Processed Electroencephalogram Monitoring and Postoperative Delirium: A Systematic Review and Meta-analysis

DELIRIUM occurs in approximately 15 to 20% of major operations in patients at least 65 yr of age.¹  Postoperative delirium is associated with prolonged length of stay, increased rates of institutionalization after discharge, mortality, and long-term postoperative cognitive dysfunction.²  The presence of postoperative delirium increases healthcare expenditure by $16,303 to $64,421 per patient, and the burden of delirium on the healthcare system ranges from $38 billion to $152 billion per year.³ 

Delirium is a geriatric syndrome with baseline patient vulnerability factors augmented by precipitating risk factors, which include both medical hospitalizations and surgical events. Baseline conditions associated with postoperative delirium include increasing age, preexisting cognitive impairment, functional impairment, sensory impairment, and institutional residence.⁴ 

In the surgical population, identification of modifiable, precipitating perioperative factors for postoperative delirium is critical. These include inadequately controlled pain, dehydration, anemia, and electrolyte abnormalities, etc.^5–7  The type, duration, and depth of anesthesia have come under scrutiny as possible factors contributing to changes in cognition.

There are several United States Food and Drug Administration–approved processed electroencephalogram monitors that are marketed as monitoring depth of anesthesia, with the two most common being the Bispectral Index (BIS) monitor (Medtronic/Covidien, USA) and the SEDline monitor (Masimo Corporation, USA).⁸  A dimensionless number (BIS or patient state index) is calculated ranging from 0, no brain activity, to 100, awake.⁹  BIS values between 40 and 60 allow sufficient anesthesia for surgery and prevention of intraoperative awareness; lower processed electroencephalogram values are associated with a deep hypnotic state.¹⁰  Other intraoperative processed electroencephalograms include the GE Datex-Ohmeda Entropy (GE Healthcare, USA), Narcotrend Compact M (Narcotrend-Gruppe, Germany), and SNAPII (Everest Biomedical Instruments, USA). Another methodology uses evoked electrical activity monitors, such as the A-line monitor (Odense, Denmark), which measures auditory evoked potentials.¹¹  The algorithms and underlying data for these indices are proprietary, and despite their development for guiding anesthesia, their utility is not clearly defined or validated.^12,13 

Whether processed electroencephalogram monitors have consistent perioperative benefits remains to be determined. There is evidence that they reduce anesthetic dose and recovery time, but their effect on intraoperative awareness and postoperative recall is controversial.^14,15  A few randomized controlled trials suggest there may be a lower incidence of postoperative delirium, as well as postoperative cognitive dysfunction with processed electroencephalogram-monitored care. The United Kingdom’s National Institute for Health and Care Excellence published 2012 guidelines recommending the use of processed electroencephalogram monitoring, especially in “high-risk” patients to improve cognitive outcomes.¹⁶  However, this recommendation is based on limited data, and there is no published meta-analysis that evaluates all trials for either postoperative delirium or postoperative cognitive dysfunction to date.

In the United States, the most recent practice advisory from the American Society for Anesthesiology (ASA) regarding processed electroencephalogram monitoring advocates its use on a “case-by-case” basis and does not address the issue of cognition.¹⁷  Therefore, this systematic review and meta-analysis was designed to examine the relationship between processed electroencephalogram monitoring and postoperative delirium/postoperative cognitive dysfunction, specifically to determine whether its intraoperative use has utility in minimizing the occurrence of postoperative delirium/postoperative cognitive dysfunction.

The primary objective of this review was to assess whether there was a significant association between the use of processed electroencephalogram and postoperative delirium/postoperative cognitive dysfunction. Inclusion criteria were randomized controlled clinical trials that provided original data, patients at least 18 yr old, and randomized to intervention use of a processed electroencephalogram during surgery versus routine nonuse of the monitor or used processed electroencephalogram to target different output values (high vs. low target values). Postoperative delirium or postoperative cognitive dysfunction must have been stated as a primary outcome and measured by a validated scale. The type of anesthetic (such as regional vs. general anesthesia) was not an exclusion criterion because we aimed to understand the potential effect of processed electroencephalogram on postoperative cognitive outcomes in any patients who require the care of an anesthesia team.

An electronic search was completed on Ovid MEDLINE, PubMed, EMBASE, Cochrane Library, and Google search without date restrictions using the following search terms: processed electroencephalogram, Bispectral Index, postoperative delirium, and postoperative cognitive dysfunction. No results in foreign languages were returned; all included articles were published in the English language. The search of these electronic databases identified 369 results, which were then pooled and duplicates were removed. One reviewer (K.K.M.) screened the remaining 265 abstracts and removed any study not meeting the inclusion criteria, specifically removing all abstracts that were not randomized controlled clinical trials or did not identify processed electroencephalogram or BIS as an independent variable and postoperative delirium or postoperative cognitive dysfunction as a dependent variable. Two reviewers (K.K.M. and J.M.L.) independently assessed the remaining 38 full-text publications to ensure that they met inclusion criteria. Both reviewers mutually agreed upon the 13 publications selected for inclusion; any disagreements were resolved by discussion (fig. 1).

Two reviewers (K.K.M. and J.M.L.) independently assessed the methodologic quality of selected studies, and discrepancies were resolved by discussion. Studies were given ratings of A (adequate), B (unclear), or C (inadequate) in three different categories: randomization, allocation concealment, and selection bias. The criteria were based on the Cochrane Risk of Bias scale and are defined in the appendix. The reviewers then used three different scales to assess each study: Physiotherapy Evidence Database scale,¹⁸  Jadad scale,¹⁹  and the Cochrane Risk of Bias assessment (table 1).²⁰ 

A data extraction form was designed to include study design (including quality measures outlined above), independent and dependent variables, anesthetic protocol, number of participants, and measured outcomes, including mean BIS with SD, incidence of postoperative delirium, and/or postoperative cognitive dysfunction between study groups with odds ratios (OR), and other publication-specific reported outcome measures at specified time points such as cognitive failure questionnaire, Mini Mental State Exam, Trail Making Test, processing speed, working memory, and verbal memory.

Although postoperative cognitive dysfunction trials were collected and extracted as outlined above, the data were ultimately excluded from the meta-analysis given the heterogeneity of the outcome measures and instead were discussed descriptively. This was the only deviation from the predefined objectives.

Analyses used OR as the effect size measured with 95% CIs, and significance was set at P < 0.05, with the study as the unit of analysis. A total of five trials with postoperative delirium as the primary outcome were included in the meta-analysis to compute the overall pooled-effect estimate examining the relationship between processed electroencephalogram and postoperative delirium.

The overall effect size analyses were computed with the STATA 12.0 (StataCorp, USA)²¹  metan function using the random option to conduct a random-effects method analyses. Additionally, Cochran’s Q statistic, which helps detect potential systematic differences in effects sizes between studies, was included in the analyses and evaluated with a chi-square test and P value. Significance in the Q test would suggest that heterogeneity exists among the studies in the analytic sample. Furthermore, the I² was calculated to represent another index of heterogeneity as the percentage of total variation caused by between-study differences.

Examination of the pooled estimate and Q statistic suggested that there was no significant heterogeneity detected in the pooled estimate, χ²(4) = 3.7 (P = 0.46). This result is further supported based on the I² = 0% in the analysis, suggesting no heterogeneity in the results. Based on the findings, the studies’ results only differ by sampling error (chance), and thus a fixed-effects model may be applied to obtain the overall effect size.

The literature search identified 39 potentially relevant articles, including one abstract.²²  The abstract was excluded because of an inability to obtain a full dataset from the corresponding author. Twenty-five publications were excluded because of failure to meet inclusion criteria: eight were review articles,^23–31  seven were cohort studies rather than randomized controlled clinical trials,^32–37  six evaluated outcome other than postoperative delirium or postoperative cognitive dysfunction,^34,38–41  two were still ongoing at the time of publication,^32,42  one used a variable other than processed electroencephalogram-guided care,⁴³  and one used a dataset from a trial already included.³⁴ 

The remaining thirteen trials were selected based on inclusion criteria and are included in this review and meta-analysis, three of which reported postoperative delirium and postoperative cognitive dysfunction as outcomes, two reported postoperative delirium only, and eight reported postoperative cognitive dysfunction only (table 2).

Five trials were included in the postoperative delirium analysis, with a total of 2,654 subjects, without crossover between standard care versus BIS monitored groups. Study populations ranged from 32 to 1,277 subjects. The mean age of patients in each trial ranged from 60 to 82 yr. The proportions of women in the reported studies ranged from 37 to 73%. Reported comorbidities included body mass index, age, sex, education, ASA status, surgery type and duration, presence of depression, preoperative cognitive status, preoperative functional assessment, and preoperative medication use (opioids, benzodiazepines). Chan et al.,⁴⁴  Radtke et al.,⁴⁵  and Sieber et al.⁴⁶  included roughly equal proportions of ASA status patients (I–IV), Jildenstal et al.⁴⁷  included only ASA I and II, and Whitlock et al.⁴⁸  included a majority of ASA IV patients.

Four of the five trials randomized to processed electroencephalogram-guided anesthesia versus unmonitored care. The trials by Chan et al.⁴⁴  and Radtke et al.⁴⁵  used a BIS-guided group (with target 40 to 60 or 50 to 60) versus a control group of routine care without BIS monitoring: the rate of postoperative delirium was 15.6% in monitored care versus 24.1% in routine care (P = 0.01) in the study by Chan et al.⁴⁴  and 16.7% versus 21.4% (P = 0.036) in the study by Radtke et al.⁴⁵  Whitlock et al.⁴⁸  used a BIS-guided group (BIS 40 to 60) versus a control group of end-tidal anesthetic concentration with goal 0.7 to 1.3 age-adjusted minimum alveolar concentration and found the rate of postoperative delirium was 18.8% in the BIS group and 28.0% in routine care (P = 0.058). Jildenstal et al.⁴⁷  used auditory evoked potential–guided anesthesia with an interventional goal of auditory evoked potentials index of 15 to 20 versus a control group of unmonitored routine care and found the rate of postoperative delirium was 0% in the interventional group versus 12.5% in the routine care group (P = 0.48).

The fifth trial by Sieber et al.⁴⁹  randomized patients into receiving two different BIS target values: at least 80 (intervention) versus 50 (control); this was the only study in the postoperative delirium analysis that employed spinal anesthesia with propofol sedation rather than using general anesthesia, as well as the only study that designated a numerically lower target goal instead of routine care without monitoring. The authors state, “the sedation criterion in the deep sedation group may be more representative of actual practice than generally appreciated”; thus, in the meta-analysis, the low-BIS group data were included with routine care data from the other trials. In this study, Sieber et al.⁴⁹  found the postoperative delirium rate to be 19% in monitored care with higher targeted BIS values versus 40% in the group with the lower targeted values (P = 0.02).

Outcomes were measured by standardized delirium screen in all postoperative delirium trials, with three using the confusion assessment method,⁵⁰  one using the confusion assessment method for the intensive care unit,⁵¹  and one using psychiatric evaluation with Diagnostic and Statistical Manual IV criteria.⁵²  The aggregated OR computed between processed electroencephalogram monitoring and postoperative delirium for all five studies using the fixed-effects model was 0.62 (P < 0.001; 95% CI, 0.51 to 0.76; I² = 0%; fig. 2). The combined results suggest that processed electroencephalogram-guided anesthesia decreased the odds of developing postoperative delirium by approximately 38%. Because the study by Sieber et al.⁴⁹  had a slightly different study goal and anesthetic management, we performed additional analysis to determine whether the overall results were different when this study was excluded. In the repeated analysis without this study, we found an OR of 0.64 (95% CI, 0.53 to 0.79, P < 0.001), which is essentially unchanged from the original result including this study.

Eleven randomized controlled clinical trials examining postoperative cognitive dysfunction were identified. Trial sizes ranged from 32 to 1,277 subjects. The mean age of included patients in each trial ranged from 37 to 75 yr. All trials used general anesthesia.

Three trials^34,45,47  previously discussed are included in this analysis because they measured postoperative cognitive dysfunction in addition to postoperative delirium as an outcome. Wong et al.⁵³  used a BIS-guided group (with target 40 to 60 or 50 to 60) versus control of routine care without BIS monitoring. Another trial from Jildenstal et al.⁵⁴  used anesthesia guided by auditory evoked potentials with a goal auditory evoked potential index (15 to 20) versus routine care. An et al.⁵⁵  used a high BIS goal of 55 to 65 versus a low goal of 30 to 40 with total intravenous anesthesia. Farag et al.⁵⁶  used a high BIS goal of 50 to 60 versus a control low goal 30 to 40. Ballard et al.⁵⁷  used a combined intervention of BIS guidance (goal 40 to 60) with peripheral capillary oxygen saturation monitoring versus routine care. Hou et al.⁵⁸  used two different BIS goals (55 to 65 vs. 40 to 50). Two trials^59,60  used a three-way randomization model with intervention groups having BIS goals 50 to 60 versus 40 to 50 versus 30 to 40 (control); Shu et al.⁵⁹  used general anesthesia, whereas Zhang and Nie⁶⁰  used total intravenous anesthesia.

Postoperative cognitive dysfunction was evaluated with a wide range of neuropsychologic test batteries that varied greatly from trial to trial. Time collection of postoperative data points also varied extensively and ranged from 1 day to 1 yr after surgery. Given this heterogeneity of outcome measurement, the extracted data were not suitable for meta-analysis. Therefore, a discussion of the studies is summarized descriptively. Direct comparison of the incidence of postoperative cognitive dysfunction is difficult, but across studies, it ranged from 0.01% at 1 day to 56% at 1 yr in the monitored groups and from 0.07% at 1 day to 84% at 1 yr in the control groups.

As described under Materials and Methods, the methodologic quality of the included studies was assessed with respect to randomization, allocation concealment, and selection bias. The risk of bias was low for three studies, with unclear assessment for two studies. The full findings are summarized in table 1.

Of the five randomized controlled clinical trials examined in the postoperative delirium meta-analysis, three^44–46  found processed electroencephalogram-guided anesthesia to be associated with significantly decreased risk of postoperative delirium. Whitlock et al.⁴⁸  found a difference in the rates of postoperative delirium between the BIS-monitored and routine care, but the difference was not statistically significant, and Jildenstal et al.⁴⁷  found no difference.

Because the trials in the current literature search vary greatly in quality, sample size, and methodology of processed electroencephalogram monitoring, meta-analysis was essential for drawing conclusions that could inform clinical practice. The combined results suggest that use of a processed electroencephalogram may be associated with lower postoperative delirium incidence. However, whether there is a causal mechanism for this decrease is unknown, although hypothesized mechanisms are discussed below.

One of the most common explanations is that the use of processed electroencephalogram monitored care allows the anesthesiologists to reduce the amount of anesthetics administered, therefore resulting in a “lighter” anesthetic depth, as shown by the continuous processed electroencephalogram number such as the BIS. This explanation suggests that anesthetic agents by themselves may be deleterious to the brain, therefore reducing the amount administered may result in a lower incidence of postoperative delirium. This hypothesis, however, is unproven by existing studies. The study by Jildenstal et al.⁴⁷  showed that by targeting BIS values of 40 to 60, doses of hypnotic agents decrease by 11 to 27%.⁹  However, this result was contrary to that reported by Radtke et al.,⁴⁵  which showed the amount of anesthetics used between the groups with versus without the use of processed electroencephalogram was similar. In fact, in the study by Whitlock et al.,⁴⁸  the authors reported that the patients with postoperative delirium actually received lower levels of anesthetics. Results from the latter study suggest that factors other than the amount of anesthetics administered may be at play that are affecting the processed electroencephalogram levels, such as patients’ baseline vulnerability.

Prior studies on the use of processed electroencephalogram-guided anesthesia focused on factors such as the BIS levels or the amount of burst suppression being predictors of postoperative delirium.^45,61  However, the assumption that the amount of anesthetic given to older patients directly contributes to acute brain dysfunction and results in subsequent delirium is unproven. Furthermore, previous studies addressing anesthetic depth and cognitive outcomes did not consider preoperative cognitive status as a potential moderator for the effects of anesthetic depth on postoperative cognitive outcomes. Specifically, one of the most important baseline patient-related factors contributing to adverse postoperative cognitive outcomes is preexisting cognitive impairment. Therefore, the depth of anesthesia may simply be a marker for patient’s baseline brain vulnerability to the effects of anesthetics. The differentiation between direct effects of anesthetic effects on the brain versus patients’ baseline vulnerability is critical to understanding the relationship between delirium and the role of the use of processed electroencephalogram-guided anesthesia.

There were several potential limitations to our meta-analysis including the number of trials available, especially with respect to publication bias, because published peer-reviewed trials tend to exclude negative trials. Additionally, two studies received an “unclear” risk of bias score, secondary to incomplete information on the allocation process and handling of exclusions. Both of these studies still received a Jadad score more than 3, indicating adequacy for meta-analysis.

Another potential limitation is that four of the five studies randomized to processed electroencephalogram-guided care with a higher target versus routine “blinded” care, established on the assumption that unmonitored anesthesia has lower monitor readings, usually a BIS of less than 60.⁶²  However, Sieber et al.⁴⁹  had both groups assigned to processed electroencephalogram-guided care (high vs. low targets), with the assumption that the “sedation criterion [low BIS target of 50] may be more representative of actual practice.” Because this represents a variation, repeat meta-analysis excluding this study was performed and did not significantly change the results.

There was also variation in the scales for reporting delirium, although the three tools (confusion assessment method, confusion assessment method for the intensive care unit, and Diagnostic and Statistical Manual IV criteria) represent, respectively, highly validated tools and gold-standard evaluation. Additionally, the mean age of patients in the included trials ranged from 60 to 82 yr, which could have biased rates of delirium and limited the generalizability of results, especially considering there are limited data for delirium incidence in patients of more than 80 yr old. Last, Whitlock et al.⁴⁸  studied a thoracic and cardiac surgery population, which is known to have higher rates of postoperative delirium,⁶³  and the findings may not be directly applicable to the noncardiac surgical patient population.

Using the fixed-effects model, the assumption is that a common effect size is generalizable only to the population collectively defined by the analyzed studies of older surgical patients. Because the studies had analogous interventions and postoperative delirium was assessed using comparable validated scales, statistical criteria were met for the use of a fixed-effects model. However, we recognize that the fixed-effects model can result in narrower CIs around the effect sizes. All results were verified by a random-effects model and were similar.

Last, we reported only on the incidence of postoperative cognitive outcomes, rather than the practical sequelae of cognitive impairment such as length of hospital stay or cost-effectiveness.⁶⁴  Future research should evaluate the extent and amount of training surrounding the use of processed electroencephalogram and the costs of equipment and supplies vis-à-vis the reduction of adverse postoperative outcomes.

An initial aim when designing this meta-analysis was to evaluate the effect of processed electroencephalogram monitoring on postoperative cognitive dysfunction, because it is still unknown whether this represents a less severe trajectory of postoperative delirium or whether it is a discrete phenomenon with varying etiologies. A recent meta-analysis from Lu et al.⁶⁵  attempted to delineate this relationship with four trials^46,55,56  (including one trial that was excluded from our study because it had an independent variable of dexamethasone administration)⁶⁶  and found no significant difference.

Although we selected a moderate number of trials for analysis (n = 11), the selected studies ultimately measured postoperative cognitive dysfunction with heterogeneous neurocognitive batteries and different timelines, which made pooled-data analysis inappropriate. Additionally, the risk of bias for the postoperative cognitive dysfunction trials was heavily weighted toward high or unclear bias, which is a departure from the delirium trials (table 1). Without a standard definition of postoperative cognitive dysfunction, it is difficult to evaluate the existing publications with meta-analysis.

In two of the trials,^59,60  the high BIS level groups actually performed significantly worse on outcome measures. Similarly, An et al.⁵⁵  showed that the rate of postoperative cognitive dysfunction in patients randomized to a lower target was 10% versus 27.5% in the higher target group. Although not a randomized controlled clinical trial, a cohort study by Deiner et al.³²  also found that more time spent in at lower levels (BIS less than 45) and burst suppression was significantly associated with lower rates of postoperative cognitive dysfunction. These results are contradictory to those shown in the delirium studies and clearly need to be confirmed by large randomized trials, including the use of standardized neuropsychologic tests and follow-up of patients at regular intervals to determine the magnitude and duration of postoperative cognitive dysfunction and its relationship with postoperative delirium. Our review also confirms the need to develop standardized definitions of postoperative cognitive dysfunction.

Delirium is a geriatric syndrome with many contributing perioperative factors. Use of processed electroencephalogram may be associated with decreased postoperative delirium incidence. However, the mechanism for this association is unknown. Specifically, whether processed encephalographic values are truly modifiable variables in the strategy to prevent or reduce postoperative delirium remains to be tested. Equally unclear is whether the observed processed encephalographic values are simply surrogate markers for the at-risk patients. Finally, the heterogeneous methods in measuring postoperative cognitive dysfunction make it difficult to assess its relationship with processed electroencephalogram.

Supported in part by National Institute on Aging of the National Institutes of Health (Bethesda, Maryland) grant Nos. R21AG04845602 and R21AG05371501A1.

The authors declare no competing interests.

• A – True randomization (i.e., random number table, computer random number generator, etc.)

• B – Indicating “randomization was done” without providing the details as described in (A)

• C – No mention of randomization, allowing choice of cohort, or other nonrandom method (i.e., medical record number, birth date, etc.)

• A – Use of central allocation or sequentially numbered opaque, sealed envelopes

• B – No mention of allocation concealment approach nor report of an approach not clearly within the bounds of (A) (i.e., mentioning sealed envelopes, but not whether they were opaque or sequentially numbered)

• C – Any approach where the research team could possibly predict allocation (i.e., open lists such as a list of random numbers), assignment envelopes without appropriate safeguards (i.e., use of unsealed, transparent, or not sequentially numbered envelopes).

• A – No missing outcome data or loss to follow-up less than 10%, reasons for missing outcome data mentioned, missing data balanced between cohorts, intention-to-treat analysis

• B – Insufficient reporting of attrition and exclusions to permit adequate judgment

• C – Loss to follow-up more than 10%, reason for missing outcome data likely to be related to true outcome, disparity in missing data between cohort groups, “as-treated” analysis