A high dose of opioids associated with a low dose of propofol has become a popular anesthetic technique. However, the influence of opioids on the electroencephalographic phenomenon related to induction of anesthesia and, thereby, on the quantitative parameters used in the depth-of-anesthesia estimation is not well known.

Twenty-seven patients were divided into three groups to receive saline, low-dose remifentanil (7.5 microg x kg x h) or high-dose remifentanil (30 microg x kg x h) during induction of anesthesia with propofol (30 mg . kg . h). Electroencephalogram was recorded from Fz electrode, and its time-frequency properties in the patient groups were analyzed from the induction of anesthesia to the occurrence of burst suppression pattern. The group differences in 14 quantitative spectral parameters used in the depth-of-anesthesia estimation were examined as well.

The time-frequency properties of electroencephalogram were different between groups. The high-frequency (greater than 14 Hz) activity during light anesthesia was decreased in remifentanil groups; whereas, increased activity in extended alpha band (7-14 Hz) and decreased activity in delta band (0.5-4 Hz) was observed during deep anesthesia. This resulted in statistically significant changes in all 14 quantitative parameters.

The effect of remifentanil on the spectrum and quantitative parameters of electroencephalogram is significant and strongly dependent on the level of anesthesia. Coadministration of opioids therefore challenges the reliability of the spectral properties of electroencephalogram in the depth-of-anesthesia estimation by using a frontal montage. Furthermore, the finding has implications for design of opioid coadministration studies.

INCREASING concentrations of anesthetics in the blood produce a continuum of electroencephalographic changes. With propofol, the changes obey roughly the following pattern. First, an increase of the high-frequency (greater than 20 Hz) activity; next, a decrease of the high- frequency activity and increase of the middle-frequency (10–20 Hz) activity; finally, a decrease of the middle-frequency activity and an increase of the low-frequency (0.5–5 Hz) activity.1,2In very deep anesthesia, the burst suppression pattern (BSP) begins. Potent inhaled GABAergic anesthetics, such as sevoflurane, isoflurane, and desflurane, follow the same electroencephalographic pattern.3

Several measures have been developed for the electroencephalogram-based depth-of-anesthesia estimation. These include both simple parameters, such as the spectral edge frequency 95% (SEF95%) and median power frequency (MPF),4and more complex commercial indices like Bispectral Index (A-2000 BIS® monitor; Aspect Medical Systems Inc., Newton, MA)5or State and Response Entropy (M-entropy® module; GE Healthcare Finland Oy, Helsinki, Finland).6Generally, in all these measures, first or higher order spectrum is utilized.7

Remifentanil is an opioid that is increasingly used in the operating room and in intensive care.8,9It has analgesic and cardiovascular effects similar to those of alfentanil but with a shorter duration of action.10The changes remifentanil produces in the electroencephalogram are characteristic of μ-receptor agonists. They consist of decreasing frequency and increasing amplitude, culminating eventually in delta activity at maximal drug effect.11,12During the induction of propofol anesthesia, the above-mentioned basic frequency progression pattern has shown to be robust against coadministration of remifentanil.13However, there is evidence that addition on remifentanil during steady-state propofol anesthesia affects to some extent, for example, bispectral index and SEF95%, suggesting some unknown contribution of this opioid to the electroencephalogram.14

This paper studies the effect of the coadministration of remifentanil on the electroencephalogram phenomenon occurring during the induction of propofol anesthesia. The effects are examined from the beginning of propofol infusion to the onset of BSP. We hypothesize that the opioids result in characteristic changes to the signal's time-frequency properties, even though the basic propofol-induced frequency progression pattern is left untouched. Furthermore, we examine if the possible changes reflect on the quantitative spectral parameters commonly used in the depth-of-anesthesia estimation.

## Materials and Methods

### Patients and Clinical Protocol

After the study was approved by the institutional Ethics Committee of South Carelia Central Hospital, Lappeenranta, Finland, 27 patients (table 1) scheduled for elective surgical procedure gave informed written consent to participate. Patients having cardiovascular or neurologic diseases, diabetes, or a body mass index greater than 30 were excluded as well as the patients using drugs that affect central nervous system. The patients were randomly assigned to one of three groups (9 each): R0, R1, and R2. Depending on the group, patients received saline (R0), low-dose remifentanil (R1, 7.5 μg · kg^{−1}· h^{−1}) or high-dose remifentanil (R2, 30 μg · kg^{−1}· h^{−1}) during the induction of anesthesia with propofol (30 mg · kg^{−1}· h^{−1}). The infusion of propofol started 1 min after the start of the saline/remifentanil infusion. Both drugs were infused at a fixed rate until the BSP was detected from the electroencephalogram channel of S/5 monitor (GE Healthcare Finland). During the induction, loss of obeying verbal command (LVC) was assessed by asking at 15-s intervals the patient to squeeze the anesthesiologist's (SM) hand. After the onset of BSP, propofol infusion rate was decreased to 18 mg · kg^{−1}· h^{−1}, by which the BSP was sustained for at least 5 min. Tracheal intubation was facilitated with 0.6 mg · kg^{−1}rocuronium. During the induction process, electroencephalogram was recorded from 17 different electrode locations according to the international 10/20 system15with an Embla polygraphic recorder (Medcare, Reykjavik, Iceland). The recorder used a sampling rate of 200 Hz and filtered the signals with a bandpass filter of 0.5–90 Hz. The recording started several minutes before the induction of anesthesia and continued for at least 5 min after the tracheal intubation. To minimize the signal artifact, the patients were told to be quiet and still before the drug infusion began. In the analysis, only the montage Fz with a common average reference was used. This montage was chosen because the modern depth-of-anesthesia monitoring is based on the analysis of frontal electroencephalogram. The data analyzed in this study was also used in our previous publication,13to which the reader is referred for complete details regarding the clinical protocol.

### Data Preprocessing

All the electroencephalographic signal processing presented in this paper was performed with the Matlab technical computing language (The MathWorks Inc., Natick, MA).

The electrooculographic artifacts were removed from the recordings by using the Automatic Artifact Removal toolbox for Matlab.||First, the electroencephalographic data from 17 channels were decomposed into spatial components by using a blind source separation technique based on second order statistics, *i.e.* , second order blind identification.16The components that could be related to electrooculographic activity were then manually detected and left out when the electroencephalographic data were reconstructed from the spatial components.

### Spectrograms

After the electrooculographic artifact removal, spectrograms were calculated from the Fz channel of electroencephalograms by using a short-time Fourier transform with a 3-s Hamming window and 2.9-s overlap. This window length was selected experimentally because it compromised the time and frequency resolution appropriately. The window was also long enough for the analysis of the lowest frequencies of interest. An example of a spectrogram with and without electrooculographic artifact is given in figure 1.

To reduce the large difference in the amplitudes of higher and lower frequencies, the spectrograms were amplitude normalized. In this procedure, the values in a specific frequency were divided by the mean value of all patients in that frequency between the start of the propofol infusion and onset of BSP. The effect of amplitude normalization can be seen in figure 1C. The amplitude normalized spectrograms were used only to visualize the changes in the frequency content of electroencephalograms. They were not used when the quantitative spectral parameters were calculated.

To examine the effect of remifentanil on the spectral characteristics of electroencephalograms, the spectrograms of different groups need to be compared. However, due to the interindividual variability in response to the anesthetic agent, the electroencephalographic spectral progression phenomenon varies in time between patients. Because this hinders the comparison, the time scale of the spectrograms was first normalized by using the method presented in our previous study.17The method is based on calculation of electroencephalographic activity in eight different frequency bands for each patient and by minimizing the mean squared error between the activity trends of different patients by time scaling. This results in patient-specific time scaling factors that can be applied to the spectrograms. In other words, the spectrograms are normalized in time by using the frequency progression phenomenon of electroencephalogram during the induction process. The method has been applied previously to the dataset used in this study, and the results showed that the time scaling factors do not differ between groups, *i.e.* , remifentanil does not affect the time normalization.13The use of the method in this study is therefore reasonable.

After the time normalization, all spectrograms have their own time scale, and they cannot be given as a function of absolute time anymore. Instead, relative time *r* is used. The *r* scale was introduced by Koskinen *et al.* ,18and its further developed version has been presented recently.13,19It can be assumed to describe the phase of the electroencephalographic spectral progression phenomenon during induction of anesthesia. As *r* values are determined from the electroencephalograms recorded during continuous infusion of propofol with a fixed rate, increasing value can be associated with deepening anesthesia. In *r* scale, the start of induction of anesthesia and LVC are used as the points of reference: the values 0 and 1 represent the start of propofol infusion and the position in which the LVC occurs, respectively. The LVC occurs at a different phase of the spectral progression in different groups; therefore, the group R0 positions, *i.e.* , the occurrence of this clinical end point without coadministration of remifentanil, were used. The LVCs of different patients occur also in a slightly different position, and therefore the median of the group R0 LVC points was set to represent the *r* value 1. In this paper, the spectral properties of electroencephalograms are analyzed continuously as a function of *r* and also separately during low *r* values (0 < *r* < 1, defined in this paper as light anesthesia) and high *r* values (1 < *r* < 2, defined in this paper as deep anesthesia). The time-normalization procedure and determination of *r* scale is illustrated in figure 2.

### Spectral Parameters

The following 14 spectral parameters were determined as a function of *r* from the time-normalized (not amplitude-normalized) spectrograms: powers in different frequency bands (total, delta, theta, alpha, and beta), relative powers in different frequency bands (delta, theta, alpha, and beta), spectral entropy (SE), relative beta ratio, MPF, spectral edge frequency 90% (SEF90%), and SEF95%. The parameters were chosen on the basis of their common use in the previous research. Because some of the signals were contaminated by high-frequency muscle artifact, the parameters were determined from the spectrum between 0.5 and 25 Hz. Due to the definition of relative beta ratio, higher frequencies were used exceptionally in the calculation of this parameter.

Total spectral, delta, theta, alpha, and beta powers were determined from the spectrograms. The frequency bands were 0.5–25 Hz for total spectral power, 0.5–4 Hz for delta power, 4–8 Hz for theta power, 8–12 Hz for alpha power, and 12–25 Hz for beta power. These parameters were chosen because the electroencephalographic activity in different frequency bands changes during the induction of anesthesia obeying a characteristic pattern.2,4,20

Relative delta, theta, alpha, and beta powers were calculated by dividing the power in the corresponding frequency band by the total spectral power. The relative powers represent classic parameters for depth-of-anesthesia estimation and were therefore included in this study.21,22

SE is one of the recently proposed measures for the assessment of depth-of-anesthesia. It quantifies the flatness of the power spectrum and is calculated as

where *f ^{l}* is the lower and

*f*the higher limit of the used frequency band, and where

^{h}*P*is the normalized

^{n}power spectrum of the signal. *N ^{f}* is the number of frequency components in the range [

*f*,

^{l}*f*] and used for the normalization of the values of SE between 0 and 1. In this study,

^{h}*f*and

^{l}*f*were 0.5 Hz and 25 Hz, respectively. With an appropriate frequency range, SE has been shown to decrease rather monotonically with increasing depth-of-anesthesia.23The parameter has been used in the M-Entropy® module (GE Healthcare Finland Oy).6

^{h}RBR is a subparameter used in the calculation of Bispectral Index.4,24It is the logarithm of the ratio of the electroencephalographic spectral power in 30- to 47-Hz band to the power in 11- to 20-Hz band:

*MPF* , *SEF90%* , and *SEF95%* represent classic parameters measured during anesthesia and have been under active research.4,21Spectral edge frequency X% is the frequency below which X% of the power in the spectrum resides. MPF corresponds to spectral edge frequency 50%.

To reduce noise in data visualization, the underlying trends of the spectral parameters were approximated with curve fitting. Since the parameters were assumed not to follow any specific pattern, the parametric fitting could not be used. Instead, a nonparametric cubic smoothing spline was applied to the data.25The spline smoothing is illustrated in figure 3A. Furthermore, the median curves of smoothed spectral parameters were determined for each group as shown in figure 3B.

### Statistical Analysis

The group differences in the values of 14 spectral parameters were statistically compared with two approaches. First, the values of each spectral parameter were compared between two specific groups continuously in different *r* values. The comparison was performed separately in all *r* values between 0 and 2 with an interval of 0.01. Since the BSP generally occurs approximately when *r* = 2,19the analysis was restricted to that value. The data were not assumed to follow normal distribution, and the comparison was performed with a nonparametric Mann–Whitney U test. *P* < 0.05 was considered statistically significant. Second, the overall effect of remifentanil on the spectral parameters was analyzed separately during light (0 < *r* < 1) and deep (1 < *r* < 2) anesthesia. For this, the spectral parameter median curves (fig. 3B) were used. However, the median curves were determined from the unsmoothed spectral parameter values (no spline smoothing) to avoid statistical bias. The spectral parameter median curves of groups R1 and R2 were compared to group R0 curves with Wilcoxon signed-rank test. The Bonferroni correction was used to adjust for multiple comparisons so that the overall criterion for rejection of the null hypothesis was *P* < 0.05. Thus, because all fourteen spectral parameters were compared between groups R0 and R1 and groups R0 and R2 during light and deep anesthesia, only *P* values less than 0.05/56 were considered significant.

The statistical analysis was performed with the statistics toolbox for Matlab® (The Mathworks Inc.).

## Results

### Spectrograms

Figure 4illustrates the median spectrograms of different groups from *r* = 0 to *r* = 2. The median spectrograms are created by choosing the group's median value at every time-frequency point. The differences between the median spectrograms of two specific groups are given as well. The median spectrograms differ markedly between groups. The following three changes can be clearly related to the increasing dose of remifentanil: (1) decrease of activity in high frequencies (> 14 Hz) during light anesthesia (0 < *r* < 1); (2) increase of activity in extended alpha band (7–14 Hz) during deep anesthesia (1 < *r* < 2); (3) decrease of activity in delta band (0.5–4 Hz) during deep anesthesia (1 < *r* < 2). These changes can be related to the coadministration of remifentanil, as they appear in a dose-dependent manner.

The median spectrogram of group R1 contains relatively strong high-frequency (greater than 30 Hz) activity during very light anesthesia (*r* < 0.5) compared to other groups. Further inspection of the electroencephalograms showed that this activity was electromyographic artifact that occurred most frequently in group R1. Because these artifacts were present also before the drug infusion began, they were considered not to be related to the administration of remifentanil.

### Spectral Parameters

The 14 median spectral parameter curves determined for each group from the time-normalized spectrograms are illustrated in figure 5. The figure also shows the results of the statistical comparison of the spectral parameters continuously in different *r* values. The group differences in spectral parameter median curves analyzed separately during light and deep anesthesia are given in table 2.

The power parameters of different frequency bands follow the changes seen in the spectrograms. In delta band, the activity decreases when the opioids are coadministered. The values of group R0 and R2 are significantly different in deep anesthesia in figure 5. Table 2shows a decrease of 66%. As a result of the dominant effect of delta activity on total spectral power, the curves of these two parameters resemble each other. In theta and beta bands, the activities do not differ statistically significantly between groups in figure 5, even though the beta activity is clearly suppressed in R1 and R2 during light anesthesia. According to table 2, the decrease is 26% in R1 and 38% in R2. The alpha activity is significantly increased in remifentanil groups during deep anesthesia. As shown in table 2, remifentanil has more than doubled the activity in this band.

The changes in spectrograms also reflect on the relative powers in different frequency bands, even though the effects in this case are more complex and require a thorough analysis. For example, in figure 5, the theta activity did not differ significantly between groups when the spectral power was analyzed, but it does with the relative power. This can be explained by the fact that, although remifentanil does not affect the theta activity itself, changing the activity in other frequencies influences the relative power parameter. Therefore, due to the lack of delta activity, the relative theta power is significantly higher in remifentanil groups during deep anesthesia. However, this does not explain why the relative theta power is significantly higher in group R2 during very light anesthesia; the delta activity does not differ there between groups. As with the delta power parameter, the high-dose remifentanil group shows significantly lower values for relative delta power during deep anesthesia in figure 5. The values of this parameter have decreased 29% (table 2). Again, due to the presence of strong delta activity, the relative alpha and beta powers are significantly lower in group R0 during deep anesthesia (fig. 5). Naturally, the increased activity in extended alpha band also influences the relative alpha power parameter in remifentanil groups. The total increase of this parameter is 179% and 293% (table 2) in groups R1 and R2, respectively.

In SE, figure 5shows no significant differences between groups during light anesthesia, although the curve of R2 seems to be slightly decreased. However, during deep anesthesia, the values in R1 and R2 are significantly higher compared to that of R0. For example, in group R2, SE has increased 15% (table 2). This can be explained by the lack of delta activity, due to which the spectrum is more flat in remifentanil groups.

Relative beta ratio was the only parameter in which higher frequencies (greater than 25 Hz) were used. Due to the electromyographic artifact, the values are significantly higher in group R1 during very light anesthesia. The artifact does not explain, however, the decreased values in high-dose remifentanil group after *r* = 1. At that point, the activity in the higher band (30–47 Hz) has ceased, and the lower band (11–20 Hz) activity dominates the values of the parameter. Hence, the lower values in group R2 come from the increased activity in extended alpha band (7–14 Hz) during deep anesthesia. However, as the changes occur only within a narrow range of *r* (1–1.5), the change reflected on the values in table 2are moderate.

The parameters MPF, SEF90%, and SEF95% show also differences between study groups. The suppression of high frequencies during light anesthesia results in a significant decrease of SEF90% and SEF95% in high-dose remifentanil group (fig. 5). All three parameters indicate consistently also that the electroencephalographic activity in saline group is in significantly lower frequencies during deep anesthesia. SEF90%, for example, increases from approximately 7 Hz to 12 Hz when the saline is changed to high-dose remifentanil. This corresponds to an increase of more than 50%, as presented in table 2.

## Discussion

The effect of remifentanil on the time-frequency properties of electroencephalogram during propofol-induced anesthesia was studied. The results show that remifentanil significantly changes the signal's spectral content in a characteristic manner, including decrease of beta activity during light anesthesia as well as decrease of delta and increase of alpha activity during deep anesthesia. The changes were strongly dependent on the level of anesthesia and reflected on the quantitative spectral parameters used in the depth-of-anesthesia estimation. For example, during deep anesthesia, SE and SEF90% were clearly increased by remifentanil; in delta activity, however, a major decrease was observed. Coadministration of opioids therefore challenges the reliability of the spectral properties of electroencephalogram in the depth-of-anesthesia estimation by using a frontal montage, which may have an impact on the indices using these properties.

As mentioned in the introduction, the changes remifentanil solely produces to electroencephalogram, *i.e.* , progression of activity from high to low frequencies, resemble somewhat the propofol-induced changes. This easily leads to the hypothesis that the effects of these two drugs on the electroencephalogram would be synergistic. However, our study shows that instead of increasing the propofol-induced changes, the coadministered remifentanil suppresses some of them (beta activity during light anesthesia and delta activity during deep anesthesia) and induces some of its own (alpha activity during deep anesthesia). According to the quantitative spectral parameters, these changes might lead to an impression of a deeper anesthesia in low *r* values (*e.g.* , lower SEF90% and beta power), but on the other hand a lighter anesthesia in high *r* values (*e.g.* , higher SE and lower delta power).

This study deepens our understanding of the total impact of remifentanil on electroencephalogram-based depth-of-anesthesia estimation. In our previous work,13we found that the infusion of remifentanil during propofol anesthesia significantly modifies the mutual relations of the electroencephalographic spectral characteristics and the clinical endpoints in a predictable and quantifiable manner. Even though it was shown that the propofol-induced basic frequency progression pattern of electroencephalogram is left untouched, this study shows that also the detailed spectral characteristics are significantly affected by remifentanil and that this phenomenon is strongly dependent on the depth-of-anesthesia. This finding further complicates the reliable usage of spectral properties of electroencephalogram for depth-of-anesthesia estimation when remifentanil is coadministered.

In the literature, the effects of remifentanil on the electroencephalogram during anesthesia have not been presented adequately before. The studies have concentrated on describing the response of the depth-of-anesthesia indices, mostly Bispectral Index, to the coadministration of remifentanil during steady-state anesthesia. The results have been controversial, with some studies suggesting the indices to be affected by the coadministration of opioids,12,14and others failing to find an effect.26,27One possible explanation for the controversy is that the effect of remifentanil on electroencephalogram is not stable or stationary during different levels of anesthesia. As presented in this paper, the nature of the electroencephalographic changes remifentanil produces seems to be strongly dependent on the phase of the propofol-induced frequency progression pattern. It is likely that the effects of remifentanil differ significantly between two steady-states, making the comparison of the results difficult without an objective control parameter of anesthetic depth. In future, the research should concentrate more on describing the effect of opioids on electroencephalogram throughout the different levels of anesthesia instead of presenting them only for example in a single steady-state condition.

Compared to the previous studies, several improvements were made in this paper. First, the effect of remifentanil on electroencephalogram is presented during the whole process of induction of anesthesia. As a result, the effect of the opioid can be seen in different levels of anesthesia and not only in a single steady-state. Second, instead of examining the effects of remifentanil by using only a few quantitative parameters of electroencephalogram, we illustrate the changes induced in the entire spectrum and how these changes reflect on the parameters. This approach clarifies the underlying spectral phenomenon that causes the changes in the parameters. Third, by applying the time normalization to the spectrograms, we were able to reduce the interindividual variability in response to the anesthetic agent and made the comparison of the data of different patients reasonable as a function of time.

We hypothesize that the effects of remifentanil on the electroencephalogram presented in this study can be generalized to some extent to all drugs with similar pharmacodynamic properties, *i.e.* , μ-receptor agonists. The studies with other opioids (fentanyl, alfentanil, and sufentanil) support this proposition because these drugs have been shown to induce similar electroencephalographic changes with remifentanil during anesthesia.12,28However, due to the difference in pharmacokinetic profiles, the beginning and duration of the effect may alter significantly. Furthermore, remifentanil is considered to be a specific μ-receptor agonist; therefore, the electroencephalographic changes induced by the drugs acting also through other receptor may differ in part from that of remifentanil.

Interestingly, tramadol also induces similar electroencephalographic changes as remifentanil during anesthesia. Tramadol is a centrally acting analgesic occasionally used during operation to prevent postoperative pain. Lately there has been debate about its use during anesthesia. When coadministered with an anesthetic, tramadol has been shown to induce electroencephalographic changes that can be associated with lightening of anesthesia.29,30These changes, including decrease of delta activity and increase of alpha activity and SEF95%, are identical to the effects of remifentanil during deep anesthesia reported in this paper. In addition to the affinity for opioid receptors, tramadol appears to have actions on the noradrenergic and serotonergic systems.31It has been suggested that the effect on electroencephalogram results from the α-adrenergic activity.29However, the changes resemble closely that of remifentanil; we therefore speculate that they are induced by opioid receptor activation. This is supported by the finding that tramadol does not affect electroencephalogram in the presence of remifentanil.32The drugs acting through the opioid receptors are known to have synergistic effect with anesthetics to the clinical state of the patients26,33; therefore, the electroencephalographic changes tramadol induced may in fact be related to increased clinical sedation.

Findings from the recent studies propose also new directions for future research. An ideal depth-of-anesthesia monitor accurately relates the electroencephalographic recordings to the clinical state of patient. The current indices have been developed to fulfill this task with a single anesthetic agent. However, this paper points out that when coadministered with propofol, the opioids change the electroencephalogram in a rather complex manner that potentially affects the reliability of the indices used in depth-of-anesthesia estimation. One possibility to overcome this problem would be detection of the presence of opioid from the electroencephalogram. This way, the observed changes in the signal could more reliably be related to the clinical state of patient. Based on the results of this study, we can expect this to be possible to some extent. Furthermore, we have studied the effect of remifentanil only from the Fz electrode; this raises the question of whether some other electrode location would be more immune to the coadministration of opioids. It would also be of interest to study whether the effects would be similar with other anesthetics. Recently, several nonlinear measures, such as approximate entropy and permutation entropy, have been proposed for the measurement of depth-of-anesthesia.34–37In the future, the effect of remifentanil on these measures should be tested as well.

## References

*versus*alfentanil: Comparative pharmacokinetics and pharmacodynamics in healthy adult male volunteers. Anesthesiology 1996; 84:821–33