Entropy and bispectral index for assessment of sedation, analgesia and the effects of unpleasant stimuli in critically ill patients: an observational study

Pain, physical discomfort and anxiety are common in critically ill patients. The underlying disease, care procedures, prolonged immobility and sleep deprivation all contribute to this [1, 2]. Both the stress response and its treatment may have a negative impact on outcome [3‐9]. Strategies aiming to reduce the amount of sedatives and analgesics administered may improve outcome and reduce the need for mechanical ventilation [8, 10, 11]. Accordingly, a reliable, objective assessment of sedation and analgesia during the course of critical illness would be very valuable.

Ideally, sedation in the intensive care unit (ICU) should result in a calm patient who can easily be aroused and has a maintained sleep-wake cycle. Reaching this ideal target is difficult [12], and some patients require deeper levels of sedation, for instance to facilitate circulatory and respiratory support [12]. In addition, patients' requirements for sedative and analgesic drugs vary substantially during the disease process and during therapeutic and supportive interventions.

The clinical assessment of sedation relies on the patient's response to external stimuli. However, the stimulus itself alters the patient's level of sedation. Monitoring electroencephalogram (EEG)-based variables can allow continuous assessment of the level of sedation, and thereby predict the patient's responsiveness. Methods and devices based on processed EEG signals are widely used to monitor the depth of anaesthesia. They have also been advocated for monitoring sedation in intensive care, although the results are controversial. Early observational studies found a good correlation between the Sedation Agitation Scale and bispectral index (BIS) or entropy values [13, 14], but other reports could not confirm better performance when compared with standard subjective assessment scores [15‐19]. A major drawback of these studies was the fact that the assessment of the sedation score and comparison with the EEG was done in patients who were clinically stable and did not have adjustments to sedation before the assessment. This reflects the difficulties of incorporating processed EEG variables into sedation protocols, because in everyday practice patients need sedation adjustment. The need for these adjustments is usually evaluated by the care teams, with the bedside nurse having a leading role in this assessment because they are with the patient most of the time. The so-called 'gold standard' of sedation can thus be considered to be protocol guided, with goals established by the physician and adjustments made by the bedside nurse.

In clinical routine, many other parameters are used (together with or without a sedation score) to decide whether analgesic or sedative drugs should be administered (including haemodynamic parameters, previous reactions to similar interventions, and sympathetic and parasympathetic reactions). In our experience, these variables do not necessarily correlate with Ramsay Sedation Scale (RSS) score. Reducing the whole sedation process to a single number is not promising; we therefore aimed to describe the indications for drug administration, and monitored patterns in clinical signs and EEG, in order to evaluate whether these patterns can predict the responses in EEG variables. We believe that it is useful to characterize how different interventions and their combinations affect EEG variables in the real-world ICU environment. Studies such as ours can determine the potential of these variables for monitoring various aspects of sedation and analgesia in the context of unpleasant stimuli.

The aim of this observational study was to evaluate different processed EEG parameters as predictors of response to sedative and opiate drugs and intratracheal suctioning, alone or in combination with drugs, during nurse-driven, protocol-guided sedation and analgesia. The interventions were administration of a sedative drug or opiate, clinically indicated endotracheal suctioning, and a combination of both. Specifically, we evaluated whether BIS and state entropy monitoring allow detection of clinically relevant distinctions between light and deep grades of sedation, and help to predict the response to unpleasant care interventions. We hypothesized that there are thresholds beyond which drugs and intratracheal suctioning do not result in significant changes in the respective processed EEG parameters, and that the thresholds for reactions to intratracheal suctioning are modified by prior drug application.

The study was approved by the ethics committee of the Canton of Bern, and written informed consent was obtained from the next of kin and, if possible, from the patient after recovery. Inclusion criteria were mechanical ventilation for 48 hours or less and expected need for further ventilation for at least 24 hours. Exclusion criteria were need for muscle relaxation, traumatic brain injury, deep coma due to intoxication or neurological injuries, severe neuropathies or myopathies, and surgery using cardiopulmonary bypass without confirmation of normal neurology before inclusion.

Routine haemodynamic monitoring and treatment were performed according to the decision of the treating physician and standard protocols. In addition, a Datex-Ohmeda S/5 Monitor (Datex-Ohmeda, GE, Helsinki, Finland) was used for measurement and storage (via WinCollect^® software [Datex-Ohmeda, GE Healthcare, Helsinki, Finland]) of the following parameters: heart rate, arterial blood pressure (systolic, diastolic and mean), pulse oximetry, end-tidal carbon dioxide tension, and respiratory pressures and volumes. BIS-Index, a processed EEG [20], was recorded via the BIS-Module of the S/5 monitor (XP-Level, smoothing time 15 seconds, using Quattro^® Sensor [Datex-Ohmeda, GE Healthcare, Helsinki, Finland]). Entropy is an EEG-derived parameter that uses nonlinear statistics to describe the order of random repetitive signals. The Entropy^® Module (Datex-Ohmeda, GE) calculates two indices: the state entropy (SE) and the response entropy (RE). The RE includes additional information about the electromyographic activity (activity higher than 32 Hz) of the face muscles [21]. The SE (range 90 to 0) and the RE (range 100 to 0) are normalized in such a way that the RE becomes equal to the SE when there is no electromyographic activity [22]. Both EEG sensors were attached on the patient's forehead in accordance with the manufacturer's recommendations. BIS and entropy sensors were randomly attached on both sides with the Fz electrode to the upper and lower forehead, respectively.

A simple computer program (annotation board) was developed to help nurses to record the following interventions, defined as events: sedative and analgesic drug bolus, increasing or decreasing continuous sedation and analgesia, intratracheal suctioning, and other potentially painful interventions (for example, chest tube insertion). In addition, the reasons for pharmacological interventions were recorded, as follows: agitation with threat to patient or nurse; agitation; insufficient sedation according to prescription; under-sedation/medical reasons (fighting the ventilator, heart-lung interaction); reduction because of over-sedation level according to prescription; anticipated painful stimulus; pain, as either indicated by the patient or perceived by the nurse subjectively, or based on vegetative signs exhibited by the patient; or opiates to sedate the patient.

A web camera with movement detector was attached above the patient's bed to facilitate the post hoc identification of the exact time of the event. Sedative and analgesic drugs were given in accordance with a standard protocol, using sedation goals (RSS score [23]) and regular assessment of sedation and pain at 2-hour intervals or more frequently. Standard doses of fentanyl were 25 to 50 μg, of midazolam were 1 to 2 mg, and of propofol were 10 to 20 mg. If more than six boli were needed in a 4-hour period, continuous infusion of the respective drug was started. A daily sedation stop was conducted unless the attending physician explicitly ordered otherwise. Reduction in continuous medication at 2-hour intervals was encouraged. Screening for delirium was not routinely conducted at that time, and so only overt delirium was detected, but no patient in this study received an antipsychotic drug (haloperidol). All medications were prescribed by the treating physician and applied by the bedside nurse, both of whom were blinded to the EEG parameters. The bedside nurse was free to administer drugs within the prescribed limits before a painful stimulus, such as intratracheal suctioning. The main reasons for administering drugs were anticipation of arterial oxygen desaturation, pain, or heart-lung interaction. The EEG-derived variables (BIS-Index, RE, SE, 60 sec mean values) and physiological parameters were recorded continuously, and were analyzed at 30 and 2 minutes before the event (time points -30 and -2), and at 2, 5 and 10 minutes after the event (time points +2, +5, +10).

The study was performed for 24 hours or until extubation, if earlier. Afterward, the camera recordings were analyzed and any missing annotations were completed. For all recorded haemodynamic, respiratory and neurological parameters, mean values over 60 seconds were calculated at 30 and 2 minutes before the intervention (stimulus or drug) and at 2, 5 and 10 minutes after the intervention. Because the events were not planned, RSS score were not available at all time points of EEG processing. Only RSS scores assessed shortly before the event (< 15 minutes) were used for further analysis.

Fifty-one patients were included in the study (Table 1). Seven patients were excluded after the study because of withdrawal of informed consent (n = 1), insufficient EEG quality (n = 4) and intermittent, unanticipated use of muscle relaxants (n = 2). The median recording time was 23 hours (from 12:30 to 27:10 hours). Altogether, 1,722 events were identified, of which 388 (23%) had to be excluded from analysis, mostly because of missing annotations and failure to classify the event clearly despite using video recordings. For in-depth analyses, we considered the 407 endotracheal suctioning episodes, the 417 sedation boli and the 378 opiate boli (Figure 1). RSS score assessments in close proximity to time point -2 (2 minutes before an event) were available for 695 events. Events with low incidences (< 5% of total) were excluded from detailed analysis (Figure 1).

In the present observational study, which included additional verification through a camera, we have created an unprecedented and large database of sedation events and unpleasant stimuli in real-life patients. Furthermore, the decision not to allocate study personnel for bedside annotations has minimized effects of the study set-up (per se) on the nurses' decisions. These data probably represent the largest study of EEG-derived parameters in an ICU population outside the setting of a controlled study. Sedation is a multidimensional concept, encompassing consciousness, amnesia, arousal, analgesia and other parameters, and is difficult to represent using a single scale. Failure in clinical practice to capture all aspects of sedation and analgesia within a sedation scale is an unsolved problem, as corroborated by the present study. The RSS score [23], for example, is unbalanced and favours sedation aspects; the Richmond Agitation and Sedation Score [17] and the Sedation-Agitation Score [24] are more balanced, but they lack the means to detect delirium or pain as a cause of agitation. No score can predict arousal in ICU patients. It is possible that failure to monitor all aspects of sedation in the present study accounts for the proportion of missing details regarding reasons for drug administration, and especially the large proportion of additional comments given as reasons for opioid administration. For example, nurses gave medication for the indication 'heart-lung interaction/fighting the ventilator' based on their observations of patient response during previous interventions despite deep sedation. The relatively high frequency of opiate and sedative administration despite deep sedation in anticipation of interventions and for diverse medical reasons (for instance, fighting ventilator and heart-lung interactions) represents further evidence of the problems associated with current sedation scores.

Taking a broader view, this might be the reason why, even now, not all sedated patients in the ICU are monitored and guided using a sedation score, as was recently confirmed by Payen and coworkers [25]. As pointed out by Carlon and Combs [26], 'If you cannot measure it, you cannot improve it.' Unfortunately, with the use of our sedation protocol, EEG-derived parameters do not add helpful information in terms of guiding the administration of sedatives or analgesics for the most frequently occurring indication, namely agitation. With respect to the second most used indication for sedatives and opiates, namely anticipated pain during nursing or endotracheal suctioning, neither EEG-derived parameters, an increase in these parameters before suctioning, nor RSS score can identify which patients might profit from prophylactic drug administration. Because patients receiving prophylactic drugs have statistically significant but clinically only slightly worse lung function parameters, and because the EEG-derived parameters reach their baseline levels faster after drug administration, we could speculate that nurses have previously noted a clinical benefit for some patients and have used prophylactic drugs in those patients who benefit most. However, these patients cannot be identified using processed EEG parameters, and it is likely that prophylactic use of drugs is an aspect of individual nurse behavior and has no rationale. For the third most often used indication for sedation, namely fighting the ventilator/heart-lung interaction, we also identified a correlation between lower EEG-derived parameters and lower sedation levels, but these parameters are unhelpful in deciding whether to extend the sedation protocol, because neither the clinical score nor the EEG parameters identify the indication and precise time point when the drug should be given.

Regarding endotracheal suctioning, it was surprising that patients who received drugs before the unpleasant stimulation had a lower RSS score than those who did not receive the medication, reflected in higher processed EEG variables, although both of these associations were not statistically significant. They also had a slightly lower EEG reading at +2 minutes as compared with -2 minutes, which mirrors the effect of the sedation bolus. It could be argued that patients receiving medication before the event have a lower EEG reading, and that this is certainly due to the medication. However, if the time between -30 minutes and -2 minutes is taken into account, then the principal component of the rise in RE/SE/BIS lies before the intervention, and so the need for suctioning elicits more arousal than the suctioning itself. In turn, the use of medication to attenuate the response to suctioning is less responsible for the return of RE/SE/BIS to baseline than the cessation of suctioning itself. So, it may be debated whether prophylactic use of drugs before suctioning should be limited to special groups of patients who cannot tolerate suctioning, such as those with heart-lung interactions or high intracranial pressure. This might reduce the total amount of drugs given to the patients and therefore decrease the total ventilation time, as demonstrated by various investigators [8, 11].

The wide overlap of the parameters RE/SE and BIS-Index precludes the use of these variables as crude parameters for discrimination of light/moderate/deep sedation in our patient population. After initial enthusiasm over the use of the BIS-Index as a parameter of sedation in ICU patients [13, 14], confirmatory studies have found the wide overlap of the BIS-Index to be problematic, although the BIS-XP technology can identify and better integrate artifactual EEGs in ICU patients [15‐17]. Still less has been published with regard to Entropy^® in ICU patients, but its use in this patient population was also discouraged in a recent report [18].

The strength of studying real patients and patient-nurse interactions is also a potential weakness of this study. Adherence to the sedation protocol was not stringent, and a significant portion of drug administration was recorded only in the nurses' notes and not in the annotation board, especially with regard to analgesic drug administration.

The lack of RSS scores collected concomitantly with processed EEG variables at all recording time points is also a limitation of the study. Concomitant assessment of the clinical degree of sedation and EEG parameters would have allowed the relationship between the two to be addressed in greater detail. Our study design did not allow this because the 'event' could not be precisely anticipated. In addition, evaluating the RSS score changes the EEG per se. There is a wide variation in methods of timing and interpretation of EEG in conjunction with clinical sedation assessment in the literature. Some authors used a steady state at least 15 minutes from the event, and manually averaged EEG values were only used when there was a stable period [13]; others collected the EEG values during the assessment [14, 17] or before assessment [15, 18], and still others took values only if the patient was not arousable (at RSS score 6) [16].

Unpleasant or painful stimuli and use of sedative and analgesic drugs are associated with significant changes in processed EEG parameters. However, clinical indications for drug administration were not reflected by these EEG parameters, and were barely reflected by sedation level before drug administration or tracheal suction. The use of a sedation score, as recommended in a recent guideline [12], is far from perfect, and the need for sedation in special circumstances such as heart-lung interactions or when patients fight the ventilator is not reflected in sedation scores. Given that the poor quality of sedation and difficulties in reaching and maintaining sedation targets cannot be resolved with currently available processed EEG methods or scores, how to achieve optimal sedation remains a major problem in the ICU.