Hypnosis and analgesia are 2 important components of anesthesia. Current electroencephalogram (EEG)-based depth of anesthesia monitors provide information on cortical activity and are therefore used as surrogate measures of hypnosis. However, because movement in response to noxious stimuli is mediated through subcortical structures,
EEG measures cannot reliably monitor the balance between nociception and antinociception/analgesia. 1 2 , The desire for indices to monitor both components of anesthesia has led to the development of new monitoring concepts. 3 4 , 5
Nociception-induced arousal responses during anesthesia are the result of ascending sensory signals, which lead to cortical activation. These arousal reactions can be blunted by very high concentrations of hypnotics or antinociceptive medication, such as opioids, in the presence of hypnotics. This illustrates the complex interaction between hypnotics and analgesics and their effect on responsiveness
and indicates that movement responses may be a sign of inadequate hypnosis, analgesia, or both. The quantification of indicators of both hypnosis and analgesia may assist in predicting movement during surgery. 6
The Bispectral Index (BIS™; Covidien, Mansfield, MA) is widely used because of its reported efficacy in defining optimal levels of hypnosis and minimizing intraoperative awareness.
Studies have shown that surgical stimulation increases the BIS and its variability, as well as increasing electromyographic (EMG) power and its variability. These responses are attenuated by the administration of analgesics. 7 8 , A derived calculation of analgesic state, the composite variability index (CVI), has therefore been developed as a function of BIS and EMG variability and is associated with an increased incidence of intraoperative somatic responses to noxious stimulation. 9 10
In contrast to BIS and CVI, which are both heuristically derived, Liley and Bojak
and Liley et al. 11 12 , have taken a more physiologically based approach to modeling the effects of anesthetic drugs on brain electrical activity, by using a detailed mathematical model of human EEG. This model aims to describe the electrical dynamics of coupled populations of excitatory and inhibitory neurons within the cerebral cortex that are driven by external inputs. Using an approximation of this model, we have developed 2 new EEG-based indices called cortical input (CI) and composite cortical state (CCS), with CCS a more robust derivative of the previously described cortical state (CS). 13 14 , CI is intended to quantify the magnitude of input to the cortex, whereas CCS characterizes the response of the cortex to an arbitrary stimulus or input. CS and CI have previously been shown to distinguish the effects of hypnotics and analgesics on cerebral electrical activity. 15 14
In this study, we aimed to investigate the predictive performance of these EEG-derived measures at different levels of hypnosis (titrated to BIS) and antinociception using remifentanil effect-site concentrations (Ce
REMI), as a surrogate, and in the presence of different stimuli: The Observer’s Assessment of Alertness and Sedation (OAA/S) scale and ulnar tetanic stimulus. Specifically, we aimed to determine (1) whether CI and CVI correlated with levels of remifentanil and how their performance compared in predicting or detecting movement in response to a noxious stimulus; (2) whether CCS correlated with levels of hypnosis as determined by BIS; (3) how the measures responded to the effects of combined stimuli, as well as the individual stimuli; and (4) whether the combined use of CCS and CI could discriminate between patients who showed movement in response to OAAS and/or tetanic stimulation and how this compared with the combined use of BIS and CVI. 16 METHODS
Patients and Procedure
Raw EEG data from a previously published study were reanalyzed,
as we aimed to quantify and compare CCS and CI with the previously recorded EEG-derived measures of hypnosis and analgesia, BIS, and CVI. All 4 measures were used in their raw time series form and reanalyzed for this article. The original study was approved by an institutional ethics committee (University Medical Center Groningen, Ethics’ Committee) and registered at 17 before patient enrollment (NCT01053611, principal investigator Michel M. R. F. Struys, trial registered in 2010). Written informed consent was obtained from 147 ASA physical status I and II patients, aged 18 to 65 years, undergoing surgery under general anesthesia. Exclusion criteria were as follows: weight <70% or >130% of ideal body weight, use of locoregional anesthetic techniques together with general anesthesia, neurological disorder or any condition or treatment that could potentially interfere with cardiovascular status, or level of consciousness during the induction (e.g., abuse of alcohol or use of benzodiazepines). In the original study, after exclusions, 120 patients were randomly allocated to 1 of the 12 study groups (see below for details). Cohorts of 10 patients per group were used as a convenience sample based on previous drug interaction studies. ClinicalTrials.gov 17
The clinical protocol has been described in the original publication.
In brief, frontal EEG activity was recorded using a Vista BIS monitor (Covidien, Boulder, CO) with 4 electrodes (BIS™ Quatro Sensor; Covidien, Boulder). The cerebral drug effect was monitored using the BIS™ (Covidien, Boulder). BIS electrodes were placed as recommended by the manufacturer. The raw EEG was recorded at a sampling rate of 128 Hz and stored for post hoc analysis. 17
In the original study, patients were allocated to 1 of the 12 groups by stratified randomization. Three different hypnotic levels, “light,” “intermediate,” and “deep,” were allocated to BIS targets of 70, 50, and 30, respectively. In the current study, the BIS 30 group has been excluded because of uncertainties about how CCS and CI are affected by burst suppression. Within each of the remaining hypnotic levels, patients were randomly allocated to 1 of the 4 groups to receive a target-predicted remifentanil effect-site concentrations of 0, 2, 4, or 6 ng/mL (Remi-0, -2, -4, and -6) using the pharmacokinetic model published by Minto et al.
18 , To ensure a stable hypnotic level, propofol was administered using a validated BIS-guided closed-loop system that controlled the predicted propofol effect-site concentration using the model from Schnider et al. 19 20 , Both propofol and remifentanil infusions were maintained for 17.5 minutes to reach a pharmacologic pseudo-steady state, such that subsequent observations of patient responsiveness could be meaningfully assessed. At the end of this period, the propofol closed-loop infusion was switched to an open-loop TCI administration, with the target concentration fixed at the achieved predicted, “steady-state” propofol effect-site concentration. 21
Eighteen minutes after the start of drug administration, we applied a sequence of stepwise graded stimuli according to the modified OAA/S scale (
Table 1) until a response was elicited or maximal stimulus intensity was reached, at which time the OAA/S score was noted. At 19 minutes, all patients with an OAA/S score <5 received a tetanic stimulus (100 Hz, 60 mA). The time taken to complete each OAA/S action was approximately 5 seconds. Therefore, if the OAA/S score was 2, it took 15 seconds to execute, meaning the tetanic stimulus was applied 45 seconds later. The tetanic stimulus was applied to generate a noxious stimulus that could evoke a subcortical nociceptive response. 22 , A 30-second exposure to a 50-mA current is a very painful stimulus that causes tetany and has been used in several studies as a standard noxious stimulus. 23 22 , Although this stimulus has not been tested in awake subjects (for obvious ethical reasons), a double burst 50-mA stimulation lasting 780 milliseconds produced a median visual analog pain intensity score of 7 to 8 in awake patients. 23 It is thus safe to assume that a 30-second 50-mA tetanic stimulus is a highly noxious stimulus. 24 Table 1:
Responsiveness Scores of the Modified Observer’s Assessment of Alertness/Sedation (OAA/S) Scale
The stimulus was applied for 30 seconds or until a purposeful response was noted. Purposeful responses were thus defined as any movement of the face, hands, and arms in response to the stimulus, other than the tetanic stimulus-induced muscle contraction itself. The stimulation electrodes were placed over the ulnar nerve on the volar side of the wrist. For ethical reasons, tetanic stimuli were not applied to patients with an OAA/S score of 5 (i.e., showing signs of wakefulness and responding to voice in normal tone/volume). The presence or absence of purposeful movement in response to tetanic stimulation was noted by the observer. Observations were continued until 180 seconds after the start of the OAA/S scale evaluation. All patients responding to either verbal and/or noxious stimuli were regarded as responders. The timeline of the study is shown in
Figure 1A. Figure 1:
Timeline of study period analyzed in (A) original published study
and (B) additional time divisions analyzed in this article. Median value over each time block was used for each measure.
We used Ce
REMI as a measure of the level of antinociception, the ulnar nerve stimulation as a standardized nociceptive stimulus, and the observed BIS to calculate the degree of hypnosis. Individuals with an OAA/S score of 0 and no movement in response to tetanic stimulation were defined as “nonresponsive,” whereas individuals with an OAA/S >0 and/or purposeful movement in response to tetanic stimulus were defined as “responsive.” CCS, CI, and Composite Variability Index (CVI)
CCS and CI were calculated using the method of Liley et al.
14 , 15 , and Frascoli et al. 25 In this method, developed by coauthor Liley, the cerebral cortex is modeled as a filter: incoming subcortical activity is filtered (dynamically transformed) by cortex to produce the recorded EEG signal. Various aspects of this process can be quantified. Specifically, both the state of the cortical filter (CCS) and the magnitude of its input (CI) can be estimated, under certain reasonable assumptions, using a variety of statistical time series methods that have been widely used to forecast trends in financial time series and other time-dependent stochastic processes. 26 The state of the cortical filter broadly characterizes the responsiveness of cortex to arbitrary input. Further details regarding this method can be found in Appendix 1 and the cited publications. 27
Previously, a simpler measure, CS was used to quantify the state of the cortical filter. Analysis involving real and simulated data has shown that the CCS gives rise to similar variations as the previously defined CS but is much less affected by noise. For this reason, we preferred to use CCS in this study. Previous studies have shown that CS (and by inference CCS) characterizes hypnotic drug action, whereas CI better represents analgesic drug effect.
14 , 15
To calculate CCS and CI, EEG data recorded at 128 Hz were resampled to 80 Hz limiting the bandwidth to 0.1 to 40 Hz. Resampled EEG time series were segmented into 2-second 50% overlapping epochs, and CCS and CI were calculated over each epoch. CCS is defined over the range 0 to −2, with more negative values reflecting increased hypnosis. To make the connection with our model as clear as possible, we chose not to scale this to the more familiar 0 to 100 range of BIS and other depth of anesthesia indices. Because our method assumes that the recorded EEG is a scaled version of CI, CI has units of μV. By hypothesis, and justified by preliminary study data, CI will be reduced during increases in antinociception. An automatic artifact rejection method detailed in Appendix 1 was applied to raw EEG data before calculating CCS and CI. This eliminated on average 30% of epochs.
CVI is dimensionless and ranges between 0 and 10, indicating a low and high level of nociceptive perception, respectively. It is calculated as the weighted combination of BIS, sBIS, and sEMG, where sBIS and sEMG are the SD of the BIS and EMG signal over the previous 3 minutes. More information can be found in the original publication.
17 Data Analysis
Offline analysis of the recorded raw EEG was later performed to calculate CVI, CI, and CCS. The measures considered during the analysis were BIS, CVI, CCS, and CI. Initially, 2 time periods, for comparison, were defined as in the original study,
(1) the baseline period, −30 to 0 seconds relative to application of OAA/S, and (2) the response period, 45 to 180 seconds after the start of the OAA/S recording ( 17 Fig. 1A). The tetanic stimulus was applied during the response period. Because the OAA/S and tetanic stimulus were applied sequentially and because of the possible effect of both stimuli on the probability of responsiveness, the systematic application of OAA/S and tetanic stimulation was seen as a combined stimulus. The value of a measure for an individual was therefore defined as its median value over the relevant time period. For example, the baseline CI for an individual was the median CI observation over the period −30 to 0 seconds relative to the application of OAA/S. In addition, for each patient, the change in measure as a result of stimulation (△parameter) was calculated as the difference between the baseline and the response period.
To further characterize changes in measures with application of the 2 stimuli, the following additional time periods were defined (
Fig. 1B) that allowed a more fine-grained temporal assessment of correlations between the applied stimuli and the processed EEG indices:
t = −20 to 0 seconds relative to the application of the OAA/S stimulus; T2—Post-OAA/S period:
t = 20 to 40 seconds after application of the OAA/S stimulus; T3—Pretetanic stimulus:
t = 40 to 60 seconds after application of the OAA/S stimulus (i.e., 20 seconds before tetanic stimulus); T4—During tetanic stimulation:
t = 60 to 90 seconds (stimulus applied for 30 seconds); T5, T6, T7, T8, T9—Post-tetanic stimulus:
t = 90 to 110 seconds, 120 to 140 seconds, 150 to 170 seconds, 180 to 200 seconds, and 210 to 230 seconds (i.e., five 20-second blocks after tetanic stimulus). Statistical Analysis
Residual normal probability plots were used to test the assumption of normality. When variance homogeneity was supported using the Levene test, analysis of variance (ANOVA) and
t tests were applied. When variance homogeneity was not supported, Kruskal-Wallis and Mann-Whitney U tests were applied. The Kruskal-Wallis H test is a rank-based nonparametric test used to determine differences between 2 or more groups. Linear mixed models were used to compare both between-subject and within-subject effects. This has the advantage over a repeated-measures ANOVA of being able to deal with missing data. Pairwise comparisons were made using Bonferroni correction with P adjusted by dividing by the number of paired comparisons to be made (for k groups,
). Residuals of the mixed-effects model were tested for normality using normal probability plots and for homogeneity of variance using the Levene tests. Associations were tested using the Spearman rank order correlation, ρ. Measures of effect size and 99% confidence intervals were calculated using Cohen’s d when comparing 2 groups (reported as: Cohen’s d [99% confidence interval]). Cohen’s d is defined as the difference between 2 means over the pooled SD and, along with the confidence interval, was calculated as described in Kadel and Kip.
To assess the ability of the measures CI, CCS, BIS, and CVI to indicate responsive and unresponsive patients, a dichotomized prediction probability,
P k, was calculated. 17 P k is an asymmetric measure of ordinal association. We calculated P k using the Somers’ D statistic in SPSS. For further details regarding the calculation, refer to Liley et al. 14
Using the time epochs defined in
Figure 1B, we investigated the ability of the combined pairwise measures (CCS/CI and BIS/CVI) to predict responsiveness to stimulation using k-means classification, a technique for data clustering and segregation. In this iterative method, a centroid is determined for each cluster and the sum, over all clusters, of the within-cluster sums of point-to-cluster-centroid distances, is minimized. The performance of k-means classification was quantified using sensitivity, specificity, positive prediction value, and negative prediction value. The 99% confidence intervals of these measures were calculated using the Wilson method. A 29 Z test for proportions was used to compare sensitivity and specificity. A linear combination of measures (CI + CCS and CVI + BIS) was also used to calculate the area under the receiver-operating characteristic (ROC) curve. The area under an ROC curve (AUROC) shows the ability of the classifier to correctly classify responders from nonresponders with areas close to 1, meaning high separability, and an area of <0.5, indicating chance or worse performance. A rough guide to classifying such separability using the AUROC is as follows: 0.9–1.0 = excellent classification; 0.8–0.9 = good classification; 0.7–0.8 = fair classification; 0.6–0.7 = poor classification; and 0.5–0.6 = failure. 30
All statistical analyses were performed using Matlab (Release 2012b; The MathWorks, Inc., Natick, MA) and IBM SPSS for Windows (version 20.0; IBM Corp., Armonk, NY). A value of
P < 0.05 was considered statistically significant. For P values between 0.01 and 0.05, statistical test assumptions were tested as described earlier and P values were reported as significant where the assumptions were met. RESULTS
Seventy-eight of a possible 80 patients were included in the analysis. The BIS 30 group was excluded to avoid a potential interference of burst suppression patterns in the raw EEG on CI and CCS calculations. In addition, 2 patients in the BIS 70, Remi-2 group were excluded. For one, there were unstable predicted plasma propofol concentrations during the response period because of a user error when switching from the closed-loop system to an open system. For the second patient, the OAA/S response was not recorded. Of these, 23 patients showed movement in response to stimulation (responders) and 55 did not show any movement (nonresponders).
Dependency of CI and CVI on Ce
At baseline, a significant effect of remifentanil was found for CVI with post hoc comparisons showing that Remi-0 had higher CVI values compared with Remi-6. CVI did not show significant differences in the Remi-2, -4, and -6 groups (
Table 2). Table 2:
The Average (SD) of Median Measures for Each Group of Patients During the Baseline and Response Periods Shown in Figure 1A
When categorizing groups into those who did not receive remifentanil (Remi-0) and those who did (Remi-2, -4, and -6), significantly higher values were found for both baseline CI (Cohen’s d: 0.65 [0.48–0.83],
P = 0.0217) and CVI (Cohen’s d: 0.72 [0.56–0.88], P = 0.0034) in the Remi-0 group. Table 3:
The Average (SD) of Median Measure Values During the Baseline and Response Periods (Figure 1A)
At baseline, BIS and CCS were significantly higher in responders (
Table 3). The ability of the measures to predict or differentiate between responders and nonresponders was assessed using the prediction probability P k (an ordinal association statistic). At baseline, both BIS and CCS showed similar 17 P k values. When assessing the change in measures at the stimulation period compared with baseline (△CCS, △BIS △CI, and △CVI, respectively), all 4 measures showed a greater change in responders. △CI and △CVI showed similar P k values (both higher than △CCS and △BIS), indicating similar performance in monitoring the response to stimulation. Correlations Between BIS with CCS and CI Figure 2:
Correlations between BIS and CCS at baseline and response periods and the 2 periods combined. CCS = composite cortical state; BIS = Bispectral Index.
To make comparisons with existing EEG-derived calculations of hypnosis, relationships between CCS and BIS were evaluated at baseline and the response period, as well as aggregated across both periods, with significant correlations found in all 3 cases (
Fig. 2). As expected, based on the assumption of independence, correlations between CI and BIS showed low R 2 values of 0.07 and 0.01 at baseline and the response period, respectively. Response of CCS, CI, BIS, and CVI to Stimuli
Across all groups, all 4 measures were significantly higher in the response period in comparison with the baseline (see
Table 2). An ANOVA with BIS and Ce REMI as factors showed that the change in measures (△ values) was not significantly different in the remifentanil groups but was significantly dependent on BIS level for △CI ( P = 0.0061), △CCS ( P = 0.0024), △BIS ( P = 0.0307), and △CVI ( P = 0.0109). In the BIS 70 group, the change was greater for all measures ( Table 2). Table 4:
Summary Data Presented According to Time Divisions T1 to T5 Shown in Figure 1B
Measures (A) CCS, (B) CI, (C) BIS, and (D) CVI from the different remifentanil groups across time. Mean and standard errors are shown (
n = 78, 6 patients had an OAA/S score of 5). Times shown relative to OAA/S stimulation are as follows: baseline t = −20 to 0 s (T1); post-OAA/S: t = 20 to 40 s (T2); pretetanic stimulus: t = 40 to 60 s (T3); during tetanic stimulus: t = 60 to 90 s (T4); five 20-second blocks post-tetanic stimulus: t = 90 to 110 s (T5), t = 120 to 140 s (T6), t = 150 to 170 s (T7), t = 180 to 200 s (T8), and t = 210 to 230 s (T9). Linear mixed models were used to compare measures between time periods and remifentanil groups. Time periods T1 (baseline) to T5 (immediately after tetanic stimulation) were considered in the analysis. No interaction between time and remifentanil level was found for any of the measures. Significant differences shown are between T1 (baseline) and other time points (i.e., post hoc analysis comparing T1 with T2 to T5, after a significant effect of time). * P < 0.05, ** P < 0.01. CSS = composite cortical state; CI = cortical input; BIS = Bispectral Index; CVI = composite variability index; OAA/S = Observer’s Assessment of Alertness/Sedation.
To characterize the effect of each of the applied stimuli (OAA/S and tetanic stimulation) on the putative indices of hypnosis and antinociception, comparisons between the additional time periods specified in
Figure 1B were made. Figure 3 shows the average and standard error of individual median values across the time divisions T1 to T9, with the following observations made ( Table 4):
A significant effect of time was found for both CI and CCS with both quantifiers increasing significantly after OAA/S stimulation (i.e., at T2). A reduction in CI and CCS was seen after OAA/S, with CI showing a similar value to the baseline at T3.
A significant effect of time was also found for BIS and CVI. A significant increase in BIS and CVI compared with baseline was reached at T3, which was later than CI and CCS. A reduction in BIS and CVI was not seen after OAA/S stimulation, and both measures continued to increase. This appears to indicate that, unlike CI, the OAA/S did not produce a short-lived phasic CVI response.
CI was significantly higher than baseline during tetanic stimulation but not after, as a reduction was seen at T5. CVI, BIS, and CCS were significantly higher both during and shortly after tetanic stimulation. Thus, after the tetanic stimulus, unlike CI, measures CCS, BIS, and CVI remained elevated.
CCS-CI and BIS-CVI Discrimination of Stimulus-Induced Movement
As indicated earlier, CCS and CI both increased significantly after OAA/S stimulation. Nevertheless, correlations between CCS and CI at baseline (T1) and after OAA/S stimulation (T2) showed low
R 2 values of 0.012 and 0.123, respectively, suggesting that they are, as per hypothesis, characterizing different physiologic consequences of the OAA/S stimulus. Given the lack of relationship between these quantifiers, they were combined to determine whether they could better distinguish responders from nonresponders compared with the individual indices alone. In contrast, BIS and CVI were significantly correlated at these time points (T1: R 2 = 0.30, P < 0.0001; T2: R 2 = 0.36, P < 0.0001) but were also combined to compare their performance. Table 5:
Performance of Combined CI/CCS and CVI/BIS to Correctly Identify Responders from Nonresponders at T2 Using k-Means Classification
CCS/CI and BIS/CVI used in combination. Responders and nonresponders are shown. A and B, Measures at T1 (baseline) and (C) and (D) at T2 (after OAA/S stimulus). At T2, k-means clustering showed CI and CCS combined were more sensitive predictors of response to stimulation than CVI and BIS combined (sensitivity 75.8% vs 42%). CCS = composite cortical state; CI = cortical input; BIS = Bispectral Index; CVI = composite variability index; OAA/S = Observer’s Assessment of Alertness/Sedation.
As illustrated in
Figure 4, A and B, at baseline for both CI/CCS and CVI/BIS combinations, responders could not be visually distinguished from nonresponders, as these 2 groups overlapped. However, at T2 (shortly after the application of the OAA/S stimulus), the 2 groups visually appeared to be better separated in the CI-CCS plane, and therefore, sensitivity and specificity of the combined measures were calculated to quantify the separation. Table 5 shows the sensitivity and specificity of the combined measures as differentiated using k-means clustering. Combined CI and CCS values showed better sensitivity ( Z = 2.72, P = 0.006) than CVI and BIS and a trend for specificity ( Z = 2.41, P = 0.0159) in differentiating responders from nonresponders. The AUROC was used to compare the performance of a simple linear combination of the measures, CI + CCS and CVI + BIS, in distinguishing responders. An area of 0.807 (99% confidence interval, 0.623–0.943) was found for CI + CCS and 0.786 (99% confidence interval, 0.631–0.899) for CVI + BIS. DISCUSSION
The aim of the current study was to describe and compare the performance of the EEG-derived quantifiers, CCS and CI, with BIS and CVI, in determining different levels of hypnosis and the nociception-antinociception balance and in their ability to predict patients’ somatic tolerance to painful stimuli. The study period was segmented in 2 ways (
Fig. 1). The first reflected the time divisions used in an original publication in which the response period included both the OAA/S and the tetanic stimuli. In a second detailed reanalysis, a more fine-grained division of time was used, which allowed a separate analysis of the effects of each of the stimuli. Our main finding in this study was that combining CCS and CI showed better sensitivity to predicting responders before application of the tetanic stimulus compared with combining CVI and BIS. With regard to the aims, the study showed that (1) at baseline and before application of any stimulus, CVI and CI both distinguished patients who had or had not received remifentanil. They did not, however, show a concentration-effect relationship as a function of remifentanil level. (2) A strong correlation between CCS and BIS was found at both the baseline and the response period. (3) After application of an OAA/S stimulus, a significant increase in CCS and CI, but not BIS or CVI, was observed. (4) Combining CCS and CI showed better sensitivity to predicting responders before application of the tetanic stimulus compared with combining CVI and BIS. 17
At baseline, before application of any stimulus, CVI showed a significant difference between remifentanil groups with post hoc analysis showing differences between Remi-0 and Remi-6. It therefore appeared that CVI could differentiate between patients receiving remifentanil and those who had not received it. However, when categorized according to remifentanil or no-remifentanil groups, both CI and CVI could differentiate between the 2 with similar effect size magnitudes. This could provide clinically useful information because existing EEG-derived indices do not distinguish opioid levels well in the unstimulated condition.
The measures, however, did not show a remifentanil concentration–dependent relationship. One reason for this may be the variability in the predictive performance of all pharmacokinetic models and, in our case, the Minto model. 17 18 , Although in a sample of individuals we would expect a mean reasonably close to the target concentration, the interindividual variability in the concentration-effect response for opiates may explain the lack of concentration-effect relationship with remifentanil levels. To standardize opioid administration to achieve similar antinociceptive states in all individuals in a group (similar to using BIS for hypnosis), a reliable control variable would be required. However, currently there is no variable sufficiently robust for closed-loop control of the antinociceptive state of anesthesia. Another possible reason for not observing the concentration-effect relationship with remifentanil levels could be a ceiling effect in the influence of remifentanil, which is concordant with previous work showing that 2 ng/mL remifentanil causes a significant decline in propofol requirements for hypnotic and noxious end points, whereas higher doses conferred little additional benefit. 19 Lack of a concentration-dependent relationship has also been reported for a variety of other putative measures of antinociception which have shown such a relationship only in the presence of stimulation. As an example, the surgical pleth index was shown to be dependent on remifentanil concentrations but only after tetanic stimulation. 31 23 , Changes in CVI have also been related to increasing remifentanil concentrations during stimulation at levels of 0 to 3 ng/mL. 32 Perhaps a more accurate quantifier of input to the modeled cortical filter (CI) is required to capture changes in remifentanil levels. Overall, it appears that CI and CVI respond similarly to Ce 33 REMI levels in the unstimulated condition.
CCS and BIS at baseline were significantly higher in responders than in nonresponders, with both showing similar
P k values ( Table 3). Because immobility during anesthesia can be achieved using hypnotics, opioids, or combinations of different drugs, movement in response to stimulation can be a result of inadequate hypnosis, analgesia, or a combination of both. 17 , Higher CCS and BIS values at baseline in responders could indicate hypnotic arousal that could contribute, in addition to nociception, to the movement observed in response to stimulation. The change in CI and CVI with stimulation was greater in responders for all measures, again indicating that the somatic response observed could be due to both hypnotic and nociceptive arousal. △CI and △CVI showed a higher 34 P k value than the baseline or stimulation periods. The values were similar for both △CI and △CVI, indicating similar performance in monitoring the response to stimulation.
It is known that BIS is a surrogate measure of the hypnotic component of anesthesia and has its limitations.
35 , However, in this study, hypnotic levels were obtained by titrating to BIS values because standardization of cerebral effect would not have been possible with a target-controlled propofol and remifentanil concentration approach. 36 When target-controlled concentrations are used, individual cerebral effects may be different in patients. However, similar BIS values are likely to be the result of similar cerebral effects as measured by EEG. We therefore compared the relationship between BIS and CCS and found a close correlation between the 2 measures. This suggests that the change in hypnotic state as mapped by BIS is well defined by CCS. In this study, CI did not show any correlation with BIS (see Results: Correlations Between BIS with CCS and CI). This is in agreement with previous studies, which have shown that CI and CCS are independent measures. 17 14
All 4 measures, CCS, CI, BIS, and CVI, were significantly elevated during the combined stimulus period (
Fig. 1A) to the baseline. For all measures, differences between the baseline and the response periods were greater in the BIS 70 group, as found previously, suggesting that the performance of the measures is, possibly dominantly, influenced by hypnotic level. 17
After OAA/S stimulation, no significant increase was seen in BIS and CVI at T2; however, these continued increasing, such that at T3 they were significantly higher than baseline. With CI and CCS, the increase was significant at T2 with CI decreasing back toward baseline at T3 (see
Fig. 3), suggesting faster response times of CI and CCS to stimulation and faster recovery of CI.
During tetanic stimulation (T4;
Fig. 1A), CI showed a significant increase in comparison with baseline, but not after stimulation, because it had again decreased. Visual inspection of the raw EEG did not show interference from the tetanic stimulus, indicating that the increase was not due to any imposed artifacts. CVI was significantly higher both during and after tetanic stimulation; however, as seen in Figure 3D, the increase appears to be initiated after OAA/S stimulation, and, therefore, it is difficult to distinguish the effect of each stimulus. A delay in processing time has been reported for the BIS. Indeed, as described, the CVI is calculated over a 3-minute window. The possible faster response time for CCS and CI could reduce the overlap in the detected effects of the 2 stimuli. 35
After OAA/S stimulation (T2, see
Fig. 1B), CCS and CI but not BIS or CVI showed a significant increase. CCS and CI showed a lack of correlation both in this study and in previous studies where CS (an alternate quantifier of CCS) and CI responded differently to target propofol and remifentanil effect-site concentrations. Because of this independence, and the significant change in both CCS and CI after OAA/S, we believe that combining these indices would provide a more accurate quantifier to predict response to stimulation. Although BIS and CVI were found to be correlated, they were used as a comparison combination measure because of their suggested use as an index of hypnosis and as a proposed index of antinociceptive state. When combined, and using k-means clustering, CCS and CI show higher sensitivity in distinguishing responders from nonresponders compared with CVI and BIS. However, given that patients who responded to either or both stimuli (OAA/S and tetanic) were regarded as responders, such a classification could either represent a prediction of response to tetanic stimulation or be an indication of response to the OAA/S assessment. On the basis of the work by Rampil, 14 which showed the spinal cord to be the site of anesthetic inhibition of motor response, it could be argued that the spinal cord–evoked movement is not considered in our methodology. Nevertheless, whether CCS and CI are classifying a response to a relatively weak intraoperative stimulus or predicting response to a noxious stimulus, the values of CCS and CI might provide some indication of a threshold above which more hypnotic or analgesic agent is required to attenuate an ongoing or expected response. 37
Unanswered by this study is whether the application of a noxious stimulus is required for the assessment of the patient’s nociceptive state. Further studies with a longer time period between stimuli and a clear differentiation of the responses to OAA/S and tetanic stimuli are needed to substantiate the ability of CI/CCS to predict response to genuinely painful stimuli. Because a simple additive combination of these quantifiers was used for ROC curves, future work could investigate the use of other functions of these quantifiers to improve discriminability.
A limitation of our analysis approach may have been our method used to eliminate electroencephalographic artifact. A less conservative method could be developed to reduce the percentage of epochs rejected and subsequently interpolated.
In summary, our findings indicate that CI and CVI both distinguished between no remifentanil and remifentanil in the absence of stimulation. With the remifentanil doses used, neither CI nor CVI showed a clear concentration-dependent relationship. CI returns to baseline more rapidly than CVI after application of both stimuli, possibly because of the faster processing time of CI compared with CVI. CCS and BIS showed strong correlations, suggesting that they behave similarly as indicators of hypnosis. When combined, and using k-means clustering, CCS and CI show higher sensitivity in distinguishing, and likely predicting, responders from nonresponders compared with CVI and BIS. This may have important implications in the development of approaches to better optimize the delivery of balanced anesthesia.
Calculation of Composite Cortical State (CCS) and Cortical Input (CI)
CCS and CI were calculated by using the method of Liley et al.,
14 , which we briefly summarize here. The mean field model of human EEG known as the Liley Model describes the bulk electrical activity generated in cortex using a set of nonlinear partial differential equations driven by white noise. 15 11 , 12 , 25 , Previous studies 26 have shown that a linearized form of this model can also be used to model the EEG as an autoregressive moving average (ARMA) time series with fixed orders of 8 and 5 (8 autoregressive terms and 5 moving average terms). 11 13 , This suggests that the EEG, to first approximation, can be understood as arising from the linear filtering, by cortex, of subcortical input. 15
From the ARMA model, CI and CCS were calculated as follows: EEG data recorded at 128 Hz were resampled to 80 Hz limiting the bandwidth to 0.1 to 40 Hz and to avoid spurious fitting to 50/60 Hz spectral peaks or any low-pass filter band edges.
Resampled EEG time series were segmented into 2-second 50% overlapping epochs. Epochs with missing data were excluded from subsequent analysis. A simple, but effective, automatic artifact rejection method was applied. The Lilliefors test was used to evaluate the null hypothesis ( 15 P < 0.01) that the amplitude distribution for a given 2-second epoch was drawn from a normal distribution. Epochs that did not have a normal distribution were rejected. This eliminated on average 30% of epochs.
Using the ARMASA MATLAB Toolbox,
ARMA models of order (8,5) were fitted to each epoch 27 s[ n]:
u[ n] is a sequence of uncorrelated random variables and a k and b k are the ARMA coefficients.
In brief, ARMASA removes the mean of the epoch and then estimates an invertible and stationary ARMA model using a variant of the Durbin method with optimal intermediate autoregressive order. This model defines a linear electrocortical filter. The poles of this filter correspond to dominant oscillatory processes in the EEG frequency bands and together with the zeros characterize the state or responsiveness of the electrocortical filter. The amplitude of the noise driving this filter is also of physiologic significance and provides a measure of subcortical input to the cortex.
14 , 15
To quantify these filter characteristics, CCS is defined as the difference between the scaled mean pole location and the scaled mean zero location of the estimated filter (
a 1 − b 1)/13. Previously, a simpler measure, cortical state (CS), was used to quantify the state of the electrocortical filter, which corresponded to the scaled mean pole location of the estimated electrocortical filter. 14 , Because the zeros of the filter are also important in characterizing it, CCS was introduced in an attempt to better define the state of the filter. Analysis involving real and simulated data has shown that the CCS gives rise to similar variations as the previously defined CS but is much less affected by noise. For this reason, we chose CCS as a more robust scalar quantification of the estimated electrocortical filter. CI, a measure of the noise driving the electrocortical filter (i.e., subcortical input), is quantified as the variance of 15 s[ n] divided by the power gain of the derived filter and is in units of μV. Previous studies have shown that CS (and by inference CCS) better characterizes hypnotic drug action, whereas CI better represents analgesic drug effect. 14 14 , 15
Missing values of CCS and CI, because of the removal of corrupted 2-second epochs, were interpolated by using a second-order Savitzky-Golay filter of span 19 seconds applied to the estimated CCS and CI values. The span and order of this filter was chosen to provide a reasonable compromise between time resolution and smoothness.
Name: Mehrnaz Shoushtarian, PhD.
Contribution: This author helped analyze the data and write the manuscript.
Attestation: Mehrnaz Shoushtarian has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Conflicts of Interest: Mehrnaz Shoushtarian is the principal scientist and holds options in Cortical Dynamics Ltd., North Perth, WA, Australia.
Name: Marko M. Sahinovic, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Marko M. Sahinovic 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.
Conflicts of Interest: None.
Name: Anthony R. Absalom, MBChB, FRCA, MD.
Contribution: This author helped design the study, conduct the study, and reviewed the manuscript.
Attestation: Anthony R. Absalom has seen the original study data and approved the final manuscript.
Conflicts of Interest: Anthony R. Absalom reported no conflicts of interest.
Name: Alain F. Kalmar, MD, PhD.
Contribution: This author helped design the study, conduct the study, and reviewed the manuscript.
Attestation: Alain F. Kalmar has seen the original study data and approved the final manuscript.
Conflicts of Interest: Alain F. Kalmar reported no conflicts of interest.
Name: Hugo E. M. Vereecke, MD, PhD.
Contribution: This author helped design the study, conduct the study, and reviewed the manuscript.
Attestation: Hugo E. M. Vereecke approved the final manuscript.
Conflicts of Interest: Hugo E. M. Vereecke reported no conflicts of interest.
Name: David T. J. Liley, MBChB, PhD.
Contribution: This author helped analyze the data and write the manuscript.
Attestation: David T. J. Liley has seen the original data, reviewed the analysis of the data, and approved the final manuscript.
Conflicts of Interest: David T. J. Liley is the co-founder and Chief Scientific Officer and holds shares in Cortical Dynamics Ltd., North Perth, WA, Australia.
Name: Michel M. R. F. Struys, MD, PhD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Michel M. R. F. Struys helped design the study, has seen the original study data, reviewed the analysis of the data, approved the final manuscript, coauthored the writing of the manuscript, and is the author responsible for archiving the study files.
Conflicts of Interest: During the last 5 years, Michel M. R. F. Struys served 2 times as a member of an advisory panel organized by Covidien Boulder, CO. He received speaker’s honoraria from Covidien. He is also a co-inventor of the closed-loop system used in the study.
This manuscript was handled by: Ken B. Johnson, MD. ACKNOWLEDGMENTS
We acknowledge the statistical advice given by Denny Meyer, PhD, Associate Professor in Biostatistics, Faculty of Health, Arts and Design, Swinburne University of Technology, Hawthorn, Victoria, Australia.
1. Velly LJ, Rey MF, Bruder NJ, Gouvitsos FA, Witjas T, Regis JM, Peragut JC, Gouin FM. Differential dynamic of action on cortical and subcortical structures of anesthetic agents during induction of anesthesia. Anesthesiology. 2007;107:202–12
2. Bergmann I, Göhner A, Crozier TA, Hesjedal B, Wiese CH, Popov AF, Bauer M, Hinz JM. Surgical pleth index-guided remifentanil administration reduces remifentanil and propofol consumption and shortens recovery times in outpatient anaesthesia. Br J Anaesth. 2013;110:622–8
3. Lobo FA, Schraag S. Limitations of anaesthesia depth monitoring. Curr Opin Anaesthesiol. 2011;24:657–64
4. Sleigh J. No monitor is an island: depth of anesthesia involves the whole patient. Anesthesiology. 2014;120:799–800
5. Kortelainen J, Seppänen T. Electroencephalogram-based depth of anaesthesia measurement: combining opioids with hypnotics. Trends Anaesth Crit Care. 2013;3:270–8
6. Shafer SL. All models are wrong. Anesthesiology. 2012;116:240–1
7. Myles PS, Leslie K, McNeil J, Forbes A, Chan MT. Bispectral index monitoring to prevent awareness during anaesthesia: the B-Aware randomised controlled trial. Lancet. 2004;363:1757–63
8. Guignard B, Menigaux C, Dupont X, Fletcher D, Chauvin M. The effect of remifentanil on the bispectral index change and hemodynamic responses after orotracheal intubation. Anesth Analg. 2000;90:161–7
9. Iselin-Chaves IA, Flaishon R, Sebel PS, Howell S, Gan TJ, Sigl J, Ginsberg B, Glass PS. The effect of the interaction of propofol and alfentanil on recall, loss of consciousness, and the Bispectral Index. Anesth Analg. 1998;87:949–55
10. Mathews DM, Clark L, Johansen J, Matute E, Seshagiri CV. Increases in electroencephalogram and electromyogram variability are associated with an increased incidence of intraoperative somatic response. Anesth Analg. 2012;114:759–70
11. Bojak I, Liley DT. Modeling the effects of anesthesia on the electroencephalogram. Phys Rev E Stat Nonlin Soft Matter Phys. 2005;71:041902
12. Liley DTJ, Cadusch PJ, Dafilis MP. A spatially continuous mean field theory of electrocortical activity. Network. 2002;13:67–113
13. Liley DT, Cadusch PJ, Gray M, Nathan PJ. Drug-induced modification of the system properties associated with spontaneous human electroencephalographic activity. Phys Rev E Stat Nonlin Soft Matter Phys. 2003;68:051906
14. Liley DT, Sinclair NC, Lipping T, Heyse B, Vereecke HE, Struys MM. Propofol and remifentanil differentially modulate frontal electroencephalographic activity. Anesthesiology. 2010;113:292–304
15. Liley DT, Leslie K, Sinclair NC, Feckie M. Dissociating the effects of nitrous oxide on brain electrical activity using fixed order time series modeling. Comput Biol Med. 2008;38:1121–30
16. Chernik DA, Gillings D, Laine H, Hendler J, Silver JM, Davidson AB, Schwam EM, Siegel JL. Validity and reliability of the Observer’s Assessment of Alertness/Sedation Scale: study with intravenous midazolam. J Clin Psychopharmacol. 1990;10:244–51
17. Sahinovic MM, Eleveld DJ, Kalmar AF, Heeremans EH, De Smet T, Seshagiri CV, Absalom AR, Vereecke HE, Struys MM. Accuracy of the composite variability index as a measure of the balance between nociception and antinociception during anesthesia. Anesth Analg. 2014;119:288–301
18. Minto CF, Schnider TW, Egan TD, Youngs E, Lemmens HJ, Gambus PL, Billard V, Hoke JF, Moore KH, Hermann DJ, Muir KT, Mandema JW, Shafer SL. Influence of age and gender on the pharmacokinetics and pharmacodynamics of remifentanil. I. Model development. Anesthesiology. 1997;86:10–23
19. Minto CF, Schnider TW, Shafer SL. Pharmacokinetics and pharmacodynamics of remifentanil. II. Model application. Anesthesiology. 1997;86:24–33
20. Schnider TW, Minto CF, Gambus PL, Andresen C, Goodale DB, Shafer SL, Youngs EJ. The influence of method of administration and covariates on the pharmacokinetics of propofol in adult volunteers. Anesthesiology. 1998;88:1170–82
21. Schnider TW, Minto CF, Shafer SL, Gambus PL, Andresen C, Goodale DB, Youngs EJ. The influence of age on propofol pharmacodynamics. Anesthesiology. 1999;90:1502–16
22. Rantanen M, Yppärilä-Wolters H, van Gils M, Yli-Hankala A, Huiku M, Kymäläinen M, Korhonen I. Tetanic stimulus of ulnar nerve as a predictor of heart rate response to skin incision in propofol remifentanil anaesthesia. Br J Anaesth. 2007;99:509–13
23. Struys MM, Vanpeteghem C, Huiku M, Uutela K, Blyaert NB, Mortier EP. Changes in a surgical stress index in response to standardized pain stimuli during propofol-remifentanil infusion. Br J Anaesth. 2007;99:359–67
24. Connelly NR, Silverman DG, O’Connor TZ, Brull SJ. Subjective responses to train-of-four and double burst stimulation in awake patients. Anesth Analg. 1990;70:650–3
25. Liley DTJ, Cadusch PJ, Wright JJ. A continuum theory of electro-cortical activity. Neurocomputing. 1999;26–27:795–800
26. Frascoli F, Van Veen L, Bojak I, Liley DTJ. Meta-bifurcation analysis of a mean field model of cortical neuronal. Physica D. 2011;240:949–62
27. Broersen PMT. Automatic spectral analysis with time series models. IEEE Trans Instrum Meas. 2002;51:211–6
28. Kadel RP, Kip KE A SAS Macro to Compute Effect Size (Cohen’s d) and Its Confidence Interval from Raw Survey Data. 2012 Durham, NC SouthEast SAS Users Group Conference
29. Brown LD, Cai TT, DasGupta A. Interval estimation for a binomial proportion. Stat Sci. 2001;16:101–33
30. Gorunescu F Data Mining: Concepts, Models and Techniques. 2011 Berlin, Germany Springer
31. Struys MM, Vereecke H, Moerman A, Jensen EW, Verhaeghen D, De Neve N, Dumortier FJ, Mortier EP. Ability of the bispectral index, autoregressive modelling with exogenous input-derived auditory evoked potentials, and predicted propofol concentrations to measure patient responsiveness during anesthesia with propofol and remifentanil. Anesthesiology. 2003;99:802–12
32. Gruenewald M, Meybohm P, Ilies C, Höcker J, Hanss R, Scholz J, Bein B. Influence of different remifentanil concentrations on the performance of the surgical stress index to detect a standardized painful stimulus during sevoflurane anaesthesia. Br J Anaesth. 2009;103:586–93
33. Ellerkmann RK, Grass A, Hoeft A, Soehle M. The response of the composite variability index to a standardized noxious stimulus during propofol-remifentanil anesthesia. Anesth Analg. 2013;116:580–8
34. Hendrickx JF, Eger EI II, Sonner JM, Shafer SL. Is synergy the rule? A review of anesthetic interactions producing hypnosis and immobility. Anesth Analg. 2008;107:494–506
35. Bowdle TA. Depth of anesthesia monitoring. Anesthesiol Clin. 2006;24:793–822
36. Bruhn J, Myles PS, Sneyd R, Struys MM. Depth of anaesthesia monitoring: what’s available, what’s validated and what’s next? Br J Anaesth. 2006;97:85–94
37. Rampil IJ. Anesthetic potency is not altered after hypothermic spinal cord transection in rats. Anesthesiology. 1994;80:606–10