The distribution of the values of SEF95 and Ratio 10 in all of the records according to Ramsay scores is shown in Figure 3. There was a wide dispersion of the values of these parameters for a given clinical level of sedation resulting in an overlapping of values between the “deep sedation” and “light sedation” populations.
The results of the relevance assessment of each parameter regarding the prediction of levels of sedation are presented in Tables 5 through 8. These results are presented as the area under the ROC curve (AUC) with the 95% confidence interval. These calculations were performed for every parameter, for the best linear combination of two parameters, for the best linear combination of all the spectral parameters, for the best linear combination of all the bispectral parameters, and for the best linear combination of all the parameters. Table 5 shows the results of comparison of records corresponding to a light clinical level of sedation (Ramsay 1 and 2) with those corresponding to the other levels (Ramsay 3 to 6). Table 6 shows the results of comparison of recordings corresponding to a deep clinical level of sedation (Ramsay 5 and 6) with those corresponding to the other levels (Ramsay 1 to 4).
Analyses reported in Tables 5 and 6 concerned all patient records. Analyses reported in Tables 7 and 8 were performed after exclusion of the records made during periods when the COMFORT 4 score was more than 2 (n = 12).
These tables show AUC values, in some cases, higher than 0.8 for certain parameters reflecting good discriminating power. Unfortunately, this must be tempered by wide 95% confidence intervals. For all the records (n = 151), the most relevant bispectral parameter for predicting the light levels of sedation was Ratio 10 and the most relevant spectral parameter was the SEF95. Their predictive power were similar (AUC = 0.829 for Ratio 10 versus 0.798 for SEF95). Bispectral analysis does not so seem to bring more precision than simple spectral analysis in the diagnosis of states of light sedation in ICU patients.
The most relevant parameters for predicting the deep levels of sedation were also Ratio 10 and SEF95. The predictive power of ratio 10 was better than that of SEF95 (AUC = 0.763 versus 0.687, respectively). Bispectral analysis provided better diagnosis of the states of deep sedation than spectral analysis.
Both types of analysis, spectral and bispectral, were better correlated to the light levels of sedation than to the deep levels of sedation. In fact, Ramsay scale became ineffective when patients were nonreactive and these patients, who were heavily sedated (Ramsay 6), varied widely in EEG status.
When the agitated patients (n = 16) were excluded, results remained unchanged. Ratio 10 and SEF95 were highly similar in predictive power for the diagnosis of lightly sedated patients (AUC = 0.804 versus 0.801, respectively). The relevance of Ratio 10 in predicting the stage of deep sedation was also more than that of SEF95 (AUC = 0.756 versus 0.684) after exclusion of the agitated patients.
The originality of our work consisted in mathematically calculating spectral and bispectral descriptors from raw EEG data to test their relevance for assessing too light or deep sedation levels in ICU patients, with the exception of those presenting a neurological disorder. The design of our study does not allow the determination of each level of Ramsay score by a linear combination of various EEG parameters. The goal was to make the EEG diagnosis of two particular clinical situations: too light, or oversedated.
The present study yielded three primary results. First, the various calculated EEG spectral descriptors exhibited large interindividual variability. Second, bispectral analysis provided the same precision as spectral analysis in distinguishing between the stages of sedation of ICU patients. Third, there was a strong correlation between spectral and bispectral parameters of the EEG, reflecting in some way the redundant nature of the two methods. Despite the absence of statistical significance, the analysis of AUC (0.804 for Ratio 10 to 0.919 for all the EEG parameters) shows an improvement in discriminative power of a linear combination of these EEG spectral descriptors into a multivariate index.
Among various spectral parameters, the SEF95 was the most relevant for predicting the level of sedation. SEF95 exhibited a high AUC (0.687) corresponding to the prediction of deep sedation levels and an even higher AUC (0.798) correlating it with states of light sedation. This result is consistent with different studies performed in patients who were sedated by midazolam (24,25). Spencer et al. (26) found no consistent correlation between the level of sedation and any single EEG spectral parameter. By performing multivariate analysis using parameters obtained from the power spectrum (e.g., tenth percentiles, mean frequency), those authors obtained a better correlation among these variables and clinical levels of sedation. They observed substantial overlap between the stages of sedation. They concluded that the substantial variability of EEG parameters among the patients and within individual patients would explain why simple spectral parameters of EEG cannot discriminate between levels of sedation. Similarly, in the present study various EEG parameters showed marked variability with an overlapping of values between the stages of light and deep sedation. This variability of EEG parameters constitutes a limitation of the EEG technique for monitoring anesthesia or sedation (9,25–27).
Bispectral analysis could improve the precision of monitoring because it theoretically provides a more complete analysis of the modifications of EEG signal in response to the administration of anesthetic drugs. We reported in our study that the most relevant bispectral parameter was Ratio 10 (AUC = 0.763 for prediction of deep sedation and 0.829 for light sedation). During the deepening of sedation, the value of the Bic function increases preferentially in low frequencies (10). Ratio 10 quantifies this tendency for relative variation in value of the Bic function in favor of low frequencies. However, the method consisting of correlating variations of sedation to variations in degree of phase coupling is not at present supported by any clearly established neurophysiologic mechanism (8,9). The first assessment of EEG bispectral analysis was performed using BIS® (Aspect Medical Systems, Natick, MA). The BIS® is a nondimensional index obtained from three EEG subparameters, the relative contributions of which are correlated to the level of anesthesia using a multivariate regression model (9,10). Studies comparing BIS® with conventional spectral parameters in the determination of the level of sedation (13,25,28–30) have established the superiority of BIS®, notably regarding the SEF95. However, the relevance of BIS® in intensive care is disappointing and the correlation between BIS® and clinical level of sedation remains weak to moderate (11–15). Moreover, all authors note large interindividual variability of the values of BIS® for a given clinical stage of sedation. Several authors (12,13) have reported that BIS® can be altered by the EMG. Vivien et al. (16) quantified the interference between EMG activity and BIS® in 45 heavily sedated ICU patients who had no neurological disorder. The administration of muscle relaxant to these patients significantly reduced the values of both BIS® (67 ± 19 versus 43 ± 10; P < 0.001) and EMG activity (37 ± 9 versus 27 ± 3 dB; P < 0.001). This interference of the BIS® by EMG activity fosters false diagnoses of insufficient sedation, leading to a risk of unjustified oversedation by the clinician. The overlap between the bandpasses of EEG signal (conventionally 0.5 to 30 Hz) and EMG activity (conventionally 30 to 300 Hz) results in artifactual contamination of the EEG signal processed by the BIS® monitor, overestimating the value of the BIS® in patients not receiving neuromuscular blockers. In the present study, we did not measure EMG activity, but exclusion of a subgroup of agitated patients did not change our results. This might be attributable to the efficacy of the analogical and digital filtering that we used before signal processing and because our analysis of EEG signal concerned a bandpass of which the superior limit of 30 Hz was lower than that used with BIS® software (47 Hz). These studies question the validity of BIS® in ICU patients in some cases, even when the most recent software versions (BIS® XP monitor) are used (16). Our method is similar to the one used to develop the BIS® except that, in our calculations, we did not include parameters accounting for the degree of burst suppression. Indeed, this level of depression of cortical activity is excessive in the framework of simple sedation. In addition, visual inspection of the raw EEG records of our study showed no isoelectric period.
The sedation protocol of our patients associated a benzodiazepine and an opioid. This type of bimodal sedation is the most used in European ICUs (1). Benzodiazepines have an EEG action different from the action of other anesthetic drugs. Although anesthetic drugs have a biphasic EEG action (initial phase of acceleration of the EEG followed with a deceleration and synchronization dosage-dependent phase), benzodiazepines only accelerate the EEG (β activation) (31). However, Greenblatt et al. (32) found that the increased β power band was lost when the subject slept. The implication is that an optimal EEG spectral parameter (SEF95, SEF50) for midazolam-based sedation might well not be optimal for sedation with propofol or other anesthetics. However, there are no comparative data concerning EEG bispectral parameters to evaluate sedation with these drugs. EEG modifications of opiates look like those recorded during sleep: deceleration and appearance of slow long waves (synchronization). These effects are modest with morphine (33).
We have chosen, in our study, the Ramsay and the COMFORT scales for the clinical evaluation of the level of sedation. Although the Ramsay scale was the first published (1974), it remains relevant (34) and is the most widely used scale in studies on the sedation in ICUs. It shows good interrater reliability with the other validated sedation scales in adult intensive care (35). We transformed the Ramsay scale into 2 Boolean variables: Light sedation (Ramsay 1 or 2) versus other levels of sedation (Ramsay 3, 4, 5, or 6) and deep sedation (Ramsay 5 or 6) versus other levels of sedation (Ramsay 1, 2, 3, or 4). This transformation permitted more reliable statistical comparison between only two, more distinct groups. We used the COMFORT scale (Table 2) it in our work to assess the spontaneous movements of the patients and isolate a subgroup of patients with frequent movements (COMFORT 4 > 2).
There was a predominance of deep sedation in the 167 raw EEG records (Fig. 2). This reflected the fact that, irrespective of instrumental monitoring of sedation, diagnosis of levels of insufficient sedation is more straightforward (e.g., agitation, pulling at endotracheal tube). Consequently light levels of sedation are easier to manage than deep levels. Overly deep sedation is frequent in ICUs, especially in surgical or polytrauma patients. Clinical scores of sedation become ineffective when the patient is nonreactive and excessive sedation is often diagnosed by delayed recovery of consciousness when medication is discontinued (1,30) The exclusion of the 16 records with excessive interference from patient movements essentially eliminated records corresponding to light levels of sedation, aggravating the imbalance between the groups. Overall, among 151 records studied, 19 corresponded to light levels of sedation (Ramsay 1 and 2) and 93 to deep levels of sedation (Ramsay 5 and 6). This imbalance reduced the statistical power of the study.
Another methodological limitation of the present study is that the study population was a convenience sample (essentially involving the availability of the materials) and not a truly random population. However, the conditions of the study simulated usual and customary bedside recording by a caregiver. In addition, for reasons of convenience (reclining position of the subjects, often hairless frontal region) a single frontocentral dipole (Fz-C3 or Fz-C4) was used. Although regional variations in EEG changes during sedation or anesthesia have been reported, practical applications remain scarce (36).
In conclusion, our results show that 60% of the patients had over-sedation (Ramsay 5 or 6). The subjective clinical techniques of assessment of sedation are not able to detect this over-sedation. The objective techniques based on spectral analysis and, more recently, bispectral analysis of spontaneous EEG might offer a solution. SEF95 and Ratio 10 are, respectively, the most relevant spectral and bispectral parameters for monitoring the level of sedation. Our results show that bispectral analysis of the EEG provides a slight improvement over simple spectral analysis in terms of predictive power despite the higher order of calculations involved. In addition, we noted a hindrance that has long been reported in the use of EEG for monitoring in anesthesia and intensive care: the large interindividual variability of results. Further studies will be necessary to evaluate continuous neuromonitoring in intensive care, not by reasoning in terms of a general population but in terms of individual clinical courses over time using basic EEG. Neither recommendations nor goals for EEG-guided midazolam-based sedation of a given ICU patient may be drawn from the present study.
The bispectrum of the temporal signal x(t) is a function of two frequencies (f1 et f2). The equation used to calculate the bispectrum B(f1,f2) is (10):
where the subscript i refers to the epoch number, L is the sum of the epochs, Xi(f) is the Fourier transformation of the ith epoch and Xi*(f1 + f2), the conjugate of Xi(f1 + f2).
The amplitude of the bispectrum B(f1,f2) is influenced by the degrees of phase coupling between each possible frequency pair f1, f2 and the frequency corresponding to their sum (f1 + f2), as well as by the power of the spectrum at frequencies f1, f2 and (f1 + f2). Consequently, it is necessary to normalize the bispectrum by calculation of the bicoherence: Bic(f1,f2). Bic(f1,f2) is obtained by dividing B(f1,f2) by the real triple product: RTP(f1,f2).
where P(f) is the power of the spectral component of frequency f.
This means that, independently of the signal power, Bic(f1,f2) expresses the degrees of phase coupling in a range from 0% (absence of phase coupling) to 100% (completely phase coupled signal).
The Ratio 10 represents the ratio of the bicoherence function in the band 1–10 Hz with respect to the total band (1– 30 Hz):
1.Soliman HM, Melot C, Vincent JL. Sedative and analgesic practice in the intensive care unit: the results of a European survey. Br J Anaesth 2001;87:186–92.
2.Park G, Coursin D, Ely EW, et al. Balancing sedation and analgesia in the critically ill patient. Crit Care Clin 2001;17:1015–27.
3.Freire AX, Afessa B, Cawley P, et al. Characteristics associated with analgesia ordering in the intensive care unit and relationships with outcome. Crit Care Med 2002;30:2468–72.
4.Jacobi J, Fraser GL, Coursin D, et al. Clinical practice guidelines for the sustained use of sedatives and analgesics in the critically ill adult. Crit Care Med 2002;30:119–41.
5.Kress JP, Pohlman AS, O’Connor, MF, Hall JB. Daily interruption of sedative infusions in critically ill patients undergoing mechanical ventilation. N Engl J Med 2000;342:1471–7.
6.Nasraway SA Jr, Jacobi J, Murray MJ, Lumb PD. Sedation, analgesia and neuromuscular blockade of the critically adult: revised clinical practice guidelines for 2002. Crit Care Med 2002;30:117–8.
7.Hug CC Jr. Monitoring. In: Miller RD, ed. Anesthesia, 2nd ed. New York: Churchill Livingstone, 1986:411–63.
8.Pandin P. Le neuromonitorage de la sédationIn: Guérit JM. éd.. L’évaluation neurophysiologique des comas, de la mort encéphalique et des états végétatifs. Marseille: Solal, 2001:213-34.
9.Rampil IJ. A primer for EEG signal processing in anesthesia. Anesthesiology 1998;89:980–1002.
10.Sigl JC, Chamoun NG. An introduction to bispectral analysis for the electroencephalogram. J Clin Monit 1994;10:392–404.
11.Simmons LE, Riker RR, Prato BS, Fraser GL. Assessing sedation during intensive care unit mechanical ventilation with the bispectral index and the Sedation–Agitation Scale. Crit Care Med 1999;27:1499–504.
12.Riess ML, Graefe UA, Goeters C, et al. Sedation assessment in critically ill patients with bispectral index. Eur J Anaesth 2002;19:18–22.
13.Riker RR, Fraser GL, Simmons LE, Wilkins ML. Validating the Sedation–Agitation Scale with the bispectral index and Visual Analog Scale in adult ICU patients after cardiac surgery. Intensive Care Med 2001;27:853–8.
14.Frenzel D, Greim CA, Sommer C, et al. Is the bispectral index appropriate for monitoring the sedation level of mechanically ventilated surgical ICU patients? Intensive Care Med 2002;28:178–83.
15.Nasraway SA, Wu EC, Kelleher RM, et al. How reliable is the bispectral index in critically ill patients? A prospective, comparative, single-blinded observer study. Crit Care Med 2002;30:1483–7.
16.Vivien B, Di Maria S, Ouattara A, et al. Overestimation of bispectral index in sedated intensive care unit patients revealed by administration of muscle relaxant. Anesthesiology 2003;99:9–17.
17.Ramsay MA, Savege TM, Simpson BR, Goodwin R. Controlled sedation with alphaxalone-alphadolone. Br Med J 1974;2:656–9.
18.Ambuel B, Hamlett KW, Marx CM, Blumer JL. Assessing distress in pediatric intensive care environments: the COMFORT scale. J Pediatr Psychol 1992;17:95–109.
19.Jasper H. Report of committee on methods of clinical exam in EEG. Electroencephalogr Clin Neurophysiol 1958;10:370–5.
20.Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839–43
21.Zar JH. Biostatistical analysis, 4th ed. Upper Saddle River, NJ: Prentice Hall, 1998:138–46.
22.Kramar A, Faraggi D, Fortune A, Reiser B. mROC: a computer program for combining tumour markers in predicting disease states. Comput Methods Programs Biomed 2001;66:199–207.
23.Le Gall JR, Lemeshow S, Saulnier F. A new simplified acute physiology score (SAPS II) based on a European/North American multicenter study. JAMA 1993;270:2957–63.
24.Veselis RA, Reinsel R, Marino P, et al. The effect of midazolam on the EEG during sedation in critically ill patients. Anaesthesia 1993;48:463–470.
25.Liu J, Singh H, White PF. Electroencephalogram bispectral analysis predicts the depth of midazolam-induced sedation. Anesthesiology 1996;84:64–9.
26.Spencer EM, Green JL, Williats SM. Continuous monitoring of depth of sedation by EEG spectral analysis in patients requiring mechanical ventilation. Br J Anaesth 1994;73:649–54.
27.Schraag S, Mohl U, Bothner U, Georgieff M. Clinical utility of EEG parameters to predict loss of consciousness and response to skin incision during total intravenous anaesthesia. Anaesthesia 1998;53:320–5.
28.Struys M, Versichelen L, Mortier E, et al. Comparison of spontaneous frontal EMG, EEG power spectrum and bispectral index to monitor propofol drug effect and emergence. Acta Anaesthesiol Scand 1998;42:628–36.
29.Sleigh JW, Donovan J. Comparison of bispectral index, 95% spectral edge frequency and approximate entropy of the EEG, with changes in heart rate variability during induction of general anaesthesia. Br J Anaesth 1999;82:666–71.
30.De Deyne C, Struys M, Decruyenaere J, et al. Use of continuous bispectral EEG monitoring to assess depth of sedation in ICU patients. Intensive Care Med 1998;24:1294–8.
31.Buhrer M, Maitre PO, Hung O, Stanski DR. Electroencephalographic effects of benzodiazepines. I. Choosing an electroencephalographic parameter to measure the effect of midazolam on the central nervous system. Clin Pharmacol Ther 1990;48:544–54.
32.Greenblatt DJ, Ehrenberg BL, Gunderman J, et al. Pharmacokinetic and electroencephalographic study of intravenous diazepam, midazolam and placebo. Clin Pharmacol Ther 1988;45:356–65.
33.Smith NT, Dec-Silver H, Sandford TJ, et al. EEGs during high dose fentanyl-, sufentanyl-, or morphine-oxygen anesthesia. Anesth Analg 1989;63:386–93.
34.Ramsay M. Measuring level of sedation in the intensive care unit. JAMA 2000;284:441–2.
35.De Jonghe B, Cook D, Appere-De-Vecchi C, et al. Using and understanding sedation scoring systems: a systematic review. Intensive Care Med 2000;26:275–85.
© 2005 International Anesthesia Research Society
36.Levy WJ. Power spectrum correlates of changes in consciousness during anesthetic induction with enflurane. Anesthesiology 1986;64:688–93.