Both entropy1 (GE Healthcare, Fairfield, Connecticut, USA) and the bispectral index (BIS, Covidien, Dublin, Ireland)2,3 are claimed to be good measures of the hypnotic component of anaesthesia. The reliability of EEG-based indices such as these to indicate the transition between different anaesthetic levels remains unclear. In addition, there may be differences between calculated index values.
The present study examines the performance of state entropy (Entropy Module) against BIS.1 State entropy is based on the spectral entropy, which is obtained by the application of Shannon entropy to the power spectrum. Hence, state entropy characterises the frequency distribution calculated from frontal EEG (0.8 to 32 Hz). For optimum response time, the length of the analysis window is individually adapted for each particular frequency, resulting in the so-called ‘time–frequency balanced spectral entropy’.4,5 A more detailed description of the algorithm is provided by Viertiö-Oja et al.5 and Särkelä et al.6 For both state entropy and BIS, the manufacturers recommend identical index ranges for all levels of anaesthesia.
Using identical EEG data, the present study evaluates prediction probability (PK) of state entropy and BIS to discriminate state transition from consciousness to unconsciousness and vice versa. So far it is not known to what extent the two monitors display similar index values when they process identical EEG signals. Information about the comparability of the indices is lacking. The method of using recorded EEG that is replayed simultaneously to both devices offers the opportunity to analyse causes of disagreement between clinical assessment of the hypnotic state (gold standard) and corresponding index values. The study also provides information about the underlying raw (time domain) EEG signals and thus allows evaluation of the potential influence of distinct EEG findings/artefacts on index calculation; indices may show misleading values due to physiological EEG pattern [e.g. beta or delta arousal, pathophysiological features (e.g. epileptiform pattern, electro-oculogram (EOG), electromyogram (EMG)] or technical artefacts (e.g. from electrode contact or cautery). In addition, if the EEG is influenced by disease, indices may be misleading. If two indices disagree, one of them or both can be ‘wrong’ due to these reasons. The purpose of this study is to evaluate how these distinct patterns affect the indices in an experimental, non-clinical setting and compare the performance with two anaesthetic regimens.
The study protocol was approved by the ethics committee of the Technische Universität München (ethical committee number 461/01, signed on 6 February 2001, Munich, Germany). Forty un-premedicated patients with an American Society of Anesthesiologists’ physical status of I or II scheduled for elective surgery under general anaesthesia gave informed written consent and were enrolled in the study (Table 1).7
The original dataset of variables from EEG and auditory-evoked potentials (AEPs) was evaluated to identify a period of wakefulness during anaesthesia.7 A two-channel referential EEG was recorded using an amplifier designed for AEP/EEG recordings during anaesthesia.8 The EEG was recorded from ZipPrep electrodes (Aspect Medical Systems, Newton, Massachusetts, USA) placed according to the manufacturer's recommendations and the international 10 to 20 system: at the left temporal region between the lateral edge of the eye and the upper edge of the ear (AT1), in the middle of the forehead (Fpz, reference), and on the left side of the forehead (F7, ground). Impedances were below 5 kΩ. The EEG was continuously digitalised using a sampling rate of 1 kHz. EEG filter settings were 0.5 Hz (high pass) and 400 Hz (low pass). Standard monitoring variables and EEG output were recorded using a personal computer.
Study design and clinical protocol
Blocked randomisation was performed to enrol 20 patients to either a sevoflurane/remifentanil (sevoflurane) group or a propofol/remifentanil (propofol) group. Anaesthesia was slowly induced and patients were asked every 30 s to squeeze the investigator's hand to determine the clinical endpoints of loss of consciousness (LOC) and return of consciousness (ROC). LOC1 was defined as a missing response to a repeated verbal command. After tracheal intubation, sevoflurane or propofol was stopped until patients followed command again (ROC1) during an episode of intended wakefulness. Tunstall's isolated forearm technique9 was used to preserve the ability to follow command during ROC1 before succinylcholine (1 mg kg−1) was given. After ROC1, sevoflurane inhalation or propofol bolus injection was recommenced until LOC2. After LOC2, anaesthesia was administered according to standard clinical practice and surgery was performed. The end of surgery was followed by emergence from general anaesthesia, and ROC2 was identified by the first verified squeeze of the hand.
Re-analysis of electroencephalogram data
EEG data were replayed simultaneously to the Entropy Module and an Aspect A-2000 monitor (Aspect Medical Systems, with XP technology with DSC-2, three-electrode set up and BIS smoothing rate of 15 s) using a digital-to-analogue converter constructed by the research group.10 The EEG player used two galvanically separated channels; an impedance check at one monitor could not affect the signal at the electrodes of the other monitor. We did not deactivate the automatic impedance check from either monitor, although it would have reduced the number of invalid data pairs. This was to ensure that the connections between the Entropy Module and Aspect A-2000 monitors and the EEG player were adequate for the entire procedure.
The sampling rate of the output was identical to the sampling rate of the recorded EEG. No additional low-pass filtering (smoothing) was performed. Signal frequency, amplitude and phase characteristics were not changed.10 A possible change in signal characteristics may have been induced by filter settings (−3 dB attenuation) during online recording. The influence of potential signal change in the slow delta and sub-delta frequencies is considered below. The EEG index values of state entropy and BIS were calculated from identical EEG signals and stored on a personal computer. Entropy values were recorded using S/5 L-Collect software (GE Healthcare, Helsinki, Finland), and BIS values were polled every 5 s using a serial hyperterminal connection (Version 5.1, Microsoft Corporation, Redmond, Washington, USA).
Continuous data were compared using Student's t-test, and categorical data with the χ2 test (two-sided tests with unequal variances, for detailed results we refer to previously published data).7 Agreement between the index values of BIS and state entropy was tested with Bland–Altman analysis using R.11
Noting the manufacturers’ recommendations for thresholds of EEG index ranges, state entropy and BIS do not always agree with clinical evaluation, displaying index values more than 80 in patients clinically assessed as unconscious and index values less than 60 in responsive (conscious) patients.
PK calculation of state entropy was based on EEG index values at anaesthetic-induced state transitions recorded 30 s before and after LOC or ROC. The interval of 30 s around hand squeezing allows a clear definition of clinical endpoints LOC/ROC, and thus the conclusion that 30 s before ROC (LOC) the patient was unconscious (conscious). State entropy and BIS values 30 s after LOC/ROC were used for PK analysis to ascertain whether the EEG monitor indicated the current clinical state.12 This approach is consistent with previously published data.7,13 The same conditions were applied to the Entropy Module.
PK values were obtained by using the PK tool that was designed by our research group to calculate overall PK values and confidence intervals (CIs).14
Pearson's correlation between state entropy and BIS was investigated during monitoring depth of anaesthesia (DOA) from light to deep hypnosis for valid data pairs of state entropy and BIS collected every 10 s during the EEG analysis period (including all anaesthetic levels).
A visual analysis of ‘raw’ time domain EEG signals from the periods of state transition was performed. All available EEG sequences that underlie the respective endpoints for the EEG-based index calculation of the conscious and unconscious state were analysed. The 40 EEG recordings include 160 state transitions between LOC or ROC (four transitions – LOC 1 and 2, ROC 1 and 2 – per recording). Three hundred and twenty EEG sequences were analysed: 160 sequences from the endpoint consciousness (LOC − 30 s, ROC + 30 s) and 160 sequences from the endpoint unconsciousness (ROC − 30 s, LOC + 30 s). The main focus of this analysis was the raw time domain signals of a 60-s time window preceding the respective endpoints for the EEG-based index calculation. Clinical assessments of consciousness and unconsciousness were performed with the isolated forearm technique. Disagreement between clinical state and index values was defined as misclassification – the respective index values were classified as erroneous.
The EEG sequences were searched for the occurrence of the most relevant criteria: physiological EEG pattern (e.g. beta or delta arousal), pathophysiological features such as epileptiform pattern (spikes, multi-spikes, sharp waves, spike and waves, periodic epileptiform discharges), different types of EOG, EMG or other technical artefacts, for example from electrode contact or cautery.
The following methodical approach was used to evaluate EEG findings: all EEG findings were considered to have potential for misleading the monitors to calculate index values that disagreed with clinical assessment. As an overall approach, the occurrences of these artefacts were evaluated in EEG signals, which resulted in index values that disagreed with clinical examination, and vice versa in EEG signals that were consistent with clinical assessment. Next, the potential influence of artefacts was separately analysed during EEG periods recorded during consciousness (i.e. artefacts which may bias a monitor towards ‘unconsciousness’) and unconsciousness (i.e. artefacts which may bias a monitor towards ‘consciousness’).
Possible differences in frequency ranges were analysed with Fast Fourier Transformation (0.25 to 60 Hz) and Spectral Edge Frequency (SEF95, calculated from 0.49 to 47.1 Hz).
Patient data are shown in Table 1.
PK analysis at the transition between consciousness and unconsciousness
A detailed description of data excluded from PK and the correlation analysis can be found in the supplemental digital content (SDC, https://links.lww.com/EJA/A62). PK values of state entropy were calculated as an overall value separately for LOC and ROC events, and for the two anaesthetic regimens (Table 2). The overall PK was 0.80 (95% CI, 0.74 to 0.84) for state entropy, with best results for the propofol group at LOC and LOC along with ROC (Table 2). BIS achieved an overall PK of 0.84 (95% CI, 0.79 to 0.88).
Correlation and agreement between state entropy and bispectral index during change from deep to light anaesthesia
The correlation coefficient for BIS and state entropy was 0.86 in the sevoflurane and 0.68 in the propofol group (overall 0.78). The Bland–Altman plot11,15 in Fig. 1 illustrates the agreement between state entropy and BIS during DOA monitoring from deep to light anaesthesia. Major differences are caused mainly by the different reaction times of the monitors as they follow changes in the anaesthetic state.16
Disagreement between electroencephalogram indices and clinical assessment at state transition
Two hundred and ninety-nine BIS and 300 state entropy index values were obtained. Table 3 gives the number of erroneous state entropy and BIS values 30 s before and after transition between consciousness and unconsciousness, and specifies the distribution of erroneous index values in the sevoflurane and propofol groups.
Statistical analysis is limited by an overall low number of index values that disagree with clinical evaluation. Therefore, only a quantitative analysis was performed. Results are given as observed frequency: 28 (9%) erroneous BIS and 41 (14%) state entropy values. For both monitors, misclassification occurred mainly at recent changes in clinical status: 71% of all misclassifications occurred after state transition, in particular 54% after LOC. In total, there were more erroneous state entropy values, mainly for the unconscious state after LOC and in the sevoflurane group.
Analysis of the raw electroencephalogram time domain signal
EEG sequences were assigned to four groups: first, EEG signals from conscious patients that were correctly assessed by the respective DOA monitor (correct conscious); second, EEG signals from conscious patients that were falsely assessed by the respective DOA monitor (false conscious); third, EEG signals from unconscious patients classified as ‘correct unconscious’; and fourth, as ‘false unconscious’. ‘Correct’ assessment by a DOA monitor was defined as being consistent with clinical assessment, and ‘false’ assessment as being inconsistent. Three independent experts, blind to the corresponding results of the DOA monitors, visually analysed 283 EEG sequences. In the event of conflict, a two-third majority was required.
Ideally, analysis of the EEG time domain signal helps to identify sources (e.g. physiological EEG pattern or pathophysiological features) that may be responsible for creating errors in the DOA monitors. This was difficult to achieve in the present study because there was a low overall incidence of erroneous index values, but conversely an overall high incidence of high-frequency signals or eye blinks in the EEG that underlay index values that were either consistent or inconsistent with clinical assessment. Thus, the present analysis allows only a descriptive overview: correct (false) index values resulted from EEG that contained high-frequency signals in at least 47% (75%) and eye blinks in at least 28% (33%) (Table 4a and b). The highest incidence was found for three categories of artefacts: high-frequency signals, eye blinks and epileptiform potentials.
An exact separation of physiological high-frequency signals into EMG, high-frequency EEG and EOG (e.g. fast eye fluttering or saccades) often seemed impossible and was not attempted. The frequency of other artefacts that were assignable, for example technical artefacts and movement artefacts, was negligible and therefore excluded from analysis. The sub-types of EOG that were classifiable were mostly eye blinks.
Visual inspection of the EEG from clinically verified conscious patients was unremarkable. High-frequency signals and eye blinks were more frequently found in EEG sequences underlying correct index values (correct conscious, Table 4a and b).
There was a trend towards a higher incidence of high-frequency signals and eye blinks (Table 4a and b) in the EEG of unconscious patients that were falsely interpreted as conscious (false unconscious, Fig. 2a and b). Correct BIS (state entropy) values resulted from EEG that showed high-frequency signals in 58% (47%) and/or eye blinks in 33% (28%). False BIS (state entropy) values had high-frequency signals in 81% (96%) and/or eye blinks in 56% (56%). Both DOA monitors were affected. By trend, erroneous index values from EEG with high-frequency signals and/or eye blinks were more often calculated by state entropy than by BIS. Vice versa, the EEG underlying correct state entropy values contained these findings less often than the EEG underlying correct BIS values.
Figure 3 illustrates the percentage of artefact-contaminated EEG sequences in conscious and unconscious patients during propofol and sevoflurane anaesthesia: the EEG sequences of unconscious patients showed by trend a higher incidence of artefacts during sevoflurane anaesthesia compared with propofol anaesthesia. The index values calculated from the EEG of conscious patients had only a very low overall incidence of disagreement (Table 3). Thus, the results from differentiation in anaesthetic groups (Fig. 3) may not provide reliable results.
Overall, the EEG sequences underlying erroneous index values during sevoflurane anaesthesia showed a higher rate of high-frequency signals (96%) and eye blinks (57%) compared with propofol anaesthesia with, respectively, 73% and 27% – particularly during unconsciousness (Table 5). Unfortunately, the absolute number of erroneous index values in both anaesthetic groups was not sufficient for a further differentiation in BIS and state entropy index errors.
The overall incidence of epileptiform potentials was low with a maximum of 3% (Table 4a and b). For a quantitative analysis, all sub-categories were summarised to evaluate an overall influence of artefacts on EEG signal processing: epileptiform potentials occurred predominantly in EEG sequences that resulted in correct index values (Fig. 2c). Thus, they do not represent significant interference for EEG processing by either monitor in our data.
Electroencephalogram frequency analysis: Spectral Edge Frequency 95 and relative frequency bands
Calculation of SEF95 and frequency bands involved the 141 EEG sequences 30 s before and after the state transition from consciousness to unconsciousness (LOC), and vice versa, the ROC. The SEF95 – calculated from the underlying EEG of false BIS and state entropy index values – showed the characteristic biphasic reaction during induction, but was higher before and declined more slowly compared to index values, which correctly indicate LOC (Fig. 4a). This trend was more pronounced during sevoflurane anaesthesia. At ROC, the SEF95 was lower for the EEG underlying false BIS and state entropy values compared with SEF95 from the EEG of correct indices (Fig. 4b), and lower again for the EEG during propofol anaesthesia.
EEG sequences for index values that falsely indicate consciousness (after clinically verified LOC) show a higher ratio of relative beta and gamma band power compared with those of index values that correctly indicate unconsciousness (Fig. 5). By trend, the EEG gamma band is more often affected by sevoflurane, whereas the EEG beta band is more affected by propofol. The EEG sequences of erroneous index values feature a higher alpha and lower delta band power compared with those of correct index values. EEG sequences for index values (after clinically verified ROC) that falsely indicate unconsciousness show a lower ratio of relative beta and gamma band power (Fig. 5), and a higher relative delta band power, compared with those of index values that correctly indicate consciousness.
In summary, differences between the EEG of erroneous BIS and state entropy values are not distinct – this may be due to the limited number of erroneous index values, in particular after ROC (Table 3).
BIS is calculated from an EEG signal derived from forehead and temple electrodes; accordingly, the EEG is prone to EMG contamination,17 for example during arousal or nociceptive stimuli. The numerical index value of BIS does not provide information about the influence of EMG on the underlying EEG signal. The Entropy Module includes EMG by analysing the facial biosignal as a surrogate measure. The information is given by the corresponding spectral entropy variable, response entropy,5 but it is not assumed to be more informative than state entropy, particularly with regard to the fast detection of intra-operative wakefulness when neuromuscular blocking agents are used.18 This analysis focused on cortical measures and excluded response entropy.
The current evaluation of the performance of state entropy and BIS is based on a challenging dataset. A study design with slow induction and emergence from anaesthesia, and EEG analysis close to the transition between consciousness and unconsciousness, represents a high requirement for a DOA monitor. BIS seems to perform slightly better during state transitions than state entropy. A re-analysis of the same data with the Cerebral State Monitor (Danmeter, Odense, Denmark) resulted in a PK of 0.75 (Cerebral State Monitor).13 Unfortunately, a direct comparison of PK values is not possible due to the different number of valid data pairs.
The ‘dream of clinical anaesthesiologists’ as expressed by Drummond19 is that index performance should be independent of the anaesthetic agent. Few studies have used more than one anaesthetic group,20–22 but we used two different anaesthetic regimens in the analysis, revealing that this goal was not entirely met. So far, no commercially available EEG-based monitor is able to satisfy these criteria.7,20,22–27 Drummond's postulate seems to assume one single measure for the different mechanisms and effects of varying anaesthetics. Given the variety of anaesthetic effects,28 this seems unlikely but not impossible.
Comparability of DOA monitors in previous studies is limited due to methodological differences.20,21,26,29–32 The present comparison is based on the analysis of identical underlying EEG data without simultaneous EEG recording. A bias towards BIS or state entropy was avoided because anaesthetic administration was not guided by one of the EEG-based monitors (e.g. BIS as reference value). The EEG player has been designed as a method of replaying stored EEG data.10 The EEG was recorded from all positions recommended for use of BIS and Entropy Module with an EEG/AEP device, developed in a European multi-centre trial (BIOMED), that introduces no proprietary filters that may alter the EEG signal.8 The original EEG was recorded with the device using passband 0.5 to 400 Hz. The cut-off frequency typically refers to the frequency where −3 dB attenuation is reached. As a result, all frequencies below 1 Hz may be attenuated and the recorded EEG signal may contain less delta and sub-delta activity than the signal originally measured at the patient's skin. According to the manufacturer, the passband for BIS is 0.16 to 450 Hz, and 0.5 to 118 Hz for Entropy. Both algorithms use frequencies below 1 Hz.3,5 As a consequence, the state entropy and BIS values obtained in our study may be higher than during online recording. The effect of this theoretical systematic error may not be equal for BIS and state entropy. For example, Laitio et al.33 reported high-amplitude delta activity during xenon anaesthesia, where state entropy values were also lower than BIS (medians 18 vs. 26). This may suggest that power in the EEG delta band has a greater influence on state entropy than on BIS. In the present analysis, state entropy showed a higher incidence of disagreement with clinical examination during LOC than BIS (Table 3) and relative delta band power was lower in the misclassified cases. It may be that state entropy puts more weight on delta band power, representing a possible limitation in the comparison between the two indices.
The EEG data were replayed to the BIS monitor via a BIS 3.4 sensor. The newer BIS XP sensor has an additional electrode that should help identify artefacts by comparing symmetries and asymmetries between two EEG channels. Our BIS version may have had weaker artefact detection and hence performance compared with the four-electrode sensor system. Nevertheless, the presented results are valid for comparison with state entropy.
A limitation of other methodological approaches that assess spectral entropy together with BIS during general anaesthesia in adults20,21,24,29–31,34–40 is that simultaneous recordings or analysis of different underlying EEG signals are used for comparison. Several studies have shown considerable disagreement in simultaneous EEG recordings.24,34 Even two identical BIS XP monitors displayed conflicting results in more than 10% of all recordings.41
In the present study, disagreement between clinical status and the corresponding index values (with thresholds according to manufacturers’ recommendations) at state transitions was identified and the underlying raw EEG was screened for potential causes. Most difficulties in indicating the correct clinical state, for BIS and state entropy, were found shortly after the transition between consciousness and unconsciousness. A higher incidence of disagreement occurred at LOC for both monitors, in particular for state entropy during sevoflurane anaesthesia. Our results from EEG frequency analysis may suggest that the algorithms of the BIS monitor and Entropy Module depend on high frequencies more than 30 Hz to calculate indices in the recommended ranges of consciousness; the EEG underlying index values that falsely indicate (un)consciousness showed a higher (lower) relative gamma band power compared with the EEG sequences underlying index values that agreed with clinical examination. Currently no comparable research data are available.
Visual analysis of our EEG data revealed that the most common artefacts were physiological in origin, consisting of high-frequency signals and eye blinks. The a posteriori identification of underlying sources of physiological high-frequency signals (EMG, high-frequency EEG or EOG, e.g. eye fluttering or saccades) from a single EEG channel may not be very reliable42 and was not used. Instead, analysis centred on the main question as to whether high-frequency signals resulted in erroneous index values, with failure to indicate the correct clinical state. The cause of these, whether it is EMG, EEG or high-frequency EOG, may not be critical. The EEG has a narrow amplitude spectrum and is prone to an overlap with artefacts, for example with the EMG or eye blinks during lighter levels of anaesthesia (LOC and ROC). In particular, there is interference in the frequency ranges of EMG and high-frequency EEG (gamma band). There is controversy about the contribution of EMG recorded from the scalp to the EEG gamma band (EEG frequencies more than 30 Hz), and subsequently to the analysis of anaesthetic depth. The detection of the gamma band by DOA monitors is aggravated by sources of interference (e.g. EMG, scalp acting as low pass filter). Separation from myogenic electrical activity has proved difficult.43 Some commercially available DOA monitors can assess the data contained in the EEG gamma band, but because most algorithms have not been published, our interpretation must remain speculative. The reliability of DOA monitors which depend on high-frequency signals to indicate consciousness may be limited with the use of neuromuscular blocking agents.17
One of the most profound threats for DOA monitoring that is based on EEG analysis is EMG generated by the activity of pericranial muscles.44 The following EMG characteristics contribute to difficulties for EEG analysis: high-amplitude, broad spectral and anatomical distributions and sensitivity to psychologically interesting processes. Furthermore, the EMG is characterised by lack of stereotypy; even spurious EMG artefacts can disguise or influence the signal across virtually the entire EEG spectrum. The most common notion is that the EMG power spectrum peaks occur at relatively high frequencies (30 to 100 Hz), but EMG signals can indeed overlap with all EEG frequencies, even as low as 2 Hz.45 This explains why changes in neurogenic and myogenic activity are often confused – sometimes with substantial consequences. Shackman et al.46 reported that alpha-blocking associated with eye-opening was weakened even when there was no high degree of EMG contamination. The individual differences of the components of EMG make high demands on the filters and tools needed to correct EMG artefacts. In summary, there is no effective solution for eliminating the influence of EMG on EEG-based analysis – even if the analysed frequency spectrum is restricted to less than 30 Hz.47,48
High-frequency signals were present in the predominant part of our EEG data but caused fewer problems for the two monitors during consciousness. High-frequency signals recorded during unconsciousness may have more often led to misclassification. In clinical practice, a patient who is mistakenly assumed to be awake might receive an unnecessary anaesthetic overdose. This may be less of a problem than patients who are not correctly identified as conscious by DOA monitors, with risk of intra-operative wakefulness and possible long-term psychological consequences.
Aho et al.49,50 investigated the influence of EMG on the behaviour of state and response entropy in two clinical studies. First, they visually verified the appearance of EMG from the original biosignal, and also from its spectral presentation in the EEG of patients under general anaesthesia without neuromuscular blockade.49 When the behaviour of entropy was analysed in the presence of EMG due to noxious stimuli, the strong EMG signal began to influence the EEG signal that was already below 20 Hz. The authors assumed that it was the EMG that was responsible for the unexpected behaviour of state entropy during nociceptive stimuli. Second, they investigated the EEG, EMG, and Entropy values before and after skin incision, and also the effect of rocuronium on entropy and EMG at skin incision during general anaesthesia. Skin incision causes simultaneous EMG and EEG arousal, but sometimes only EEG arousal was detected.50 Both beta and EMG arousals increased state entropy and response entropy. The power spectra of EEG and EMG overlap significantly at frequencies of 20 to 50 Hz. The authors concluded that when numerical values mislead, the raw signal is required for the correct interpretation. The design of these studies does not permit comparison with ours, not least because we did not use noxious stimuli. In general, they confirm the influence of EMG on the reliability of state entropy – notably the falsely high index values that can result.
The artefact produced by an ‘eye blink’ is seen in the EEG of virtually every conscious individual. Eye blinks were present in approximately one-third of all EEG sequences from unconscious patients, and in 50% of those conscious. By trend, both DOA monitors (more by state entropy than BIS) misinterpreted the EEG when eye blinks occurred, more in unconscious than conscious patients.
A verified response to a simple command was used to indicate consciousness. The present approach is dichotomous; the patient is either responsive or not. Unresponsiveness is classified as unconsciousness. The short interval of 30 s between the first response and asking the patients to squeeze the hand allows an analysis that is close to the transition between consciousness and unconsciousness. This is different from a ‘fully awake’ or a ‘fully anaesthetised’ state. During these lighter levels of anaesthesia, eye blinks may frequently occur in unresponsive patients.
The overall higher rate of erroneous state entropy values suggests that state entropy is less reliable than BIS in correctly indicating the clinical state during general anaesthesia, in particular identifying unconsciousness. It may be assumed that high-frequency signals and/or eye blinks interfere with signal processing and mislead both monitors into indicating index values in the recommended ranges for consciousness. The higher incidence of high-frequency signals and eye blinks in the EEG underlying erroneous state entropy values suggests that they have a higher influence on the index calculation for the Entropy Module than for the BIS monitor. A substance-specific influence of the anaesthetic agent on the EEG cannot be excluded. Epileptiform patterns may be more frequent with sevoflurane,51,52 and muscle artefact more frequent with propofol, even in deep anaesthesia.42,49,50 The present analysis shows high-frequency signals and/or eye blinks that seem more frequent in the sevoflurane than in the propofol group, particularly during EEG sequences from unconscious patients that gave index values indicating consciousness. In general, this might be due to epileptiform patterns during sevoflurane anaesthesia, but the visual EEG analysis did not confirm this. The sevoflurane mask induction in this study may have led to a higher rate of excitation and movement artefacts and thus entail a higher incidence of EMG and/or eye blinks during LOC. Also, emergence from sevoflurane anaesthesia may involve a higher risk of excitation than emergence from propofol anaesthesia. Unfortunately, the absolute number of erroneous index values in the anaesthetic groups was insufficient for a comparison between BIS and state entropy.
The interpretation of these results must remain speculative. The overall incidence of erroneous index values was too low for a statistically significant quantitative analysis, and descriptive analysis does not permit a definite conclusion. Because the exact algorithm of BIS calculation has not been revealed in detail by the manufacturer, it is not known how the configurations of BIS and state entropy, including filter settings, influence signal processing. This makes it impossible to evaluate the impact of artefacts on the EEG signal or to ascertain if these are causal for erroneous index calculation. Unfortunately, most artefacts, in particular the EMG and EOG, are not characterised by a stereotypical response. Inevitably, a uniform explanation for failure of DOA monitors seems unlikely.
Our results confirm that EEG recordings of conscious patients are often contaminated with artefacts, in particular high-frequency signals, including EMG and eye blinks. These artefacts were discussed as possible causes for aberrant classification of the hypnotic level. Beyond that, DOA monitors may have other flaws. Those that apply frequency-based methods, such as spectral analyses in the bispectrum (BIS) or spectral entropy (state entropy) need stability of the EEG over a longer analysis period. Analysis of a time series assumes that the data are stationary. In statistical terms this means that the mean, variance and structure do not change with time. The term that describes this is ‘stationarity’. Index calculation of BIS and state entropy requires a time interval up to 60 s,3,5 but the EEG signal reflects highly complex cortical activity, particularly during lighter levels of anaesthesia, and is assumed to be predominantly non-stationary. Kreuzer et al.53 determined the duration of EEG sequences that can be considered stationary at different anaesthetic levels of sevoflurane and propofol anaesthesia. They revealed possible conflicts between the EEG segment length used for index calculation and availability of the required duration of stationary signal episodes, in particular during lighter levels of anaesthesia, wakefulness and LOC. For conscious patients, EEG stationarity was maintained for sequences only up to 12 s and this increased with increasing DOA. Furthermore, EEG stationarity during LOC was strongly influenced by the anaesthetic used, with a significantly higher probability for static sequences with propofol than sevoflurane.
In summary, there was a trend towards a higher incidence of high-frequency signals and eye blinks during sevoflurane compared with propofol anaesthesia. A higher probability of non-stationary EEG sequences from conscious patients of the sevoflurane group53 may account for the higher rate of false index values of BIS and state entropy. High-frequency signals and eye blinks were present in the majority of all EEG sequences and may in particular account for index values that falsely indicate consciousness. Compared with BIS, state entropy gave more false classifications of the clinical state at transition between consciousness and unconsciousness. Results of the present study support the view that EEG indices should always be interpreted in context and with other clinical signs.
Acknowledgements relating to this article
Assistance with the study: none.
Financial support and sponsorship: this study was funded from departmental sources and a grant (KKF) from the Technische Universität München.
Conflicts of interest: none.
2. Bruhn J, Bouillon TW, Shafer SL. Bispectral index (BIS) and burst suppression: revealing a part of the BIS algorithm. J Clin Monit Comput
3. Rampil IJ. A primer for EEG signal processing in anesthesia. Anesthesiology
4. Shannon C. A mathematical theory of communication. Bell System Techn J
5. Viertiö-Oja H, Maja V, Sarkela M, et al. Description of the entropy algorithm as applied in the Datex-Ohmeda S/5 entropy module. Acta Anaesthesiol Scand
6. Särkelä M, Mustola S, Seppänen T, et al. Automatic analysis and monitoring of burst suppression in anesthesia. J Clin Monit Comput
7. Schneider G, Hollweck R, Ningler M, et al. Detection of consciousness by electroencephalogram and auditory evoked potentials. Anesthesiology
8. Jordan C, Weller C, Thornton C, Newton DE. Monitoring evoked potentials during surgery to assess the level of anaesthesia. J Med Eng Technol
9. Tunstall ME. Detecting wakefulness during general anaesthesia for caesarean section. Br Med J
10. Kreuzer M, Kochs EF, Pilge S, et al. Construction of the electroencephalogram player: a device to present electroencephalogram data to electroencephalogram-based anesthesia monitors. Anesth Analg
11. R Development Core Team. R: A language and environment for statistical computing. Vienna: R Development Core Team; 2006.
12. Zanner R, Pilge S, Kochs EF, et al. Time delay of electroencephalogram index calculation: analysis of cerebral state, bispectral, and Narcotrend indices using perioperatively recorded electroencephalographic signals. Br J Anaesth
13. Pilge S, Blum J, Kochs EF, et al. Does the cerebral state index separate consciousness from unconsciousness? Anesth Analg
14. Jordan D, Steiner M, Kochs EF, Schneider G. A program for computing the prediction probability and the related receiver operating characteristic graph. Anesth Analg
15. Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res
16. Kreuzer M, Zanner R, Pilge S, et al. Time delay of monitors of the hypnotic component of anesthesia: analysis of state entropy and index of consciousness. Anesth Analg
17. Messner M, Beese U, Romstock J, et al. The bispectral index declines during neuromuscular block in fully awake persons. Anesth Analg
18. Kawaguchi M, Takamatsu I, Kazama T. Rocuronium dose-dependently suppresses the spectral entropy response to tracheal intubation during propofol anaesthesia. Br J Anaesth
19. Drummond JC. Monitoring depth of anesthesia: with emphasis on the application of the bispectral index and the middle latency auditory evoked response to the prevention of recall. Anesthesiology
20. Kaskinoro K, Maksimow A, Langsjo J, et al. Wide inter-individual variability of bispectral index and spectral entropy at loss of consciousness during increasing concentrations of dexmedetomidine, propofol, and sevoflurane. Br J Anaesth
21. Vakkuri A, Yli-Hankala A, Talja P, et al. Time-frequency balanced spectral entropy as a measure of anesthetic drug effect in central nervous system during sevoflurane, propofol, and thiopental anesthesia. Acta Anaesthesiol Scand
22. Hocker J, Raitschew B, Meybohm P, et al. Differences between bispectral index and spectral entropy during xenon anaesthesia: a comparison with propofol anaesthesia. Anaesthesia
23. Schneider G, Kochs EF, Horn B, et al. Narcotrend does not adequately detect the transition between awareness and unconsciousness in surgical patients. Anesthesiology
24. White PF, Tang J, Romero GF, et al. A Comparison of state and response entropy versus bispectral index values during the perioperative period. Anesth Analg
25. Dahaba AA. Different conditions that could result in the bispectral index indicating an incorrect hypnotic state. Anesth Analg
26. Vanluchene AL, Struys MM, Heyse BE, Mortier EP. Spectral entropy measurement of patient responsiveness during propofol and remifentanil. A comparison with the bispectral index. Br J Anaesth
27. Avidan MS, Jacobsohn E, Glick D, et al. Prevention of intraoperative awareness in a high-risk surgical population. N Engl J Med
28. Lee U, Ku S, Noh G, et al. Disruption of frontal-parietal communication by ketamine, propofol, and sevoflurane. Anesthesiology
29. Lysakowski C, Elia N, Czarnetzki C, et al. Bispectral and spectral entropy indices at propofol-induced loss of consciousness in young and elderly patients. Br J Anaesth
30. Bonhomme V, Deflandre E, Hans P. Correlation and agreement between bispectral index and state entropy of the electroencephalogram during propofol anaesthesia. Br J Anaesth
31. Iannuzzi M, Iannuzzi E, Rossi F, et al. Relationship between Bispectral Index, electroencephalographic state entropy and effect-site EC50 for propofol at different clinical endpoints. Br J Anaesth
32. Mahon P, Greene BR, Greene C, et al. Behaviour of spectral entropy, spectral edge frequency 90%, and alpha and beta power parameters during low-dose propofol infusion. Br J Anaesth
33. Laitio RM, Kaskinoro K, Sarkela MO, et al. Bispectral index, entropy, and quantitative electroencephalogram during single-agent xenon anesthesia. Anesthesiology
34. Schmidt GN, Bischoff P, Standl T, et al. Comparative evaluation of the Datex-Ohmeda S/5 Entropy Module and the Bispectral Index monitor during propofol-remifentanil anesthesia. Anesthesiology
35. Ellerkmann RK, Liermann VM, Alves TM, et al. Spectral entropy and bispectral index as measures of the electroencephalographic effects of sevoflurane. Anesthesiology
36. Aime I, Verroust N, Masson-Lefoll C, et al. Does monitoring bispectral index or spectral entropy reduce sevoflurane use? Anesth Analg
37. Lefoll-Masson C, Fermanian C, Aime I, et al. The comparability of bispectral index and state entropy index during maintenance of sufentanil-sevoflurane-nitrous oxide anesthesia. Anesth Analg
38. Schultz A, Siedenberg M, Grouven U, et al. Comparison of Narcotrend Index, Bispectral Index, spectral and entropy parameters during induction of propofol-remifentanil anaesthesia. J Clin Monit Comput
39. Rinaldi S, Consales G, De Gaudio AR. State entropy and bispectral index: correlation with end tidal sevoflurane concentrations. Minerva Anestesiol
40. Ellerkmann RK, Soehle M, Alves TM, et al. Spectral entropy and bispectral index as measures of the electroencephalographic effects of propofol. Anesth Analg
41. Niedhart DJ, Kaiser HA, Jacobsohn E, et al. Intrapatient reproducibility of the BISxp monitor. Anesthesiology
42. Aho AJ, Yli-Hankala A, Lyytikainen LP, et al. Can electromyographic arousal be detected visually on the Datex-Ohmeda S/5(TM) Anesthesia Monitor? Acta Anaesthesiol Scand
43. Nottage JF, Morrison PD, Williams SC, Ffytche DH. A novel method for reducing the effect of tonic muscle activity on the gamma band of the scalp EEG. Brain Topog
44. McMenamin BW, Shackman AJ, Greischar LL, Davidson RJ. Electromyogenic artifacts and electroencephalographic inferences revisited. NeuroImage
45. Goncharova II, McFarland DJ, Vaughan TM, Wolpaw JR. EMG contamination of EEG: spectral and topographical characteristics. Clin Neurophysiol
46. Shackman AJ, McMenamin BW, Slagter HA, et al. Electromyogenic artifacts and electroencephalographic inferences. Brain Topogr
47. McMenamin BW, Shackman AJ, Maxwell JS, et al. Validation of ICA-based myogenic artifact correction for scalp and source-localized EEG. NeuroImage
48. McMenamin BW, Shackman AJ, Maxwell JS, et al. Validation of regression-based myogenic correction techniques for scalp and source-localized EEG. Psychophysiology
49. Aho AJ, Lyytikainen LP, Yli-Hankala A, et al. Explaining entropy responses after a noxious stimulus, with or without neuromuscular blocking agents, by means of the raw electroencephalographic and electromyographic characteristics. Br J Anaesth
50. Aho AJ, Yli-Hankala A, Lyytikainen LP, Jantti V. Facial muscle activity, response entropy, and state entropy indices during noxious stimuli in propofol-nitrous oxide or propofol-nitrous oxide-remifentanil anaesthesia without neuromuscular block. Br J Anaesth
51. Jaaskelainen SK, Kaisti K, Suni L, et al. Sevoflurane is epileptogenic in healthy subjects at surgical levels of anesthesia. Neurology
52. Voss LJ, Sleigh JW, Barnard JP, Kirsch HE. The howling cortex: seizures and general anesthetic drugs. Anesth Analg
53. Kreuzer M, Kochs EF, Schneider G, Jordan D. Nonstationarity of EEG during wakefulness and anaesthesia: advantages of EEG permutation entropy monitoring. J Clin Monit Comput