Secondary Logo

Journal Logo

Does the Cerebral State Index Separate Consciousness from Unconsciousness?

Pilge, Stefanie MD*; Blum, Jasmin MD; Kochs, Eberhard F. MD; Schöniger, Stephan-Andreas MD; Kreuzer, Matthias MSc; Schneider, Gerhard MD*

doi: 10.1213/ANE.0b013e31823007cd
Technology, Computing, and Simulation: Research Reports

BACKGROUND: The Cerebral State Monitor™ (CSM) is an electroencephalogram (EEG)-based monitor that is claimed to measure the depth of hypnosis during general anesthesia. We calculated the prediction probability (PK) for its ability to separate consciousness from unconsciousness in surgical patients with different anesthetic regimens.

METHODS: Digitized EEG recordings of a previous study of 40 nonpremedicated, adult patients undergoing elective surgery under general anesthesia were replayed using an EEG player and reanalyzed using the CSM. Patients were randomly assigned to receive either sevoflurane-remifentanil or propofol-remifentanil. The study design included a slow induction of anesthesia and an episode of intended wakefulness. CSM values at loss and return of consciousness were compared. PK was calculated from values 30 seconds before and 30 seconds after loss and return of consciousness.

RESULTS: The PK for the differentiation between consciousness and unconsciousness was 0.75 ± 0.03 (mean ± SE). For sevoflurane-remifentanil, PK was 0.71 ± 0.04. For propofol-remifentanil, PK was 0.81 ± 0.03.

CONCLUSIONS: The ability of CSM for separation of consciousness and unconsciousness was comparable to other commercially available EEG-based indices.

Published ahead of print October 14, 2011 Supplemental Digital Content is available in the text.

From the *Department of Anesthesiology I, Helios Klinikum Wuppertal, University Witten/Herdecke, Wuppertal; and Department of Anesthesiology, Klinikum rechts der Isar, Technische Universität München, Munich, Germany.

Supported by departmental sources and a grant from the university (to KKF).

The authors declare no conflicts of interest.

This report was previously presented, in part, at the annual meeting of the American Society of Anesthesiologists, October 2005, Atlanta, GA.

Reprints will not be available from the authors.

Address correspondence to Gerhard Schneider, MD, Department of Anesthesiology I, Helios Klinikum Wuppertal, University Witten/Herdecke, Heusnerstr. 40, 42283 Wuppertal, Germany. Address e-mail to

Accepted July 6, 2011

Published ahead of print October 14, 2011

On the basis of auditory evoked potentials (AEPs) and/or electroencephalography (EEG), numerous different monitors are available to assess the hypnotic component of anesthesia. The Cerebral State Monitor™ (CSM) (Danmeter, Odense, Denmark), a portable handheld device, captures EEG from electrodes on the forehead and the mastoid. Calculation of cerebral state index (CSI) integrates a complex algorithm that uses 4 variables processed from the spontaneous EEG:

  1. β ratio = log
  2. α ratio = log
  3. β-α ratio = log
  4. Burst suppression ratio (percentage of time in a 30-second window where the EEG amplitude is <3.5 μV).

All variables correlate individually with the level of hypnosis. As a single measure, their correlation coefficient is low; a more reliable assessment may be achieved when the 4 variables are used as inputs of a fuzzy logic system (adaptive neuro-fuzzy inference system [ANFIS]).1 Output of the ANFIS is the CSI as a dimensionless number between 0 and 100. The underlying fuzzy logic system was designed to use the best applicable variable or a combination to reach a higher correlation with the level of hypnosis. It does not depend on a mathematical function that deduces the correlation between values calculated from the EEG and the patient's clinical state. The exact algorithm of CSI calculation, e.g., the selection of adequate variables by the ANFIS for the respective frequency content of the EEG, has not been revealed in detail by the inventor.

Only a few studies with the CSI have been published. In the present study, the CSM was tested for its ability to separate consciousness from unconsciousness in patients undergoing general anesthesia with 2 different anesthetic regimens.

Back to Top | Article Outline


Selection of Patients

Digitized EEG recordings of a previous study in adult patients undergoing elective surgery under general anesthesia2 were replayed with the EEG player and then reanalyzed with the CSM.3

Forty nonpremedicated, ASA physical status I or II patients scheduled to undergo elective surgery under general anesthesia were enrolled in the study; no geriatric patients were included.2 Patients gave informed, written consent to the protocol, which was approved by the ethics committee of the Technische Universität München (Munich, Germany). Patients with contraindications to the study drugs, a history of psychiatric or neurologic diseases, drug abuse, or medication known to affect the central nervous system, patients who were pregnant, and those with an indication for rapid sequence induction were excluded from the study.

Back to Top | Article Outline


Standard monitoring variables were recorded on a personal computer, including noninvasive arterial blood pressure, heart rate, oxygen saturation, end-tidal carbon dioxide concentration, sevoflurane concentration, and respiratory variables.

Back to Top | Article Outline

Clinical Protocol

Patients were randomly assigned to receive either anesthesia with sevoflurane and remifentanil (sevoflurane group) or total IV anesthesia with propofol and remifentanil (propofol group). Blocked randomization was performed to enroll 20 patients in each group. Without premedication, anesthesia was slowly induced with sevoflurane in oxygen by mask (sevoflurane group) or with propofol injections (0.7 mg · kg−1, followed by 20 mg every 30 seconds [propofol group]). In both groups, the minimal remifentanil infusion rate was 0.2 μg · kg−1 · min−1. Patients were asked every 30 seconds to squeeze the investigator's hand. To differentiate a response to command from involuntary movement, the response was verified by an immediate repetition of the command, which also required a response. Loss of consciousness (LOC1) was defined as a missing response to verbal command (Fig. 1). Tunstall's isolated forearm technique4 (IFT) was used to preserve the ability to follow commands before succinylcholine (1 mg · kg−1) was given. After tracheal intubation, sevoflurane or propofol was stopped until patients followed commands again (return of consciousness [ROC1]). Afterward, sevoflurane inhalation (5 vol%) or propofol bolus injection (20 mg every 20 seconds until LOC), followed by continuous infusion, was recommenced. LOC2 was reached when patients stopped responding to commands again. Sevoflurane, propofol, and remifentanil were administered according to standard clinical practice, and surgery was performed. At the end of surgery, remifentanil, sevoflurane, and propofol were discontinued and patients were asked again every 30 seconds to squeeze a hand. ROC2 was defined as the first verified squeeze of hand (Fig. 1).

Figure 1

Figure 1

Back to Top | Article Outline

Calculation of CSM Values

Using an amplifier designed for AEP/EEG recordings during anesthesia,5 a 2-channel referential EEG was recorded with electrode positions at the left temporal region between the lateral edge of the eye and the upper edge of the ear (AT1), on the right mastoid (M2), in the middle of the forehead (Fpz, reference), and on the left side of the forehead (F7, ground, electrode positions according to the international 10-20 system). EEG filter settings were 0.5 Hz (high pass) and 400 Hz (low pass). Skin was prepared with alcohol to keep impedances <5 kΩ. The EEG was continuously digitized and recorded on a personal computer with a rate of 1 kHz per channel.

The digitized EEG data, recorded with a sampling frequency of 1 kHz and filter bandwidth of 0.5 to 400 Hz from electrodes M2 and Fpz (reference), were replayed using a digital to analog converter constructed by the research group.3 A multifunction data acquisition card (DAQ, PCI 6036E; National Instruments™ Corp., Austin, TX) converted stored EEG data into a continuous analog output signal. Sampling rate of the output was identical with the sampling rate of the recorded EEG. No additional low pass filtering (smoothing) was performed. Only resistors and no additional capacitors were used. The signal was replayed with a resolution of 6 nV. Signal frequency, amplitude, and phase characteristics were not changed. The CSM (software version 1.008.012) was connected to the analog output to calculate CSI values. Using CSM Link™, a tool for wireless connection, the calculated index values were transferred to a laptop computer and stored for further analysis.

Back to Top | Article Outline

Statistical Analysis

The aim of the current study was to evaluate the ability of the CSM to separate consciousness from unconsciousness, in particular, at the transitions between the hypnotic levels. For this purpose, prediction probability (PK) was calculated from CSI pairs at predefined time points representing the conscious and the unconscious state. Time points were chosen close to the transitions between consciousness and unconsciousness, clinically verified as LOC and ROC. Thus, CSI values 30 seconds before and after LOC and ROC contributed to the analysis: the interval of 30 seconds to ask the patients to squeeze a hand allows a clear definition of the clinical end points LOC/ROC, and thus allows the conclusion that 30 seconds before ROC (LOC) the patient was unconscious (conscious). We intended to ascertain that the EEG monitor indicated the current clinical state by allowing a computing time of 30 seconds after LOC/ROC. Unconsciousness is represented by CSI values at time points ROC − 30 seconds and LOC + 30 seconds; analog: consciousness is represented by CSI values at time points LOC − 30 seconds and ROC + 30 seconds. The CSM did not always calculate an index value at the predefined time points. In these cases, the closest value to the respective time point that had been calculated during the corresponding clinical state was selected and the time interval (without index value) was listed. The selected CSI values were used to calculate PK as described by Smith et al.6 The PK for discrimination between consciousness and unconsciousness was calculated over all valid data pairs and for both anesthetic regimens separately. PK values were calculated with an Excel (Microsoft®, Redmond, WA) Macro (PKMACRO); differences between the sevoflurane and the propofol group were calculated with a different Excel Macro (PKDMACRO) using t scores for paired data. Both Excel Macros were provided by Smith et al.6 The PKMACRO computed statistics for grouped and paired data, and the compared PK values CSI-conscious and CSI-unconscious, using the worksheet results generated by PKMACRO. Overall significance level was set to P < 0.05 using the Bonferroni correction. In addition, a general linear model for repeated measurement with pairwise post hoc comparisons (P < 0.05, Bonferroni correction) was applied to CSI data obtained 30 seconds after LOC1, ROC1, LOC2, and ROC2. The allocated drug combinations were used as between-subject factor; LOC1, ROC1, LOC2, and ROC2 were used as within-subject factor.

A PKTool was used to identify the optimal thresholds of index values obtained by ROC analysis; in addition, PK values calculated by the Excel Macro (PKMACRO) were verified. The PKTool is a software that was developed by the authors7 and will be available on the Open TCI Web site ( PK was calculated for dichotomous states of CSI recorded during consciousness (30 seconds before LOC and 30 seconds after ROC, “CSI-conscious”) and unconsciousness (recorded 30 seconds after LOC and 30 seconds before ROC, “CSI-unconscious”). A detailed description of the PK algorithm is provided in the Appendix.

Patient characteristics and demographic data in both groups were calculated and compared with t tests and χ2 test (2-sided tests with unequal variances; for detailed results, we refer to previous published data2). Data are presented as mean ± SD, unless stated otherwise. Logistic regression was calculated and plotted using SigmaPlot 8.0 (Systat Software Inc., San Jose, CA).

Back to Top | Article Outline


Mean CSI Values

The Pk for the differentiation between consciousness and unconsciousness was 0.75 ± 0.03 (mean ± SE). For sevoflurane-remifentanil, Pk was 0.71 ± 0.04. For propofol-remifentanil, Pk was 0.81 ± 0.03. Table 1 displays mean CSI values during sevoflurane and propofol anesthesia at LOC1, ROC1, LOC2, and ROC2 (allowing computing time of 30 seconds, respectively). The observed mean threshold value to distinguish between conscious and unconscious was 72.5 ± 15.9 (mean ± SD) over all events (LOC1, LOC2, ROC1, and ROC2) and over both drugs. Additionally, receiver operating characteristic analysis was performed to obtain the maximal sensitivity and specificity at the optimal threshold (g) for separation between consciousness and unconsciousness: g (CSI) = 75 (sensitivity: 0.74; specificity: 0.71). Demographic data did not show significant differences between the sevoflurane group and the propofol group (Table 2).

Table 1

Table 1

Table 2

Table 2

Back to Top | Article Outline


According to the manufacturer, a CSI range of 40 to 60 reflects adequate anesthesia and a range of 80 to 90 drowsiness. These numerical thresholds are recommended for EEG monitoring of anesthetic depth. The agreement between the numerical classification of the conscious state according to the manufacturer's recommendations and the clinical assessment with the IFT was examined. Conflicting results are described as misclassifications. Considering a CSI value of 60 as the numerical threshold for the transition between consciousness and unconsciousness (CSI >60 conscious, CSI ≤60 unconscious), the CSM reported at 109 time points (35.3%) an index value that did not indicate the actual level of consciousness (e.g., the patient squeezed the investigator's hand to command but CSM indicated general anesthesia). If one assumes that drowsiness, but not a deeper level of anesthesia, allows the patient to follow a command (i.e., CSI ≥80 conscious, CSI <80 unconscious), 99 CSI signals (32.0%) were misclassified. A detailed report of misclassifications is given in Table 3. A χ2 test showed that the misclassifications occur among all patients and not in the same subjects; most have a misclassification for all periods.

Table 3

Table 3

The logistic regression curve in Figure 2 illustrates the increasing probability for correctly detecting consciousness when CSI values increase. The logistic regression coefficients A (a constant) and B (logit coefficients) were obtained for the sevoflurane and propofol group. The regression curve follows the function


Figure 2

Figure 2

Sevoflurane coefficients were A = −4.859 and B = 0.60 and propofol coefficients were A = −6.146 and B = 0.089. Hence, odds ratio exp(B) was 1.062 in the sevoflurane group and 1.094 in the propofol group. Odds ratio exp(B) indicates the increase of odds = P/(1 − P) = exp(A + B · CSI), i.e., odds is the ratio between the probability that a CSI value indicates consciousness and the probability that a CSI value indicates unconsciousness. Odds ratio exp(B) >1 indicates that exp(B) does not randomly deviate from 1 but that the states 0 (unconscious) and 1 (conscious) have a significant effect on the CSI. That is the case in both groups. The 95% confidence intervals of exp(B) were 1.035 to 1.090 for sevoflurane and 1.061 to 1.127 for propofol. In the propofol group, the odds ratio is higher than in the sevoflurane group. This indicates a more pronounced differentiation between consciousness and unconsciousness than in the sevoflurane group.

Most misclassifications occurred at LOC2 (37.37%) followed by LOC1 (29.29%), ROC1 (20.2%), and ROC2 (13.13%). Figure 3 shows index values at LOC and ROC (allowing 30-second computing time).

Figure 3

Figure 3

Back to Top | Article Outline

Internal Error Messages

At 40 of 316 analyzed time points, the CSM did not calculate an index value but displayed an error message. Thirty (75%) of these error messages occurred in the sevoflurane group and only 10 (25%) occurred in the propofol group. CSM displayed the most internal error messages at the event ROC2 (45%) followed by ROC1 (25%) and LOC2 (27.5%). Only 1 error message was shown at LOC1 (2.5%). The error signals may be caused by low signal quality or electromyogram or movement artifacts. All monitors rely on proprietary algorithms, including different algorithms for artifact detection. As a consequence, an analysis of monitor reactions to artifacts must remain speculative. Compared with other devices, the CSM integrates a mastoid electrode, which may be prone to muscle artifact.

In case of missing data points, the closest calculated CSI value was determined. Table 4 gives an overview of the time interval between the clinical event and the closest time when CSM calculated an index value. The time interval was between 1 and 174 seconds. In 7 cases (all after ROC2), no index values were obtained despite of an extended time window; in most cases, tracheal extubation had already been possible and EEG recordings were stopped.

Table 4

Table 4

Back to Top | Article Outline


CSI values at LOC during sevoflurane anesthesia were significantly higher than at LOC during propofol anesthesia. There was no significant difference in CSI values at ROC. Different resulting CSI values during LOC may have been attributable to different time delays of CSI calculation during deepening and lightening of anesthesia.8,9

The results of the current study indicate that the goal of predicting depth of anesthesia without hysteresis (deepening versus lightening of anesthesia) and independent of anesthetic drugs10 with CSM has not been reached.

In the present study, the ability of the CSM to separate consciousness from unconsciousness was analyzed at the transition between these levels. First, the study design included an episode of intended wakefulness. Second, because of a slow induction and emergence from anesthesia, there is a stepwise transition between the anesthetic levels. Third, the short interval of 30 seconds between asking the patients to squeeze a hand allows a clear definition of the analyzed clinical end points LOC and ROC. By comparing index values 30 seconds before and after LOC and ROC, the patients' states should have been in neither a fully awake nor a fully anesthetized state, but in the transition.

In our study, the IFT was used to detect consciousness. A verified response to a simple command was used as a clinical sign of consciousness. With this definition, periods of intact working memory are detected. Working memory has a limited capacity (a few seconds) and is the basis of perception. Other studies defined ROC as the ability to respond to a more complex or conditional command.11 The ability to follow a complex or conditional command reflects higher memory functions. Wang and Russel11 provided a classification of given responses when the IFT is used, based on increasing levels of consciousness as follows: grade 0 = no movement or response; grade 1 = reflex movement only (i.e., not contingent on command); grade 2 = response to simple command (i.e., contingent on command) but unable to respond to a more complex or conditional command; grade 3 = response to conditional command (e.g., “if you are comfortable, squeeze your hand twice”); and grade 4 = conditional response indicates patient is in pain.

In contrast to the response to a complex or conditional command (grade 3), our end point detects a more basic level of perception (grade 2). Although explicit (conscious) memory may require grade 3 responsiveness, it is still unclear at which level implicit (unconscious) memory occurs. Because anesthesia should prevent both explicit and implicit memory, we prefer the more conservative approach and focus on responsiveness (grade 2) as a sign of intact working memory, which is the basis of perception and memory. The present approach is a dichotomous approach, i.e., the patient is either responsive or not. In the context of anesthesia monitoring, the Observer's Assessment of Alertness/Sedation (OAA/S) scale is often used. It is a different approach to evaluate responsiveness in clinical studies12 with a more detailed evaluation of patient responses. As a consequence, it provides a more detailed classification of sedative effects of anesthetics. For our purpose, we decided to omit the information about subanesthetic effects and focused on unconsciousness rather than sedation. Regarding the detection of unconsciousness, the OAA/S scale uses a similar clinical concept as the IFT. The OAA/S scale cannot be used when neuromuscular blocking drugs are present and does not include complex or conditional commands. Clinical measures of consciousness are important components in the development process of an EEG-based monitor. The underlying algorithms of index computation remain proprietary. At present, the majority of commercially available EEG-based monitors were validated with different clinical scales during their development process; e.g., the manufacturers of the Bispectral Index (BIS), CSI, and Patient State Index (PSI) monitors referred to the modified OAA/S score to define which EEG-based subparameters or a combination of them correlated best with the clinical state. In contrast, the basic principle of Narcotrend monitor development was to identify EEG subparameters that discriminate best between visually determined EEG substages. It seems that with entropy-based EEG monitoring, variables are calculated directly from the EEG. The only commercially available variable is the so-called “time-frequency balanced spectral entropy” which is implemented in the Datex-Ohmeda Entropy™ Module. Aside from the underlying algorithm, each parameter was developed on the basis of a separate training dataset with different patients or volunteers and different anesthetic regimens. As a consequence, the performance of a monitor can only be as good as the corresponding database. Furthermore, the ability of an index value to recognize the hypnotic state can only be as reliable as the underlying method to classify the clinical state during the development process.

Calculated from the identical EEG data, BIS™ (Aspect Medical Systems, Newton, MA) obtained a similar result with a PK of 0.74.2 With a very similar study design, the PSI™ (Physiometrix, North Billerica, MA) separated consciousness from unconsciousness with a PK of <0.7,13 and the PK of Narcotrend® (MonitorTechnik, Bad Bramstedt, Germany) was only 0.5 for this task.14 A direct comparison of these PK values is not possible because of a different number of valid data pairs (BIS) or a slightly different study design (PSI, Narcotrend).

Hoymork et al.12 analyzed the clinical performance of the CSM compared with BIS in 55 patients receiving remifentanil and a target-controlled infusion of propofol. Evaluation of the current state was based on the OAA/S score. The PK for separation of awake and LOC (0.83) was similar to results for the propofol group in our study (0.81). During analysis of trend curves, in 13% of the patients, the CSM displayed values indicating an awake state during unconsciousness. As in our data, the CSM was not able to differentiate between consciousness and unconsciousness. The choice of a CSI value of 60 as a threshold between consciousness and unconsciousness may result in a bias toward detection of consciousness (even in unconscious patients). This bias reflects the view that it may be the greater dilemma to produce unintended consciousness rather than to consider a patient being conscious while he or she is anesthetized. Numerical thresholds according to the manufacturer are recommended for EEG monitoring of anesthetic depth. Our study revealed a considerable percentage of misclassifications between numerical and clinical classification of the hypnotic state. Receiver operating characteristics analysis was performed to obtain the maximal sensitivity and specificity at the optimal threshold g for differentiation between consciousness and unconsciousness: g (CSI) and g (BIS) are both well positioned between the upper threshold of 60 for general anesthesia and the lower limit of 80 for consciousness. Both misclassifications and the results of the receiver operating characteristic analysis support a critical view on thresholds recommended by the manufacturer. Information from processed EEG monitoring should always be accompanied by clinical assessment. Numerical thresholds may differ according to the particular EEG monitor, the anesthetic drugs, or within patient variability.

Zhong et al.15 compared the performance of the CSM to the BIS during target-controlled infusion of propofol with stepwise changes of the level of sedation, and concluded that the performance of the CSI was comparable to the performance of BIS. CSI and BIS values were compared at different sedation levels according to a modified OAA/S scale. Both indices showed almost identical Spearman correlation coefficients when compared with the modified OAA/S scale. Furthermore, PK values derived when comparing levels “awake,” “loss of verbal contact,” and “loss of response” were similar for the CSI and BIS. One study compared the CSM and BIS during propofol anesthesia in 15 patients16 and found that the overall performance of both monitors during propofol induction was similar: both the CSM and BIS proved a comparable overall PK to detect the level of anesthesia, which was estimated with propofol concentrations at the effect site (PK BIS 0.87 and PK CSI 0.86). Only 6 studies analyzed the CSI during inhaled anesthesia.1722 A comparative study19 analyzed BIS and CSI in 38 patients undergoing day surgery with different anesthetic drugs without muscle relaxants. Both indices decreased with increasing levels of sedation. However, neither monitor could give a reliable assessment of the levels according to the OAA/S scale, because indices recorded at distinct levels overlapped in their values. When both indices were compared, median index values were similar during surgery (BIS: 50 [14–89]; CSI: 51 [7–88]) and correlation between both indices was acceptable, i.e., both monitors seemed to give similar information. On rare occasions, CSI values deviated >100% from the BIS.

Time delay of CSI calculation may be a factor that contributes to impairment of PK. Time delay of CSI has been examined, e.g., at the transitions between “awake” and “general anesthesia” with artificially generated EEG signals9 as well as recorded and replayed EEG.3,8,9 The amount of time required to indicate a new level depends not only on the levels to be detected but also on the order in which these levels were applied; i.e., the delay at the change from “awake” to “general anesthesia” was different from the delay at the change from “general anesthesia” to “awake,” although identical EEG signals were used. There were considerable differences between calculation times at increasing and decreasing index values; e.g., time delay for indicating the transition from awake to general anesthesia was 53 seconds and vice versa from general anesthesia to awake was 45 seconds.9

A potential drawback may be the method of replaying stored EEG data. The EEG player has been designed as a method of replaying stored EEG data. An agreement between index values recorded from patients and from replayed EEG signals has been demonstrated for BIS calculated by the A1000 and A2000, for the Narcotrend, and the CSM.3

EEG was recorded from all positions recommended for use of BIS and CSM, i.e., EEG replayed to the BIS monitor was recorded from the forehead, whereas the CSM integrates a mastoid electrode. EEG filtering during its recording should not influence the results because EEG was recorded with an EEG/AEP device, developed in a European multicenter trial (BIOMED), which was designed for EEG recordings in the operating room.2 The use of this recording device reserves the advantage that no proprietary filters are introduced that could alter the EEG signal. The control of filter settings enabled us to estimate alterations of the EEG signal and led us to the conclusion that the frequencies used to calculate CSI are not influenced by the recording device. EEG was replayed without any additional filtering routines. The chosen frequency range of 0.5 to 400 Hz during recording contains the frequencies used by the CSM (6–42.5 Hz) for index generation.

Results of the logistic regression analysis must be observed with caution, because the underlying CSI data are not normally distributed. Analysis was performed to evaluate a particular topic: the index performance at the transitions between consciousness and unconsciousness.

Consideration of 4 transitions between consciousness and unconsciousness per patient includes evaluation of intraindividual parameter stability in PK statistics and completes interindividual assessment. Basically, repeated measurements may bias statistical evaluation, but effects were assumed to be negligible because measurements from identical clinical states were from distinct time points. To maintain statistical evaluation comparable to previous investigations,2 possible dependence was not considered for analysis.

These first observations do not provide insights into underlying mechanisms nor do they answer the question of which EEG-based index most reliably reflects the actual status of the patient. Underlying algorithms remain proprietary. Thus, valid interpretation of EEG analysis and reasonable transfer of these findings to clinical context are not possible. Research work has to cope with descriptions of index performance under certain (clinical or designed) conditions.

EEG-based monitoring does not provide a possibility to predict intraoperative responsiveness in advance. In addition, there is currently no EEG-based monitor that unfailingly detects wakefulness. None of the commercially available indices of the hypnotic state reliably separates responsiveness from unresponsiveness.2,8,9,1214,23

In the present study, performance of the CSM was superior during propofol anesthesia and therefore not independent from anesthetic regimens. The overall PK of 0.75 of the CSI for discrimination between consciousness and unconsciousness is comparable to the performance of other EEG-based indices. However, because of the lack of ability to perfectly separate consciousness from unconsciousness (with a PK of 1), there remains a considerable overlap between EEG-based classifications of anesthetic states for the CSI and other commercially available EEG-based indices.

Back to Top | Article Outline


Name: Stefanie Pilge, MD.

Contribution: This author wrote the manuscript, helped to design the study, conduct the study, and analyze the data.

Attestation: Stefanie Pilge has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files.

Name: Jasmin Blum, MD.

Contribution: This author helped to conduct the study, analyze the data, and write the manuscript.

Attestation: Jasmin Blum has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Eberhard F. Kochs, MD.

Contribution: This author helped to design the study and write the manuscript.

Attestation: Eberhard F. Kochs has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Stephan-Andreas Schöniger, MD.

Contribution: This author helped to design the study, conduct the study, and analyze the data.

Attestation: Stephan-Andreas Schöniger has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Matthias Kreuzer.

Contribution: This author analyzed the data, wrote the manuscript, helped to design the study, and conduct the study.

Attestation: Matthias Kreuzer has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Gerhard Schneider, MD.

Contribution: This author designed the study, conducted the study, helped to analyze the data, and write the manuscript.

Attestation: Gerhard Schneider has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

This manuscript was handled by: Dwayne R. Westenskow, PhD.

Back to Top | Article Outline


PK was calculated for dichotomous states of the cerebral state index (CSI) recorded during consciousness and unconsciousness.

= (k1, …, kM) represents the index values; and = (z1, …, zM) with zi ∈ {“CSI-consious”,“CSI-unconsious”} ordered as “CSI-conscious” > “CSI-unconscious” is the observed hypnotic state when the parameter value was obtained. PC defines the probability that 2 pairs (ki, zi) and (kj, zj) (i, j ∈ {1, …, M}, i ≠ j) and zi ≠ zj are concordant, i.e., zi > zj implies ki > kj whereas PD is the probability of 2 pairs being discordant, i.e., zi > zj implies ki < kj. Pt implies that 1 value is allocated to both states (identical relation), i.e., zi ≠ zj implies ki = kj. Then, prediction probability (PK) can be defined as:

Back to Top | Article Outline


1. Jang JSR. ANFIS: adaptive-network-based fuzzy interference system. IEEE Trans Syst Man Cybern 1993;23:665–85
2. Schneider G, Hollweck R, Ningler M, Stockmanns G, Kochs EF. Detection of consciousness by electroencephalogram and auditory evoked potentials. Anesthesiology 2005;103:934–43
3. Kreuzer M, Kochs EF, Pilge S, Stockmanns G, Schneider G. Construction of the electroencephalogram player: a device to present electroencephalogram data to electroencephalogram-based anesthesia monitors. Anesth Analg 2007;104:135–9
4. Tunstall ME. Detecting wakefulness during general anaesthesia for caesarean section. BMJ 1977;1:1321
5. Jordan C, Weller C, Thornton C, Newton DE. Monitoring evoked potentials during surgery to assess the level of anaesthesia. J Med Eng Technol 1995;19:77–9
6. Smith WD, Dutton RC, Smith NT. Measuring the performance of anesthetic depth indicators. Anesthesiology 1996;84:38–51
7. Jordan D, Steiner M, Kochs EF, Schneider G. A program for computing the prediction probability and the related receiver operating characteristic graph. Anesth Analg 2010;111:1416–21
8. Zanner R, Pilge S, Kochs EF, Kreuzer M, Schneider G. Time delay of electroencephalogram index calculation: analysis of cerebral state, bispectral, and Narcotrend indices using perioperatively recorded electroencephalographic signals. Br J Anaesth 2009;103:394–9
9. Pilge S, Zanner R, Schneider G, Blum J, Kreuzer M, Kochs E. Time delay of index calculation: analysis of cerebral state, bispectral, and Narcotrend indices. Anesthesiology 2006;104:488–94
10. 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 2000;93:876–82
11. Ghoneim M. Awareness During Anesthesia. Oxford: Butterworth-Heinemann, 2001
12. Hoymork SC, Hval K, Jensen EW, Raeder J. Can the cerebral state monitor replace the bispectral index in monitoring hypnotic effect during propofol/remifentanil anaesthesia? Acta Anaesthesiol Scand 2007;51:210–6
13. 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. Br J Anaesth 2003;91:329–35
14. Schneider G, Kochs E, Horn B, Kreuzer M, Ningler M. Narcotrend does not adequately detect the transition between awareness and unconsciousness in surgical patients. Anesthesiology 2004;101:1105–11
15. Zhong T, Guo QL, Pang YD, Peng LF, Li CL. Comparative evaluation of the cerebral state index and the bispectral index during target-controlled infusion of propofol. Br J Anaesth 2005;95:798–802
16. Cortinez LI, Delfino AE, Fuentes R, Munoz HR. Performance of the cerebral state index during increasing levels of propofol anesthesia: a comparison with the bispectral index. Anesth Analg 2007;104:605–10
17. Anderson RE, Barr G, Jakobsson JG. Cerebral state index during anaesthetic induction: a comparative study with propofol or nitrous oxide. Acta Anaesthesiol Scand 2005;49:750–3
18. Anderson RE, Jakobsson JG. Cerebral state index: comparison between pairwise registrations from the left and the right sides of the brain. Br J Anaesth 2006;97:347–50
19. Anderson RE, Jakobsson JG. Cerebral state monitor, a new small handheld EEG monitor for determining depth of anaesthesia: a clinical comparison with the bispectral index during day-surgery. Eur J Anaesthesiol 2006;23:208–12
20. Anderson RE, Sartipy U, Jakobsson JG. Use of conventional ECG electrodes for depth of anaesthesia monitoring using the cerebral state index: a clinical study in day surgery. Br J Anaesth 2007;98:645–8
21. Nishiyama T. Cerebral state index vs. bispectral index during sevoflurane-nitrous oxide anaesthesia. Eur J Anaesthesiol 2009;26:638–42
22. Revuelta M, Paniagua P, Campos JM, Fernandez JA, Martinez A, Jospin M, Litvan H. Validation of the index of consciousness during sevoflurane and remifentanil anaesthesia: a comparison with the bispectral index and the cerebral state index. Br J Anaesth 2008;101:653–8
23. Messner M, Beese U, Romstöck J, Dinkel M, Tschaikowsky K. The bispectral index declines during neuromuscular block in fully awake persons. Anesth Analg 2003;97:488–91
© 2011 International Anesthesia Research Society