Of the 25 experts who were initially invited to join the panel, 21(84%) agreed and participated. The Delphi process was completed after 3 rounds and the statements defining key terms were used by the panelists to generate consensus definitions for 3 parameters used for ICU DOS. From the formulation of 89 statements, consensus was reached for a total of 28 recommendations pertaining to the 4 main domains previously identified (Table 1). The level of evidence was categorized with GRADE whenever possible. An internal review from the panelist and an external review from a non-voting expert (TB) were used to finalise the consensus document.
This document is aimed to be reviewed after a five-year time period, or whenever new evidence from the literature could support the change of an existing recommendation.
The majority of ICU patients require sedation and previous guidelines [14,15]suggested the routine use of clinical scales in order to monitor sedation. These scales are ordinal scales that can accurately assess consciousness levels during mild to moderate sedation, but they are unreliable when consciousness has been lost. Moreover, scoring involves stimulating the patient which itself alters the level of sedation and patient comfort. We therefore reached consensus and evidence suggesting the use of pEEG when sedation scoring in not possible, such as during deep sedation (RASS -4 or -5) or under neuromuscular blockade. (Table 1 Q1.1)
A correlation between pEEG-based index (e.g. BIS, PSI, SE, qCON) values and the administered dose of intravenous and inhalational anesthetic agents has been demonstrated: the progressive deepening of sedation induces a corresponding progressive reduction in pEEG-based index values . Current data on pEEG monitoring in critically ill patients in ICU are less definitive and more controversial. During neuromuscular blockade, BIS detected oversedation in 18 deeply sedated ICU patients and was shown to highly correlate with the Sedation Agitation Score in 63 mechanically ventilated patients, although the correlation varied among medical, surgical and trauma patients. Similarly, Riker et al. found good correlation between BIS and the SAS but noted that electromyographic interference could affect the accuracy of BIS in cardiac patients receiving neuromuscular blocking agents, a group in whom residual electromyographic activity can cause spurious BIS elevations . Other DOS devices have not yet been evaluated in ICU patients receiving muscle relaxants [19,20]. However, it is important to remember that evaluation of brain-stem reflexes and response to pain stimulation remain paramount, since monitoring systems should be used to supplement, and not replace, the clinical examination. Furthermore, they can assist the clinician during the management of TBI [21,22] and to markedly reduce the total dose of sedative used to achieve the same level of clinical sedation  (Table 1 Q 1.2)
The ideal timing for initiating DOS monitoring in ICU patients has yet to be investigated, but given the risks of under- and over-sedation the experts consider that DOS monitoring should be started as soon as possible when deeper sedation than a RASS of 0 or -1 is required . (Table 1 Q1.3)
Patients admitted to ICU can require multiple interventions (e.g. central line placement, bedside tracheostomy, change of burns dressings, thoracic drainage placement) for which deep sedation, analgesia, and sometimes the administration of neuromuscular blocking agents are required. Several studies have shown an association between pEEG burst suppression and negative outcomes (e.g. delirium, duration of mechanical ventilation and mortality) [24,25]. The correlation between duration of over-sedation (in terms of cumulative time spent in BS) and outcome has not yet been studied but BIS has shown clear reliability to detect deep sedation in mechanically ventilated patients . (Table 1 Q1.4)
Patients on ECMO have numerous risk-factors for delirium such as hypoxia, reduced cerebral perfusion and the potential for vascular microemboli. Burst suppression associated with over-sedation might limit the beneficial effect of ECMO on cerebral metabolism and negatively impact patient outcomes . (Table 1 Q1.5)
Processed EEG provides a compressed and simplified view of the raw EEG signals, allowing for potential evaluation by non-neurophysiologists who can alert the neurophysiologists when required because of matters of concern [28,29].Some technical and physiological limitations of these parameters when applied to the bedside for sedation assessment should be considered: pEEG monitoring to guide sedation of patients in intensive care is not able to distinguish specific signatures of each drug used for sedation [30-32]. It requires specific knowledge to recognize the features and changes in pEEG values associated with sedation with ketamine, nitrous oxide (faster EEG oscillations, higher index values) and dexmedetomidine (profound slow oscillations, low index values in an awakable patient) . Nonetheless integration of pEEG parameters in a wide monitoring platform in addition to the raw EEG can facilitate accurate control of sedation level and differentiate between sedation and sleep [33-35]. (Table 1 Q2. 1)
Spectral Edge Frequency (SEF95) was shown to correlate with the level of sedation. In critically ill patients sedated with midazolam and morphine SEF95 was modestly able to detect deep sedation levels (area under the receiver operating characteristic curve, AUC 0.687) but was a better predictor of light sedation states (AUC 0.798) .Overall SEF has limitated value when used to titrate sedative administration. It is not influenced by drug-specific EEG changes, and may show paradoxical changes when ketamine, nitrous oxide, dexmedetomidine or a combination of drugs is used, or in the elderly or very sick patients with a low baseline voltage. As it is a summary statistic major changes in the frequency power spectrum may occur without a significant change in SEF. (Table 1 Q2.2)
Processed EEG monitors were launched primarily as ‘hypnosis monitors’ during surgery [37-39].The use of the output index ranges suggested by the manufacturer for light and deep sedation might be an effective tool as a first approach to guide and individualize sedative drug dosing schemes integrated in a multimodal monitoring strategy in critically ill patients [40,41]. Some limitations of this technology are represented by the influence of EMG/artifacts and patient conditions (e.g. brain damage) since they can alter the pEEG number [39,41-44]. Therefore, the pEEG number should be verified by the concordance with the raw EEG rhythm [45,46]. (Table 1 Q2.3)
It may be challenging for clinicians to interpret a sedation state from the unprocessed electroencephalogram in real-time. The presence of a spectrogram makes it easier to interpret the subtle changes across the range of EEG frequencies. Furthermore, by presenting longer time frames than raw EEG, slower variations occurring over time are more likely to be identified. Different sedatives, acting on different neuronal circuits by different mechanisms, have distinct EEG signatures which produce different spectrogram patterns. Propofol-induced unconsciousness for example is associated with slow-delta and alpha oscillations .EEG spectral patterns in ICU patients have a standardized nomenclature with high inter-rater agreement and can be a useful tool for EEG screening .Age and co-morbidities decrease the EEG amplitude, and weaker power in the alpha band increases the propensity for burst suppression which is a phenotype of a vulnerable, frail brain . The possibility to adjust the power scale of the spectrogram increases utility as it can increase the visibility of such “weak” bandwidths. (Table 1 Q2.4)
BS has been associated with a higher risk of delirium and mortality in critically ill patients [48,49]. It is important to distinguish unintended BS resulting from overdosing of sedative drugs, from therapeutic induced BS which might be potentially useful in situations of low cerebral blood flow and altered metabolism such as refractory intracranial hypertension and the treatment of refractory status epilepticus. Despite a lack of clear evidence to support this practice and large variability in the degree of EEG suppression achieved, BS remains incorporated in many pragmatic refractory status epilepticus treatmeant algorithms [50-54]. (Table 1 Q2.5)
Assessment of sedation
Besides the differences in pharmacodynamics and pharmacogenetics in relation to anaesthetic and sedative drugs, other factors may influence the EEG signal: cerebral blood flow, hypothermia, age, brain damage and others. When interpreting the pEEG, all possible causes of patient variability must be taken into consideration including different EEG spectra during burst and suppression periods [32,65]. (Table 1 Q3.1)
Although pEEG has the advantage of being easily interpreted by doctors and nurses who are not experts in neurophysiology, it is very vulnerable to artefacts caused electromyographic signals from shivering or facial movements or from interference from electrical signals from nearby machines (body thermoregulating systems, haemofiltration and ECMO machines) . Efforts to improve artefact detection and removal and signal-to-noise ratios are underway [57-61]. Until these systems have been validated, subjective sedation scoring systems should be considered more reproducible than pEEG for patients who are lightly sedated and for whom neurological evaluation is possible, particularly when the risks of artefact exposure are high [62,63]. (Table 1 Q3.2)
Processed EEG devices generally appear best suited for sedative titration during deep sedation or in patients who have received neuromuscular blockade, although observational data suggest potential benefit with lighter sedation as well . In non-paralyzed patients, electromyographic signals may impair the utility of the displayed index value . (Table 1 Q3.3)
Processed EEG devices cannot, and should not replace the clinical validated scales but rather be supplemental to them, since these later are globally more informative of the clinical sedation status of ICU patients. (Table 1 Q3.4)
EEG-based monitoring devices are well suited to facilitate sedative titration during deep sedation and especially when neuromuscular blocking agents have been administered (eg. in patients with ARDS, those requiring prone positioning, venous-arterial and venous-venous ECMO). . (Table 1 Q3.5)
Competences, Training and Support
The use and recourse to technologies, upon which clinicians increasingly rely on, requires adequate and appropriate training [66,67]. (Table 1 Q4.1)
Initially, clinicians require knowledge and experience in order to apprehend the principle of pEEG monitoring and to interpret the basic EEG waveforms, spectrogram and processed indices during sedation in the healthy brain and with minimal interference, for which a 30-min training session is considered sufficient . This is followed by a further period of training and experience to understand the influence of other pathologies/conditions/artifacts on the pEEG index . Finally, clinicians should be tautght to be aware of the limits and advantages of each pEEG monitor used. Frequent use of pEEG in within the ICU, combined with multidisciplinary teaching, is warranted to improve the performance of clinicians when using pEEG monitoring to manage patients and estimate their prognosis. (Table 1 Q4.2)
In recent years, experience has been gained in anaesthesia and intensive care medicine regarding the teaching of ultrasound-based techniques in relatively short periods of time (days to weeks). The use ofinteractive teaching approaches and simulation seem to be very effective [70-74].
Bombardieri and colleagues reported the application of simulation for training in pEEG and found a significant improvement in clinicians without prior EEG training in identifying EEG waveforms corresponding to different hypnotic depths and also in recognizing when the hypnotic depth suggested by the EEG was discordant with the pEEG index  (Table 1 Q4.3).
The participation of neurospecialists in the faculty of pEEG teaching and training courses provides effectiveness and quality in two respects: firstly, because of their gained knowledge of the EEG signal itself in all its aspects (basic and theoretical, physiological and pathophysiological, neurological and pharmacological) and secondly, their more advanced experience regarding EEG training [71-74,76].
Literature is divided in regards to the balance between support and continuing education, and regular recertification . It is also not clear which authority should be responsible for certification: academic, regional or even national level, professional or scientific society .
The quality of the practice of using any technology to improve patient care and outcomes must be guaranteed. Telemedicine–based solutions have been used with growing effectiveness in high as well as low- and middle-income countries. How this is implemented will depend on local circumstances [70,71].
A graphical representation (trend vs. spectrogram) represents a third reading level of the EEG that should be implemented and simultaneously visible with the raw trace and other derived parameters such as pEEG index, SR and MEF/SEF. Graphical representations convey information about the effect of general anaesthesia or sedation, the spectral signature of the drug, and patterns associated with age of the patient [71-74]. Trends or spectrograms can reveal the occurrence of excessive EEG suppression, or an increase in activity level consistent with a non-convulsive seizure episode, both of which as amenable to rapid therapeutic intervention, and specific patterns may help the clinician to assess the patient’s underlying condition and formulate a prognosis (i.e. delta/theta septic encephalopathy, renal or hepatic failure, BS due to severe brain damage) [76-79]. (Table 1 Q 4.4-9)
Agenda for future research
Depth of sedation monitoring of critically ill patients remains a challenging topic due to the contradicting results in the literature . The main barriers to the routine use of these monitors in ICU are represented by: 1) the lack of knowledge, especially outside the neurological ICU, 2) the lack of validation of the use of the monitors for prolonged sedation, 3) the lack of a standardization between monitors based on different EEG analysis algorithms, 4) the financial constraints limiting availability of the monitors and finally, and 5) the unknown effect of excessive sedation on the long-term outcome of these patients although it has been demonstrated that a prolonged duration with (bilateral) BIS 0 values serves as a better outcome predictor after OHCA as compared to a single observation . In fact, their use may become necessary in order to understand if delirium or cognitive impairments after ICU are related to oversedation or other EEG abnormalities.
Processed EEG could be a useful tool to predict outcome as BIS values are correlated with the prognosis of patients with coma in ICU, and can be a useful marker for estimating the prognosis of comatose patients  including when they are witdrawm from sedation .
This consensus suggests that the time has come to implement DOS monitoring with pEEG monitors in sedated critically-ill patients especially in those deeply sedated and/or receiving neuromuscular blocking agents. Artifacts from electrical interference from other machines, from patient movement or concomitant neurological diseases can reduce the utility of this monitoring. We therefore suggest that pEEG index values should always be interpreted in the context of the other available quantitative EEG parameters (such SEF and SR), and to assess the raw EEG trace, the density spectral array and the clinical status of the patient.