and the power spectrum. Two dimensional bispectral data matrices representing all frequency pairs in the frequency band between 1 and 30 Hz were produced for statistical analysis: bispectral density, bispectral real triple product, and bicoherence . The bispectral variables were analyzed using a multivariate logistic regression model to produce BIS, a univariate probability function . In comparison to our prior study , BIS was rescaled to compensate for a difference in high-pass filter settings (1 Hz vs 0.5 Hz).
Statistical analysis was with analysis of variance and Neuman-Keuls post hoc testing, nonpaired Student's t-tests, and logistic regression as appropriate. Mean values for BIS during the 1 min prior to incision were used to determine intergroup differences (responders versus nonresponders in each of the treatment groups). Results are expressed as mean +/- SD. P values <or=to 0.05 were considered statistically significant.
Fifty patients were enrolled in the study. By randomization, 31 patients were assigned to receive propofol/alfentanil and 19 patients to receive isoflurane/alfentanil. Baseline demographic variables and preincision EEG variables are summarized in Table 1. There were no significant differences among any of the demographic variables. The number of patients in each group is not equal. Among the first 20 patients in the propofol/alfentanil group, only four moved in response to skin incision due to relatively high initial target plasma concentrations of propofol and alfentanil. Therefore, additional patients were enrolled in an attempt to obtain a more balanced distribution of responders and nonresponders. Nevertheless, the propofol/alfentanil group still had a significantly lower overall response rate (7/31, 23%) compared to the isoflurane/alfentanil group (12/19, 63%).
The two treatment groups differed significantly in all of the preincision calculated EEG variables including BIS, SEF 95%, median frequency, and relative delta power (P <or=to 0.001) Table 1.
(Table 2) summarizes the mean EEG values and diagnostic test performance results for movers and nonmovers in both treatment groups. BIS is more accurate at predicting movement than the other EEG variables. A significant difference is found between the BIS values for movers versus nonmovers within each of the two treatment groups Figure 1. Response group (P <or=to 0.00001) and treatment group assignment (P <or=to 0.01) were significant factors in the analysis of variance but their interaction was not.
The Newman-Keuls post-hoc test was performed to differentiate between the treatment groups and the response rate. The results appear in Figure 1 and are described in the legend. BIS was successful at differentiating between movers and nonmovers in each of the two treatment groups; the BIS values for the nonmovers in the isoflurane group were not statistically different from the movers in the propofol/alfentanil group.
(Table 3) contains the mean hemodynamic variables (mean blood pressure and heart rate at baseline and immediately prior to incision) for the two groups separated into movers versus nonmovers. The hemodynamic variables in the two groups do not vary significantly and thus are not useful in predicting patient response.
(Figure 2 and Figure 3) show the dose-response relationships for each of the individual patients in the two treatment groups. Logistic regression analysis showed that the assigned target propofol concentrations did not predict response in this population, whereas BIS was a significant (P <or=to 0.0005) predictor. Preincision expired isoflurane levels alone were significant (P <or=to 0.01), but BIS still performed better (P <or=to 0.003) than measured dose alone. The corresponding probability of response curves calculated by logistic regression for both treatment groups are shown in Figure 4.
Fourier analysis of the EEG generates information about the frequency, power, and phase of the signal. In conventional power spectral analysis, the phase information is discarded. Bispectral analysis utilizes this phase information to examine interfrequency relationships in the EEG. The EEG is a complex signal and it may be that important information relating to anesthetic effect is contained in the phase relationships of the EEG.
In a previous study, we demonstrated that BIS predicted movement in patients anesthetized with isoflurane using a stepwise logistic regression technique to produce the most accurate combination of bispectral and power spectral variables to predict movement . In such a study the regression analysis will produce a best fit for data in the particular sample dataset. We attempted to mitigate this with subjects from a larger reference database using various anesthetics. Thus, the danger of spurious results due to inappropriate fitting to the study dataset should have been reduced. However, prospective confirmation of the accuracy of the BIS is needed to assess the clinical utility of the analysis technique.
The purpose of this study was to determine whether BIS can be used to predict movement with a combination of anesthetics, namely propofol/alfentanil and isoflurane/alfentanil. To assure a predictable reduction in the MAC of isoflurane, a computer-assisted infusion device was used to keep an approximately constant plasma concentration of alfentanil over time. Similarly, the same method was used in the propofol/alfentanil group to maintain constant plasma levels of alfentanil, while the propofol level was increased or decreased in 0.5-micro gram/mL plasma concentration increments with a modified Dixon updown method.
In a MAC study it would be expected that approximately 50% of patients would move at 1 MAC and that a small proportion would move at 1.25 MAC. The anesthetic treatment doses were chosen in the expectation that approximately 50% of the patients would move in response to skin incision. In this study, the incidence of movement in the propofol/alfentanil group was low due to a relatively high starting dose of propofol. More patients were entered into the propofol/alfentanil group in an effort to obtain a more balanced distribution of patient response to skin incision. Nevertheless, we still observed a significant difference in response ratios between our two treatment groups.
Our dose response results highlight some of the current dosing limitations faced by practicing anesthesiologists. When using inhaled anesthetics, such as isoflurane, in combination with opioid infusions, measurement of end-tidal concentrations does provide some useful indication of likely patient response. For IV anesthetics such as propofol or alfentanil, real time measurement of actual blood levels is not yet possible. Although pharmacokinetic-based infusion devices have been developed to deliver drugs to predetermined target blood levels, their impact on improving clinical care must still be demonstrated. In this study, assigned target propofol concentrations could not be used to predict patient response. Measurement of actual preincision propofol concentrations might be expected to achieve better correlation to patient response, but it was not attempted in this study.
Anesthesiologists currently lack a direct measure of anesthetic effects on the brain that spans all anesthetics and is applicable at clinically used ranges of anesthesia [3,7]. Anesthetic dose is generally given based on a constellation of autonomic signs. Since the target site of action of general anesthetics is the brain, it would not be unreasonable to expect a neurophysiologic measure of anesthetic effect to exist. Such a measure should be sensitive enough to detect insufficient anesthesia and be able to predict recovery from anesthesia. It should also be independent of anesthetic used, and should correlate with anesthetic concentration at the site of action. Movement in response to skin incision during anesthesia represents a standard test of anesthetic effect, and we have used this to assess the predictive value of the EEG bispectrum.
Consistent with the observed differences in response rates, the preincision BIS values Table 1, and interestingly, all of the other measured EEG variables, were significantly different when comparing the two treatment groups. The propofol/alfentanil group had lower BIS values than the isoflurane/alfentanil group probably due to relatively higher doses of anesthetic. The fact that very few of the original 20 patients in the propofol/alfentanil moved can also be attributed to the relatively high doses of propofol or a greater synergy between propofol and alfentanil compared to isoflurane and alfentanil.
It is not clear from our data whether BIS performs similarly with both isoflurane and propofol. Evidence for a drug-dependent effect includes the observation that both movers and nonmovers in the propofol group had significantly lower mean BIS levels compared to their respective response groups in the isoflurane/alfentanil-treated patients Figure 1. Furthermore, movers in the propofol group could not be distinguished from isoflurane nonmovers. However, the probability of response curves obtained by logistic regression for each treatment group were almost identical Figure 4, suggesting that the unequal distribution of responses in the two treatment groups contributed to the apparent imbalance in means. The impact of concomitant opioid analgesic administration could not be addressed in this study, since all patients received identical alfentanil infusions. Preliminary results from our ongoing studies suggest that opioid administration can complicate interpretation of the relationship between EEG measures and the probability of patient response (PS Sebel, personal communication, 1994).
These findings support the possibility that different anesthetics have different effects on the EEG. BIS may not be independent of anesthetic technique, but is a predictor of movement in response to skin incision when comparing movers to nonmovers in the same treatment group. BIS is a markedly better predictor of patient movement than any of the other variables in both treatment groups Table 2, including available dose information (expired isoflurane or target propofol). This suggests that interfrequency phase coupling, a nonlinear dynamic property of the EEG, which is measured with BIS, may contain clinically useful information in the assessment of anesthetic adequacy.
Increases in arterial blood pressure and heart rate in response to skin incision are considered indicators of inadequate anesthesia, but there is not always good correlation between movement and hemodynamic response. In this study, preincision hemodynamic variables did not predict patient movement in response to skin incision. This demonstrates the importance of developing a monitor of anesthetic effects in that hemodynamic and other autonomic responses cannot be relied on as good predictors of adequate anesthesia. The reasons that patients without apparent cardiovascular disease do not mount a sympathetic response are unknown.
In conclusion, BIS appears to be a good predictor of movement within treatment groups but may not be independent of the anesthetic regimen used.
1. Brazier MAB, Finesinger JE. Action of barbiturates on cerebral cortex: electroencephalographic studies. Arch Neurol Psychol 1945;53:51-6.
2. Falconer A, Bickford RG. Electroencephalography in anesthesiology. Springfield, IL: Charles C Thomas, 1960.
3. Sebel PS, Heneghan CP, Ingram DA. Evoked potentials. A neurophysiological indicator of depth of anaesthesia? Br J Anaesth 1985;57:841-2.
4. Levy WJ, Shapiro HM, Maruchak G, Meathe E. Automated EEG processing for intraoperative monitoring: a comparison of techniques. Anesthesiology 1980;53:223-6.
5. Rampil IJ, Holzer JA, Quest DE, et al. Prognostic value of computerized EEG analysis during carotid endarterectomy. Anesth Analg 1983;62:234-41.
6. Rampil IJ, Matteo RS. Changes in EEG spectral edge frequency correlate with the hemodynamic response to laryngoscopy and intubation. Anesthesiology 1987;67:139-42.
7. Scott JC, Ponganis KV, Stanski DR. EEG quantitation of narcotic effect: the comparative pharmacodynamics of fentanyl and alfentanil. Anesthesiology 1985;62:234-41.
8. Hudson RJ, Stanski DR, Saidman LJ, Meathe E. A model for studying depth of anesthesia and acute tolerance to thiopental. Anesthesiology 1983;59:301-8.
9. Buhrer M, Maitre PO, Hung OR, et al. Thiopental pharmacodynamics. Defining the pseudo-steady-state serum concentration-EEG effect relationship. Anesthesiology 1992;77:226-36.
10. Sidi A, Halimi P, Cotev S. Estimating anesthetic depth by electroencephalography during anesthetic induction and intubation in patients undergoing cardiac surgery. J Clin Anesth 1990;2:101-7.
11. White PF, Boyle WA. Relationship between hemodynamic and electroencephalographic changes during general anesthesia. Anesth Analg 1989;68:177-81.
12. Dutton RC, Smith WD, Smith NT. Does the EEG predict anesthetic depth better than cardiovascular variables? [abstract] Anesthesiology 1990;73:A532.
13. Dwyer R, Rampil I, Eger EI,II, Bennett HL. The EEG does not predict movement in response to surgical incision at 1.0 MAC isoflurane [abstract]. Anesthesiology 1991;75:A1025.
14. Schwilden H, Schuttler J, Stoeckel H. Quantitation of the EEG and pharmacodynamic modelling of hypnotic drugs: etomidate as an example. Eur J Anaesthesiol 1985;2:121-30.
15. Schwilden H, Schuttler J, Stoeckl H. Closed loop feedback control of methohexital anesthesia by quantitative EEG analysis in humans. Anesthesiology 1987;67:341-7.
16. Schwilden H, Stoeckel H, Schuttler J. Closed-loop feedback control of propofol anaesthesia by quantitative EEG analysis in humans. Br J Anaesth 1989;62:290-6.
17. Stanski DR. Monitoring depth of anesthesia. In: Miller RD, ed. Anesthesia. 4th ed. New York: Churchill Livingstone, 1994:1127-59.
18. Huber PJ, Kleiner B, Gasser T, Dumermuth G. Statistical methods for investigating phase relations in stationary stochastic processes. IEEE Trans Audio Electroacoust 1971;19:78-86.
19. Sigl JC, Chamoun NG. An introduction to bispectral analysis for the electroencephalogram. J Clin Monit 1995. In press.
20. Sebel PS, Bowles SM, Saini V, Chamoun N. EEG bispectrum predicts movement during isoflurane anesthesia. J Clin Monit 1995. In press.
21. Kenny GNC, White M. A portable computerised infusion system for propofol. Anaesthesia 1990;45:692-3.
22. Dixon WJ. The up-and-down method for small samples. Am Stat Assoc J 1965;60:967-78.
23. Rosenblatt M, Van Ness JW. Estimation of the bispectrum. Ann Math Statist 1972;36:1120-36.
© 1995 International Anesthesia Research Society
24. Brillinger DR. An introduction to polyspectra. Ann Math Statist 1965;36:1351-74.