Abstract
Background. Power spectral analysis is a well-established method for the analysis of EEG signals. Spectral parameters can be used to quantify pharmacological effects of anaesthetics on the brain and the level of sedation. This method, in numerous variations, has been applied to depth of anaesthesia monitoring and has been incorporated into several commercially available EEG monitors. Because of the importance of EEG spectral analysis, we evaluated the performance of each frequency in the power spectrum regarding detection of awareness.
Methods. Ninety artefact-free EEG segments of length 8 s were obtained from a database that contains perioperatively recorded EEG data. For the present analysis, EEG data were selected from 39 patients with propofol–remifentanil or sevoflurane–remifentanil anaesthesia with a period of awareness. Half of the EEG segments were recorded during periods of awareness as defined by an adequate response to the command ‘squeeze my hand’. The other half were from unresponsive patients. The power spectral density was calculated for each segment. The performance of each frequency bin of the power spectrum as a detector of awareness was assessed with a remapped prediction probability rPK, i.e. the prediction probability PK mapped to a range of 0.5–1.
Results. The remapped prediction probability was high (rPK>0.8) for low frequencies (<15 Hz) and for high frequencies (>26 Hz), with a minimum (rPK<0.55) at 21 Hz. Indentations in the ‘performance spectrum’ occur at the power-line frequency (50 Hz) and its harmonics and at 78 Hz, probably caused by the continuous impedance measurement of another device used in parallel. With the exception of the indentations, the remapped prediction probability of the high frequencies (>35 Hz) was >0.95.
Conclusions. The best performance for the detection of awareness was achieved by EEG power spectral frequencies from >35 Hz up to 127 Hz. This frequency band may be dominated by muscle activity. The frequency band between 15 and 26 Hz may be of limited value, as reflected by lower rPK values.
Spectral analysis is one of the standard methods used for quantification of the EEG. The power spectral density (power spectrum) reflects the ‘frequency content’ of the signal or the distribution of signal power over frequency. Several parameters derived from the power spectrum have been used, including total power, spectral band power, and median and spectral edge frequency. Several EEG-based monitors for assessment of depth of anaesthesia are (at least in part) based on spectral analysis of the EEG. The bispectral index (BIS, Aspect Medical Systems, Newton, MA) is in part based on the beta ratio, the spectral power of two subcomponents of the beta band.1 The Patient State Index (PSI, Physiometrix Inc., North Billerica, MA) is based on the fronto-occipital relation of spectral power components of the EEG. The Datex Entropy Module (Datex-Ohmeda Division, Instrumentarium Corporation, Helsinki, Finland) calculates state entropy (SE) and response entropy (RE), which are both based on entropy analysis of the power spectrum of EEG and frontal EMG. The SNAP Index (VIASYS Healthcare Inc., Conshohocken, PA), which has been withdrawn from the market, was also based on spectral analysis of classical EEG and higher-frequency components.2
In the present study we examined which parts of the EEG power spectrum are most useful for discrimination between awareness and responsiveness. Usually, the power spectrum is analysed on the basis of broader frequency bands, which represent the sum of power of several smaller frequency bands (bins). In the present analysis, we used a different approach and calculated a ‘performance spectrum’. This spectrum represents the performance of each single EEG power spectral frequency bin to discriminate between awareness and unresponsiveness. The spectral band width was 1 Hz for each frequency bin tested.
Methods
We analysed a set of perioperatively recorded EEG data from a database containing EEG and standard monitoring parameters of patients and volunteers recorded during sedation and anaesthesia.3 The data were from a study with 40 patients who underwent elective surgery under general anaesthesia.4 This study received institutional ethics committee approval and patients gave written consent. Exclusion criteria were rapid sequence induction, medication with CNS-affecting drugs, pregnancy, and psychiatric or neurological diseases. Blockwise randomization was performed to assign the patient to one of the following groups: (1) sevoflurane–remifentanil (≤0.1 µg kg−1 min−1); (2) sevoflurane–remifentanil (≥0.2 µg kg−1 min−1); (3) propofol–remifentanil (≤0.1 µg kg−1 min−1); (4) propofol–remifentanil (≥0.2 µg kg−1 min−1). Monitoring parameters were continuously recorded and included non-invasive blood pressure, heart rate, oxygen saturation, oxygen, carbon dioxide and sevoflurane concentrations and respiratory variables. At EEG electrode positions, the skin was prepared with alcohol to maintain impedances <5 kΩ. A two-channel referential EEG was recorded from electrode positions AT1, AT2, Fz (reference) and Fp1 (ground) using the A-1000 EEG monitor (BIS version 3.3, Aspect Medical Systems Inc., Newton, MA, USA). The high pass was set to 0.25 Hz, no low pass was used, and the notch filter (50 Hz) was enabled. The EEG was continuously digitized at 256 Hz per channel and simultaneously recorded on a personal computer. Unpremedicated patients received remifentanil infusion at either 0.1 or 0.2 µg kg−1 min−1 according to the group assignment. Patients were tested for responsiveness every 30 s. with a repeated command to squeeze an investigator's hand placed in the patient's hand. A repeated (i.e. verified) response to command was classified as ‘awareness’, whereas absence of a response was classified as ‘unresponsiveness’. Anaesthesia was induced with sevoflurane by inhalation through a facemask (groups 1 and 2) or by intravenous propofol injections (0.7 mg kg−1, followed by 20 mg every 30 s [groups 3 and 4]). Loss of consciousness (LOC1) was defined as the time when the patient stopped squeezing the hand on command. Following LOC1, additional propofol or sevoflurane was administered to increase the level of hypnosis. A blood pressure cuff on the right arm was inflated and maintained for 5 min to occlude the circulation of the right forearm and retain the ability to move the arm to command while succinylcholine (1.0 mg/kg) was given for intubation of the trachea (Tunstall's isolated forearm technique).5 Then, while remifentanil infusion was continued, sevoflurane or propofol was stopped until patients gave a verified response to command (return of consciousness [ROC1]). Sevoflurane or propofol bolus injection, followed by continuous infusion, was then recommenced. Loss of consciousness 2 (LOC2) was noted when patients stopped following commands to squeeze the hand. Remifentanil was administered with the predefined remifentanil infusion rates. During surgery, sevoflurane and propofol were maintained according to clinical practice. At the end of surgery, commands to squeeze the hand were given every 30 s. Sevoflurane, propofol and remifentanil were discontinued and return of consciousness (ROC2) was noted at the first verified (i.e. repeated) response to command.
Results
Owing to an error in the file structure of a data file, only 39 recordings could be included in the analysis. The plot of rPK values (Fig. 1a) illustrates that frequencies of ∼20 Hz have the lowest performance in detecting awareness. Frequencies below this minimum show a negative correlation between frequency bin power and the detection of awareness (PK<0.5), and frequencies above the minimum show a positive correlation (PK>0.5) (Fig. 1b).
Performance of single-frequency bins between 1 and 127 Hz as detector of awareness. (a) rPK, a modified PK with values remapped to the interval between 0.5 and 1 to obtain the performance of the parameters as a detector of awareness independent of the negativity or positivity of the correlation. (b) Original PK values, showing that frequencies >21 Hz decrease with awareness, whereas higher frequencies increase.
Discussion
The results demonstrate that frequencies between 36 and 127 Hz show a high prediction probability (rPK>0.95) for the separation of ‘awareness’ and ‘unresponsiveness’. The selection of more than one data pair per patient may have increased the overall PK. On the other hand, the selection of only one data pair per patient would have ignored within-patient variability. As a consequence of the current selection, data pairs may not be truly independent. We have accepted this limitation, as it has been accepted from the introduction of PK analysis.7 In addition, the main focus of the present analysis is the relation between the different frequency bins, which is maintained.
We found indentations for the power-line frequency of 50 Hz and its harmonic at 100 Hz. An additional indentation is located at 78 Hz, and may be caused by the continuous impedance check performed by the Physiometrix PSA-4000 device used in parallel with the recording A-1000 device. This suggests that signal quality may severely decrease when other devices interfere. Most likely, these indentations at 50 and 100 Hz are caused by power-line interference present in the signal. Some of the EEG segments used revealed an increased power of the 50 Hz component relative to its neighbouring frequency bins and thus indicated the presence of 50 Hz interference.
The high performance of higher-frequency components suggests that facial muscle activity may have influenced or possibly dominated the results, as EMG activity during awareness is higher than during unresponsiveness. None of our patients had been under complete neuromuscular block. Under the given circumstances, we cannot reliably differentiate between EEG and EMG activity because the EEG gamma band and the EMG frequency range overlap. As a consequence, the present results do not indicate whether the EEG or the EMG are the reason for the good performance of high frequencies.
Classical EEG analysis is usually performed in frequency bands below 30 Hz, i.e. the delta, theta, alpha and beta bands. The results of our study show that frequencies in the beta band perform quite differently. An increase of the lower part of the beta band (<21 Hz) indicates an increasing probability of unresponsiveness, whereas an increase of the higher part of the beta band (>21 Hz) indicates an increasing probability of awareness. Owing to this oppositional behaviour it may be reasonable to split the beta band into two parts when analysing beta-band power.
Several EEG monitors of depth of anaesthesia also use higher-frequency bands. The BIS components Beta Ratio and Sync Fast Slow use frequencies up to 47 Hz.1 Beta Ratio incorporates frequencies between 11 and 20 Hz (i.e. mainly beta activity) and between 30 and 47 Hz (i.e. gamma activity). The SNAP index, which has now been withdrawn from the market, was calculated from a low-frequency band (0.1–40 Hz) and a high-frequency band.2 The frequency range 40–80 Hz is omitted because it may be dominated by EMG. However, our study did not show differences between the frequency ranges 40–80 Hz and 80–127 Hz. The Datex Entropy Module calculates state entropy (SE) from the EEG-dominated 0.8–32 Hz frequency band and the response entropy (RE) from the 0.8–47 Hz frequency band, which includes EMG activity.8 These examples show that higher-frequency components (>30 Hz) of EEG or EMG may be useful for depth of anaesthesia monitoring. Currently, it is uncertain whether the high-frequency signal components reflect activity of the main target organ of anaesthesia, the brain (EEG), or just an indirect and unspecific parameter, i.e. muscle activity (EMG). Further studies are required to evaluate the performance of these components during complete or varying neuromuscular blockade. The results presented here indicate that the high-frequency components are well suited to detect awareness, but also that the performance to detect awareness may be limited when artefacts are present.
This study was supported by B. Braun Melsungen AG, Germany.
References
Wong CA, Fitzgerald PC, McCarthy RJ. Comparison of depth of anesthesia indices (SNAP vs Bispectral) during balanced general anesthesia in patients undergoing outpatient gynecologic surgery. www.asa-abstracts.com
Ningler M, Schneider G, Stockmanns G, Schäpers G, Kochs E. Datenbank zur Unterstützung studienübergreifender Auswertungen von Signalen für das Narkosemonitoring.
Schneider G, Gelb AW, Schmeller B, Tschakert R, Kochs E. Detection of awareness in surgical patients with EEG-based indices—bispectral index and patient state index.
Tunstall ME. Detecting wakefulness during general anaesthesia for caesarean section.
Smith WD, Dutton RC, Smith NT. Measuring the performance of anesthetic depth indicators.
Leslie K, Sessler DI, Smith WD, et al. Prediction of movement during propofol/nitrous oxide anesthesia. Performance of concentration, electroencephalographic, pupillary, and hemodynamic indicators.
Viertiö-Oja H, Meriläinen P, Paloheimo M, et al. Optimization of response time of spectral EEG entropy enables early warning of emergence from unconsciousness. Abstracts of the 5th International Conference on Memory, Awareness and Consciousness
Comments
Dear Sir,
with interest I read the above-mentioned article by Dressler et al. I wonder whether or not the reader's attention is drawn to the significance of their findings for clinical practice.
This I thought after reading what clinical observation exactly was correlated with EEG readings in Dressler's study. The authors claim to detect awareness, however, closely looked at, clinical awareness was only defined as a form of response to verbal command.
In clinical practice of theatre anaesthesia it certainly would be far more informative to find out which interpretations of EEG readings (if any) would ascertain the absence of repsonse, not to verbal command but, to noxious stimuli since no anaesthetist would dare rely on absence of response to verbal command in order to proceed with anaesthetic manipulations on the patient. Why should we infer from the useful analysis of EEG interpretation in the state of unresponsiveness to verbal command to the a clinically relevant analysis of EEG interpretation in the state of unresponsiveness to noxious stimuli? Vanluchene et al (1) found no correlation between several EEG monitoring modes and response to noxious stimuli.
Not only in the light of Vanluchene A L G et al's article (1) in this journal it is unlikely that the findings of Dressler's above-mentioned article will signify a breakthrough in the attempt to invent a technical monitor which provides data on which to rely is safer than clinical observation in anaesthetic practice.
With kindest regards,
Dr R t Wörden, D.E.A.A. Hamburg
1 Vanluchene A L G et at; BJA 93(5):645-54 (2004)
Conflict of Interest:
None declared