New Composite Index Based on Midlatency Auditory Evoked Potential and Electroencephalographic Parameters to Optimize Correlation with Propofol Effect Site Concentration: Comparison with Bispectral Index and Solitary Used Fast Extracting Auditory Evoked Potential Index

After institutional ethics committee approval (University Hospital, Ghent, Belgium), informed consent was obtained from 13 patients (10 women, 3 men) with an American Society of Anesthesiologists physical status of I, aged 18–65 yr, who were scheduled to undergo ambulatory gynecologic or urologic surgery. Exclusion criteria included weight less than 70% or more than 130% of ideal body weight, neurologic disorder, and recent use of psychoactive medication, including alcohol.

An 18-gauge intravenous line was positioned at a large left forearm vein. Every patient received approximately 100 ml crystalloid fluid during the study period. No fluid load was given before induction. No patient received preoperative medication. No other drugs, including opioids, were given during the study period. All patients maintained spontaneously ventilating (even at high levels of burst suppression) via  a facemask delivering 100% oxygen at 6 l/min.

Heart rate and noninvasive blood pressure, arterial oxygen saturation, and capnography were recorded at 1-min intervals using an S-5 monitor (Datex, Helsinki, Finland). Capnography was monitored by putting the side stream sample line in the facemask of the patient. This implies the occurrence of an error for quantification of the partial pressure of carbon dioxide, but it enables monitoring of respiratory rate and free airway. A silent operation room was obtained for all patients.

At the start of the study period, 1% propofol was administered at 300 ml/h via  a computer-assisted continuous infusion device (RUGLOOP**). This system captures all monitored data while driving a Fresenius Modular Infusion Pump connected to a Fresenius Base A (Fresenius Vial Infusion Systems, Brézins, France) via  an RS-232 interface. During the propofol infusion, the corresponding Ce^PROPis calculated in a time-synchronized way by RUGLOOP using a three-compartment model enlarged with an effect site compartment, previously published by Schnider et al.  15,16The calculated Ce^PROPwas computed to yield a time to peak effect of 1.6 min after bolus injection,17as also published by Schnider et al.  15,16and clinically confirmed by Struys et al.  18The propofol infusion was continued at the same speed until burst suppression levels of 80% or higher were achieved. However, infusion was stopped earlier if the mean arterial blood pressure became lower than 50 mmHg.

Before the drug administration was started, all patients were asked to close their eyes and relax for 2 min. During that time, signal quality, impedance of the electrodes, and the adequate detection of all parameters by RUGLOOP were verified. Baseline measures were performed during the first 5 s after starting the pumps, when Ce^PROPwas 0 in all patients. Averaging of the data was performed using a 10-s interval.

BIS-XP (version 4.0) was derived from the frontal electroencephalogram (At-Fpzt) and calculated by the A-2000 BIS® monitor using four BIS® sensor electrodes (Aspect Medical Systems, Inc.). The BIS value ranges from 0 to 100. The smoothening time of the BIS monitor was set at 15 s.

We recorded raw electroencephalographic data including the MLAEP data with a prototype version of the A-Line® monitor. Three electrodes (A-Line® AEP electrodes; Danmeter A/S) were positioned at mid forehead (+), left forehead (reference), and left mastoid (−). Impedance was always lower than 5 kOhm. In this protocol, MLAEP was elicited by a bilateral click at 9 Hz, with a 2-ms duration and an adaptable click intensity set automatically by the monitor according to the measured signal-to-noise ratio (SNR) of the raw MLAEP signal.19Artifact rejection and 25- to 65-Hz band-pass filtering was conducted previously to the extraction of the MLAEP.

All raw MLAEP data were stored on a compact flash card (SanDisk, Sunnyvale, CA) connected to the prototype A-Line®. Post hoc , both AAI^1.5and AAI^1.6were extracted from these data on a time synchronized basis.

The AAI^1.5is calculated using a fast extracting method called ARX, enabling us to extract information from the MLAEP within 2–6 s. This protocol has been published elsewhere.2The AAI^1.5value ranges from 0 to 100.

The AAI^1.6is a composite index using three sources of information resulting in an index value ranging between 0 and 100. First, an ARX is used to extract information from the raw MLAEP wave during periods with high SNR. The method used is based on the AAI^1.5protocol and is described elsewhere.4When the MLAEP is reduced to a flat line, the SNR becomes too low to extract a useful AAI calculation. At that time, the electroencephalographic-based information will determine the AAI^1.6calculation. Finally, when burst suppression patterns are detected, this information becomes the major factor in the calculation of the composite index. The AAI^1.6is mathematically formulated as

where AAI^1.5is the result of the ARX model. E^{30–47 Hz}and E^{10–20 Hz}are the results of a power analysis in the raw electroencephalographic spectrum using higher and lower frequencies, respectively. BS is the percentage of burst suppression patterns detected during the last 30 s. k⁰, k¹, and k²are functions of the SNR, the detection of iatrogenic artifacts (translated in a signal quality index), and the auditory stimulus intensity. The SNR is determined during the averaging process of the raw signal. It evaluates the detection quality of the signal under investigation. For MLAEP, the occurrence of the electroencephalogram, electromyogram, and other artifacts will induce a lower SNR. For the electroencephalogram, only electromyographic and iatrogenic artifacts will be taken into account. The signal quality index is calculated based on the number (and type) of artifacts during a certain period of time. If the amount of artifacts per period of time is higher than a percentage of this time, no AAI^1.6is calculated because it is not reliable. If the number of artifacts per period of time is less than this value, an index is calculated, but the signal quality index bar is decreased on screen, indicating a lower accuracy of the calculated index. When this occurs, an artifact code is transferred to the RUGLOOP program automatically. These data were deleted post hoc  from further analysis.

The burst suppression parameter is defined as the percentage of time where the power of the electroencephalogram is smaller than 5 μV over a period of time. However, the calculation of burst suppression differs between monitors. For BIS and AAI^1.5, the burst suppression algorithm details are published previously.7 

In the AAI^1.6, the burst suppression detection algorithm is based on a maximum likelihood cumsum algorithm, with variable probability functions, using a preprocessed signal after artifact detection and filtering. The main property of this method is that low-amplitude signal criteria are based on probabilities and not on a fixed number. Second, the signal has no fixed segmentation, allowing a higher time resolution for detecting changes in comparison with other methods as applied in AAI^1.5and BIS. Finally, burst suppression is evaluated over a window of 30 s in AAI^1.6, instead of 22 s in the former version.

Because it might be hypothesized that no additional information on loss of consciousness is revealed for AAI^1.6values between 60 and 100,3the A-Line® monitor offers the possibility to adjust the range of the index toward a 0–60 scale (defined as AAI1.6⁶⁰). However, by setting all values higher than 60 to 60, it has to be proven that no information on cerebral anesthetic drug effect is lost, which is crucial when using the monitor in a pharmacodynamic protocol or in a clinical setting. Therefore, applicable performance measurements (baseline variability, individualized Spearman rank correlation, prediction probability analysis) were calculated on both scales.

The significance level was set at 5% unless otherwise reported.

The baseline variability is calculated by computing the coefficient of variance on the electroencephalographic data points obtained during the first 5 s of the protocol, before any drug has been delivered. In this setting, the baseline variability will mainly reflect interpatient variability because of the short duration of measurement.

A Pearson correlation was calculated between AAI^1.5and AAI^1.6.

The relations between the burst suppression and each measure of anesthetic drug effect are plotted. A Spearman rank correlation was calculated. For each scatterplot, a model was fitted to the data using curve estimation function from Sigmaplot2000® for Windows® (SPSS Inc., Chicago, IL). The curve estimation function produces curve estimation regression statistics and related plots for different curve estimation regression models including linear, logarithmic, inverse, quadratic, cubic, power, compound, S-curve, logistic, growth, and exponential. A separate model is produced for each dependent variable, together with its regression coefficients, predicted values, residuals, and predictive intervals. After this, the most appropriate regression model can be selected.

The ability of the different indicators to describe the anesthetic drug effect was evaluated using prediction probability (P^K), which compares the performance of indicators having different units of measurements. P^Kwas calculated using a custom spreadsheet macro P^KMACRO, developed by Smith et al.  13,14The P^Kvalue was calculated for every individual patient and for all parameters studied. We evaluated the predictive capacity of BIS, AAI^1.5, and AAI^1.6for detecting the calculated Ce^PROP, which has a known correlation with the anesthetic drug effect.18The calculated Ce^PROPreached at every 5 s was used as an endpoint for P^Kcalculation. Ce^PROPwas calculated with a precision of two decimals. A P^Kof 1 for BIS, AAI^1.5, or AAI^1.6would mean that BIS, AAI^1.5, or AAI^1.6always decreases (increases) as the patient reaches higher (lower) drug concentrations according to the Ce^PROP. Such an indicator can perfectly predict the anesthetic drug concentration. Alternatively, a P^Kvalue of 0.5 would mean that the indicator is useless for predicting anesthetic drug concentration. The jackknife method was used to compute the SE of the estimate, based on the assumption that all assessments were independent.13,14After having evaluated normal distribution, a Student t  test with Bonferroni correction was used to evaluate significant difference between P^Kmeans.

As an additional nonparametric approach, the Spearman rank correlations between Ce^PROPand BIS, AAI^1.5, or AAI^1.6(both scales) were individualized by computing the correlation first for every patient i separately, defined as R  ⁱ. The reported Spearman rank correlation, R , is a weighted average of the R  ⁱ(weighted according to the number of observations for each patient). In this way, R  retained its usual interpretation. The confidence intervals on R  were obtained by the bootstrap method in which the hierarchical nature of the data were incorporated by resampling within patients. Equality of two correlation coefficients was tested at the 5% level of significance by constructing the 95% confidence intervals of the difference (confidence intervals were also computed with the bootstrap technique). All bootstrap calculations were based on 10,000 simulation runs.

The relation between Ce^PROPand the electroencephalographic measures of anesthetic drug effect were analyzed using a sigmoid E^maxmodel,

where Effect is the electroencephalographic effect being measured (BIS, AAI^1.5, or AAI^1.6[both scales]), E⁰is the baseline measurement when no drug is present, E^maxis the maximum possible drug effect, Ce is the calculated effect site concentration of propofol, Ce⁵⁰is the effect site concentration associated with 50% maximal drug effect, and γ is the steepness of the concentration–response relation curve. The model parameters were estimated using NONMEM V (Globomax LLC, Hanover, MD). Interindividual variability was modeled using a log-normal distribution,

where Pⁱis the parameter value (E⁰, E^max, y or Ce⁵⁰) in the i  th patient, P^tvis the typical value of the parameter in the population, and η is a random variable with a mean of 0 and a variance of ω.2Individual variability is reported as ω, the SD of η in the log domain, which is approximately the coefficient of variance in the standard domain. Residual intraindividual variability was modeled using a standard additive error model.20 

The population characteristics were as follows: weight, 64.0 ± 3.8 kg; age, 38 ± 6.0; height, 168 ± 7.8 cm; sex, 10 women and 3 men. All patients remained within a safe hemodynamic and respiratory clinical state.

All measured data were included in the analysis. Figure 1shows the raw data over time for the four electroencephalographic measures (BIS, AAI^1.5, AAI^1.6, AAI1.6⁶⁰) and Ce^PROP. All patients remained in a hemodynamic and respiratory safe condition during the study period.

The interpatient baseline variability was calculated for all studied electroencephalographic measures (BIS, AAI^1.5, AAI^1.6, AAI1.6⁶⁰). The coefficient of variance on the first 5 s of measurement before administration of propofol is shown in table 1.

AAI^1.5and AAI^1.6(both scaled between 0 and 100) were compared as shown in figure 2. This shows a strong correlation between both AAI versions (Pearson correlation R = 0.91209833, P < 0.001). We notice a large variability at high values of both parameters and a tendency for AAI^1.5to remain on a higher value at higher levels of Ce^PROPin comparison with AAI^1.6. For AAI^1.5, no data lower than 7 are observed in contrast to AAI^1.6.

The relations between the percentage of burst suppression and BIS, AAI^1.5, and AAI^1.6are plotted in figure 3together with the most appropriate curve estimation. For BIS, a cubic polynomial regression curve was considered to be the best fit. For AAI^1.5and AAI^1.6, a sigmoid regression curve with three parameters was selected as the best fit. For AAI1.6⁶⁰, identical results to AAI^1.6were found. Therefore, no graph is presented. The Spearman rank correlations between burst suppression and BIS, AAI^1.5, and AAI^1.6are −0.728, −0.551, and −0.871, respectively.

The P^Kvalues for BIS, AAI^1.5, AAI^1.6, and AAI1.6⁶⁰are shown in table 2. The individualized Spearman rank correlations between Ce^PROPand BIS, AAI^1.5, AAI^1.6, and AAI1.6⁶⁰are also included in table 2.

Figure 4shows the behavior of the four electroencephalographic measures versus  Ce^PROPfor all patients. With increasing Ce^PROP, all measures decreased. The relations of Ce^PROPto BIS, AAI^1.5, AAI^1.6, and AAI1.6⁶⁰are modeled and shown in figure 5. The typical parameter values for each population model including the coefficient of variance (as a measure for interindividual variability in the standard domain) are found in table 3. The SD for each model (as a measure of the intraindividual variability in the log domain) was 5.63 for BIS, 13.67 for AAI^1.5, 7.24 for AAI^1.6, and 4.02 for AAI1.6⁶⁰.

This study shows that the application of a composite index (AAI^1.6) based on MLAEPs and spontaneous electroencephalography optimizes the prediction of Ce^PROPas compared with a solitary fast extracting MLAEP measure (AAI^1.5). Comparing several monitors for cerebral hypnotic drug effect demands a systematic approach. The methodology used in this trial is based on a former publication from Vanluchene et al. ,7where spectral entropy is compared with BIS and AAI^1.5.

The A-Line® monitor offers the possibility to reduce the upper scale limit to 60, based on the rationale that values between 60 and 100 suffer from a high interpatient variability and offer no additional information on loss of consciousness.2Although this adjustment might be interesting on a clinical level, it is crucial to prove whether this scale is applicable to pharmacodynamic studies investigating subhypnotic levels of anesthesia (when consciousness is not lost), without losing any information on the propofol drug concentration. Therefore, we also investigated the performance accuracy of the AAI1.6⁶⁰as an individual index.

High baseline variation, as defined by the coefficient of variance, might decrease the predictive ability of the electroencephalogram-derived measures when used to detect cerebral drug effect, as stated by Bruhn et al.  11BIS showed the lowest baseline variability. In contrast, AAI^1.5and AAI^1.6showed a high but comparable coefficient of variance. We found a major improvement in baseline variability in AAI1.6⁶⁰in comparison to AAI^1.5and AAI^1.6. Although it is potentially interesting when applying the A-Line® monitor clinically to monitor loss of consciousness, one might question that it does not make sense to reduce baseline variability by cutting of data higher than 60, if this new scale causes a decreased sensitivity for the detection of subhypnotic levels of anesthetic drug effect. This reduced upper limit only makes sense if the other performance measures reveal a beneficial effect.

Burst suppression is a benign electroencephalographic pattern frequently seen in healthy brain at deep levels of anesthesia. Because the detection of burst suppression represents an important component to the level of hypnotic effect, it might be helpful to take into account burst suppression analysis when measuring deep levels of anesthetic drug effect because it might optimize the discriminating power of the investigated index at these levels. For BIS, burst suppression is included in the algorithm. Various authors have found that above a burst suppression ratio of 40%, a linear correlation exists between BIS and burst suppression ratio, indicating that the burst suppression ratio is the only determinant factor for BIS values below 30.7,21Our results for BIS, as shown in figure 3, are in agreement with these previously published findings. If burst suppression was excluded from the algorithm, BIS would become very resistant at excessive levels of anesthetic drug effect. In contrast, the calculation of the AAI^1.5is only based on the MLAEP. As already observed in previous work, the MLAEP flattens at patient-specific levels of anesthetic drug effect, thereby limiting its discriminating power.7–9This phenomenon results in a maximum decreased AAI^1.5“plateau” level higher than 0. As shown in figures 2 and 4, the lowest observed AAI^1.5values are ceiled around 7. Beyond this level, no additional information on cerebral drug effect can be obtained. When taking in consideration that the ED⁹⁵for loss of consciousness is around 19 for AAI^1.5, the discriminating power might become very low after loss of consciousness. For AAI^1.6, the electroencephalogram and burst suppression are included as a component of the composite index as described in the Materials and Methods section. This results in an increased discriminating power at deep levels of anesthetic drug effect. As shown in figure 3, the absolute values of AAI^1.6are able to decrease to 0 at increasing levels of burst suppression. This improvement between AAI^1.5and AAI^1.6is also reflected in the statistically significant better individualized Spearman rank correlation for AAI^1.6(table 2).

When studying performance accuracy, the question is how well the observed measure, which is the electroencephalographic response, predicts the unobserved “underlying” state of the patient, which is represented by the Ce^PROP. Therefore, P^Kwas calculated for BIS, AAI^1.5, AAI^1.6, and AAI1.6⁶⁰. The underlying theoretical background of P^Kcalculations has been published elsewhere.7In our study, we did not obtain any significant difference between P^Kvalues, although AAI^1.6tended to have the highest median value.

As an additional analysis for correlation between parameters and Ce^PROP, we calculated the Individualized Spearman rank correlation coefficients for each measure. This weighted nonparametric statistical approach depicts the nonlinearity in the system much more accurately as compared with P^Kanalyses. Results confirm the significant increase in correlation between AAI^1.6versus  Ce^PROPin comparison to AAI^1.5, thereby proving that the combination of MLAEP and derived measures from the spontaneous electroencephalogram (power spectrum analysis and burst suppression calculation) increases the accuracy to measure cerebral anesthetic drug effect. Initially, no difference was found between BIS and AAI^1.5or between BIS and AAI^1.6when scaled to 100. However, when the upper limit of the scale was reduced to 60, remarkable results were observed. Figure 4shows that a high variability and a high amount of nonlinearity are observed in all measures when scaled to 100. For AAI^1.6, one can say that the signal between 60 and 100 nearly behaves as “random noise.” By eliminating this part of the AAI^1.6range, the high variability resulting in a high level of nonlinearity is drastically reduced. This results in a very high individualized Spearman rank correlation as shown in table 2. However, one can raise the question of whether a constant number around 60 offers more information compared with the complete scale, both when predicting propofol anesthetic drug effect and when evaluating a clinical level of sedation before loss of consciousness is obtained. This must be studied in further research. We were not able to evaluate the clinical level of sedation in this population because we wanted to avoid any stimulation during induction.

Figure 5shows the pharmacodynamic modeling for all measures versus  Ce^PROP. Previously, the relation between measures of anesthetic drug effect and Ce^PROPwas observed following a sigmoid E^maxmodel. Therefore, in our study, the relation between the measures of anesthetic drug effect and Ce^PROPwas also modeled using a sigmoid E^maxmodel, and the model parameters were estimated using a population approach. In NONMEM, the parameters in the individual are weighted in a Bayesian manner toward the mean for the population, based on the variance of the individual parameters. Examining our raw data, we found for all parameters a single sigmoid regression to be adequate for describing the range of concentrations used in this study. When looking at figure 5, one can conclude that the smoothest regression is found in AAI1.6⁶⁰. This is reflected in an optimization of both the interindividual and the intraindividual variability in the population. In table 3, the coefficients of variation of the typical values of the cerebral drug effect measures reveal a major decrease in the interindividual variability of the slope (γ) when AAI1.6⁶⁰is compared to AAI^1.6. Simultaneously, a decrease in intraindividual variability at baseline is observed for AAI1.6⁶⁰, reflected by the lower SD compared with AAI^1.6. However, we want to stress the fact that the regression analysis for AAI1.6⁶⁰should be interpreted cautiously because of the effect of rejection of data on the NONMEM analysis. It might cause a distortion on the data input in such a way that the behavior of the population regression is closer to the a priori  chosen E^maxmodel. Further research is needed to clarify whether this result is an artificial overestimation of the performance parameters or whether AAI1.6⁶⁰is indeed able to maintain a high performance as a hypnotic drug effect monitor.

We conclude that the use of the composite index AAI^1.6, which combines information from the MLAEP, spontaneous electroencephalogram, and burst suppression, increases the correlation with the cerebral drug effect of propofol as compared with the AAI^1.5, which is a solitary fast extracted MLAEP index. By adjusting the upper scale limit to 60, the performance measures are optimized even more. However, the implications of this newly chosen upper scale limit should be further explored on both the clinical and the pharmacodynamic level.