^{1}

^{2}

^{2}

^{3}

^{4}

^{1}

^{2}

^{*}

^{1}

^{2}

^{3}

^{4}

Edited by: Agnes Gruart, Pablo de Olavide University, Spain

Reviewed by: Anders Ledberg, Stockholm University, Sweden; Giancarlo Vanini, University of Michigan, United States

*Correspondence: Jamie Sleigh

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Oscillations in the electroencephalogram (EEG) at the alpha frequency (8–12 Hz) are thought to be ubiquitous during surgical anesthesia, but the details of how this oscillation responds to ongoing changes in volatile anesthetic concentration have not been well characterized. It is not known how often alpha oscillations are absent in the clinical context, how sensitively alpha frequency and power respond to changes in anesthetic concentration, and what effect increased age has on alpha frequency. Bipolar EEG was recorded frontally from 305 patients undergoing surgery with sevoflurane or desflurane providing general anesthesia. A new method of detecting the presence of alpha oscillations based on the stability of the rate of change of the peak frequency in the alpha range was developed. Linear concentration-response curves were fitted to assess the sensitivity of alpha power and frequency measures to changing levels of anesthesia. Alpha oscillations were seen to be inexplicably absent in around 4% of patients. Maximal alpha power increased with increasing volatile anesthetic concentrations in half of the patients, and decreased in the remaining patients. Alpha frequency decreased with increasing anesthetic concentrations in near to 90% of patients. Increasing age was associated with decreased sensitivity to volatile anesthesia concentrations, and with decreased alpha frequency, which sometimes transitioned into the theta range (5–7 Hz). While peak alpha frequency shows a consistent slowing to increasing volatile concentrations, the peak power of the oscillation does not, suggesting that frequency might be more informative of depth of anesthesia than traditional power based measures during volatile-based anesthesia. The alpha oscillation becomes slower with increasing age, even when the decreased anesthetic needs of older patients were taken into account.

Oscillations in the human electroencephalogram (EEG) in the alpha frequency band (8–12 Hz) were first reported by Berger (

Alpha oscillations during anesthesia are thought to be ubiquitous (Gugino et al.,

One of the most common measures of the alpha signal is the magnitude of power over the alpha range (usually 8–12 Hz). The problem with this simple measurement is that it will conflate both the underlying broad-band noisy activity (1/

During the transition into and out of general anesthesia, the frequency of the alpha oscillation shifts according to anaesthetic concentration, as observed by Long et al. (

The aims of this study are therefore threefold:

To note the incidence of alpha oscillatory activity in a clinical population receiving a volatile anesthetic.

To characterize changes in the oscillatory components of alpha power (peak, broadband and oscillatory alpha power), alpha frequency, and the spectral gradient to changing volatile anesthetic concentration.

To characterize the effect of age on the relationship between alpha frequency and volatile anesthetic concentration.

This observational study was approved by the New Zealand Health and Disability Ethics Committee (Ref. 12/CEN/56), and all patients gave informed written consent before being included. Physician anesthetists were allowed to provide anesthetic care according to their own clinical judgment. Volatile gas anesthetic (VGA) concentrations at the end of each breath cycle (end-tidal) were recorded every 5 s from the S5 Anesthesia Monitor using the S5 Collect program (both from GE Healthcare, Helsinki, Finland) in units of Minimum Alveoli Concentration (MAC) to allow comparison between different gas types. Brain effect-site VGA concentrations were estimated using a delay model with a half-time equilibrium constant (_{eo}) of 144 s as reported in McKay et al. (_{e}MAC. VGA concentrations were adjusted for age according to norms set out in Nickalls and Mapelson (

Frontal bipolar EEG (with FP_{Z} as reference, and either FP_{1} or FP_{2} as active electrode) was recorded from 305 patients undergoing surgery at the Waikato District Health Board in Hamilton, New Zealand using either a Bispectral Index^{®} (BIS^{®}, from Aspect Medical Systems, Newton, MA, USA, with a sampling rate of 128/s) or Entropy (GE Healthcare, Helsinki, Finland, sampling rate: 100/s) anesthetic depth monitor. Researchers interested in analyzing data from this database can submit requests online at

A Fast-Fourier Transform (FFT) was used to calculate power spectral densities, using the multi-taper method developed by Thomson (

As the precise relationship between narrowband alpha power and the underlying broadband spectral gradient during changing levels of anesthesia is not known, we used the convention set out in Leslie et al. (

_{e}MAC

When a genuine sustained oscillation in the alpha range is present, peak alpha power will be consistently higher than the spectral gradient, and peak frequency will be centered on a stable frequency. If the peak alpha frequency suddenly shifts, this indicates that peak frequency has jumped from one oscillatory peak to another within the 7–17 Hz allowable range, and that the oscillatory alpha power is within the range of noise on the spectral gradient. In this way, a maximal value of the rate of change in peak frequency can be used as a threshold to assess if a sustained oscillation is present or not. Note that if the peak frequency of the oscillation is not stable at a given frequency, a value of oscillatory alpha power will still be given. In this instance the oscillatory alpha power value will be a measure of the noise in the spectrum around the spectral gradient and thus depend on the time window length and frequency resolution chosen.

Practically,

Peak alpha frequency was measured from the spectrum (Figure

The first derivative of the peak alpha frequency (the rate of change of peak frequency per second) was calculated (Figure

When a threshold of 1 standard deviation of change in peak frequency was breached, sustained oscillatory alpha activity was considered absent (Figure

Individual concentration-response slopes were calculated by fitting a linear robust regression (using a bisquare weighting function with a tuning constant of 4.685) of peak alpha frequency against C_{e}MAC (Figure _{e}MAC also shown in

Concentration-response relationships were also fitted for peak (Figure

Our analysis began at start of surgery and ended at the beginning of gas flush, and any period of EEG showing burst suppression did not enter the concentration-response regression. All correlations were calculated using Pearson’s Correlation Coefficient.

Of the original 305 patients, 19 patients were excluded as dropouts, 18 rejected due to problems with VGA recordings, and 10 excluded as they received a non-volatile based anesthesia (propofol).

Of the 258 patients included for analysis, patient ages ranged from 18 years to 90 years (median 64 years, interquartile range 27 years). Patient gender was split evenly (128 females, 130 males), and the majority of patients received sevoflurane (188, 73%) with the remaining receiving desflurane. A wide range of surgical disciplines entered the analysis, the most prevalent being general surgery (41%), followed by vascular surgery cases (26%), with gynecology and urology (25%). The remaining surgery types (8%) were plastics, thoracic, with one ENT and neurological case each. Median operation length was 71 min (IQR 104 mins), and the median C_{e}MAC over the operation for all patients was 0.96 (IQR 0.22). Median effect-site opioid concentration during the operation was 0.81 ng/ml fentanyl-equivalents (IQR 0.74).

An ongoing oscillation in the alpha range was not always present; 48 patients (19%) showed alpha activity for less than half of the surgical anesthesia period. Retrospective _{e}MAC concentrations during the operation (0.99 vs. 0.95, _{e}MAC was age-adjusted (1.18 vs. 1.07 C_{e}MAC,

In our analysis, concentration-response curves were only fitted to C_{e}MAC values when an alpha oscillation was detected. Patients may have received higher C_{e}MAC concentrations over the operation than the concentrations where alpha was observed. In this instance (where the maximal concentration over the operation was higher than the maximal concentration where alpha was observed) the maximal C_{e}MAC value of the concentration-response fitting represents the limit at which alpha oscillations occur for that patient. This is in contrast to where alpha oscillations were observed right up until the highest C_{e}MAC value that patient received over the operation; in this instance the maximal C_{e}MAC could simply represent the maximal concentration the anesthetist was willing to give to that patient, and not the natural concentration limit of the alpha oscillation _{e}MAC more than the maximal C_{e}MAC of the concentration-response regression. In this patient subgroup (median age 73 years, IQR 20 years) the median maximal C_{e}MAC values of the concentration-response regression was 1.01, with an inter-quartile range of 0.29 C_{e}MAC. The mean maximal age-adjusted C_{e}MAC value was 1.17 (IQR 0.30) and as such represents an estimate of the upper volatile anesthetic concentration limits where the alpha oscillation occurs.

The following results focus on the sign of the concentration-response slope, where a positive concentration-response slope indicates an increase in frequency or power, and a steepening of the spectral gradient to increasing anesthetic concentration, a negative concentration-response slope a decrease in the same measures to increasing anesthetic concentration. For the concentration-response analysis the 48 patients with alpha activity for less than 50% of the maintenance anesthesia period were excluded. A further 46 patients were also rejected due to having only a very small C_{e}MAC range (0.15 C_{e}MAC during surgery), which made fitting the linear regression unreliable. This _{e}MAC range less than this value had non-physiological concentration-response slope values. C_{e}MAC values are non-age-adjusted unless otherwise mentioned. There was no statistically significant effect of type of volatile anesthetic on the concentration-effect slopes for alpha frequency, any alpha power measures, or spectral gradient (all

The mean patient C_{e}MAC where an alpha oscillation was present was 0.84, and the mean peak alpha frequency at this C_{e}MAC concentration was 9.2 Hz. The mean concentration-response slope for alpha frequency was −2.8 Hz/C_{e}MAC (SD 2.7), indicating a near to 0.5 Hz frequency slowing with each 0.2 C_{e}MAC increase in anesthetic concentration. Eighty-eight percent of patients (145/164) had a negative concentration-response slope for alpha frequency. Figure

_{e}MAC. Scatterplots of concentration-response slopes for peak alpha frequency against: peak alpha power

The mean peak alpha power at mean C_{e}MAC (0.84 MAC) was 12.4 dB. An even distribution of positive and negative concentration-response slopes for peak alpha power were noted to increasing anesthetic concentration (Figure _{e}MAC (SD 6.8). Eighty-six (52%) patients increased peak alpha power with increasing anesthetic concentration, whilst 78 (48%) showed the opposite—a decrease. Magnitude of the peak alpha power-C_{e}MAC concentration-response slope was not associated with age, maximal or minimal C_{e}MAC, maximal or minimal peak alpha frequency, mean opioid concentration (all

The mean broadband alpha power at mean C_{e}MAC was 3.3 dB. The mean concentration-response slope for broadband alpha power was 6.8 dB/C_{e}MAC (SD 6.4). One-hundred and fourty-four patients (88%) had a positive concentration-response slope for broadband power (see Figure

Given that the majority of patients (also 88%) showed a decrease in alpha frequency to increasing C_{e}MAC, we would expect broadband alpha power to increase to increasing C_{e}MAC simply due to the point of measurement sliding up the negative spectral gradient (see Figure _{e}MAC was −0.92 dB/Hz, and the mean concentration-response slope for spectral gradient was −0.19 dB/Hz/C_{e}MAC (SD 0.61). Two-thirds of patients (110/164) had negative concentration-response slopes, indicating an increasing steepness of the spectral gradient to increasing C_{e}MAC, and one third (54/164) had positive slopes. The concentration-response slope of the spectral gradient was not correlated with age (

The mean oscillatory alpha power at mean C_{e}MAC was 8.9 dB. The mean concentration-response slope for oscillatory alpha power was −5.7 dB/C_{e}MAC (SD 6.1). One-hundred and fourty-one patients (86%) had a negative concentration-response slope for oscillatory alpha power (Figure

The maximum and minimum peak alpha frequency of the concentration-response curves both decreased with age (_{e}MAC values of the concentration response curves also decreased with age (_{e}MAC concentrations with age result in a left shift of the concentration-response curves.

_{e}MAC _{e}MAC

When age-adjusted C_{e}MAC values were used, minimum C_{e}MAC did not change with age (_{e}MAC values of the concentration response curves now increased with age (_{e}MAC values are used.

In summary, in most patients, peak alpha frequency decreases in response to increasing anesthetic concentration. While peak alpha power can either increase or decrease with increasing anesthetic concentration, broadband alpha will usually increase, resulting in a decreasing oscillatory alpha power. The increase in broadband alpha power is not simply a mathematical consequence of a slide up the negative spectral gradient with slowing alpha frequency, but is contributed to in some degree by an increasing steepness of the spectral gradient itself with increasing anesthetic concentration.

This study reveals new findings concerning the incidence, concentration-responsiveness, and effect of age on frontal oscillatory alpha activity during volatile based clinical anesthesia.

A frontal alpha oscillation under anesthesia was inexplicably absent in around 4% of our patients. This rate is slightly lower than some older studies looking at occipital alpha in awake patients with eyes closed (summarized in Vogel and Götze,

That alpha power only increases to increasing anesthetic concentration in half our patients is a novel finding. Most previous studies have focussed on the induction or emergence period where large changes in anesthetic concentration and behavioral response are guaranteed (e.g., Purdon et al., _{e}MAC range where alpha was observed were not correlated with sign of the concentration-response curve slope, and discounts this biphasic effect as an explanation. Inspection of the concentration response figures showed that the more traditional use of a sigmoid fitting was not warranted for the alpha power measure; in a few cases (around 8 patients), alpha power saturated at a maximal level, and the use of a sigmoid would have been more suitable, but the sign of the concentration-response slope would not have changed, and is very unlikely to change the results.

That alpha power increases for only half of our patients may also be due to a limitation of spectral analysis, such as the effect of a non-sinusoidal shaped waveform. An increase in peak spectral power could be caused by an alpha waveform increasing in amplitude, or may also be due to an alpha waveform with the same amplitude becoming more tightly sinusoidal. Alternative methods, such as wavelets, would be needed to tease out these subtle effects of wave morphology.

In contrast to alpha power, the slowing of alpha frequency to increasing anesthetic concentration in the majority of patients is a clear finding, and suggests that frequency might be more informative of depth of anesthesia than traditional power measures during volatile-based surgical anesthesia.

Slowing of the occipital alpha frequency can also be seen in response to subtle physiological changes such as decreased temperature (e.g., Chang et al.,

Regarding the possible mechanisms of alpha slowing, Hughes and Crunelli (

A decrease in alpha frequency under sevoflurane anesthesia with increasing age has been previously noted by Purdon et al. (

The spectral gradient itself was sensitive to anesthetic concentration in two-thirds of our patients. We chose to use a log-linear fit for the spectral gradient as this historically had a better fit to the data than the log-log fit (Sleigh et al.,

A number of observations in this study also have relevance to any depth of anesthesia measures based on the alpha waveform. While the indexes of the two current primary commercial monitors (the Bispectral Index, or BIS^{®}, and Entropy Module from GE Healthcare) are not based specifically on the alpha waveform, some newer proposals are (for example Sleigh et al.,

To conclude, in this study we have demonstrated that clear oscillatory alpha activity was absent in near to 5% of our clinical population, and that while peak alpha frequency shows a consistent slowing to increasing volatile gas concentration during surgical anesthesia, the peak power of the oscillation does not, only increasing in around half of the patient group. The underlying broadband spectral gradient became steeper with increasing concentration in two-thirds of patients. We have also shown that the alpha oscillation becomes slower with increasing age, even when the decreased anesthetic needs of older patients were taken into account.

DH completed EEG recordings and analysis and wrote the manuscript. LJV, PSG and JS wrote the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding for this project was provided by the James S McDonald Foundation (Grant Award No. 220020346 to PSG). The results were independently derived and do not reflect any endorsement on the part of the James S. McDonnell Foundation. PSG’s research efforts are supported in part by a Career Development Award #BX00167 (PI: PSG) from the United States Department of Veteran Affairs, Biomedical Laboratory Research and Development Service. Many thanks also to Matthias Kreuzer for providing the burst-suppression detection code, and to Joel Winders for help with completing the EEG recordings.

The Supplementary Material for this article can be found online at: