Can intraoperative processed EEG predict postoperative cognitive dysfunction in the elderly?

Abstract

Purpose

Postoperative cognitive dysfunction (POCD) is a serious and costly complication of the elderly; even mild impairment has the potential to impact overall well-being. Anesthesiologists continue to search for ways to manipulate intraoperative technique to optimize postoperative cognition in the elderly. Depth of anesthesia during surgery is one area which has shown some promise for short term outcomes such as delirium. However excessive depth has both positive and negative associations with longer term postoperative cognitive dysfunction. We hypothesize that this uncertainty is due to the inability of median depth to capture the amount of burst suppression or electrical silence. In this study our purpose is to identify the intraoperative processed EEG parameters which are most closely correlated with postoperative cognitive function.

Methods

To explore the association of several processed EEG parameters with POCD we performed a retrospective analysis of a cohort of 105 patients over age 68 scheduled for major surgery under general anesthesia. Intraoperative medications, hemodynamics, processed electroencephalography (EEG) and cerebral oxygen saturation were recorded. All patients had a cognitive battery before surgery and repeated at 3 months afterwards.

Findings

105 patients enrolled, 77 (73.3%) patients completed the 3-month cognitive testing. The incidence of POCD was 27%; the median BIS value for patients who developed POCD was similar to patients who did not (46 vs. 42 minutes). However patients who developed POCD spent less time with BIS < 45 (82.8 vs. 122.5 minutes, p=.01) and burst suppression (35 vs. 96 minutes, p=.04). Hypotension, cerebral desaturation, and use of total intravenous anesthesia were similar between patients with and without POCD.

Implications

Patients who developed POCD spent less time in EEG burst suppression and less time in deep states. Burst suppression may be protective for postoperative cognitive dysfunction. Further work is needed to definitively identify the role of burst suppression in context of other patient and intraoperative variables to prevent POCD.

Introduction

Postoperative cognitive dysfunction (POCD) is a serious and costly complication in the elderly. It is often reported that elderly patients were “never the same” after surgery and this complaint has been validated with neuropsychological tests. According to one study, more than 40% of surgical patients over 60 years of age had cognitive decline at hospital discharge and 12.7% were impaired at 3 months ¹. Mild impairment in executive function or memory after surgery has the potential to impact overall well-being, even if transient. More than 17 million patients over the age of 65 undergo inpatient surgery and every year millions of elderly surgical patients are discharged from the hospital with a decrease in their cognitive abilities ².

Many intraoperative factors have been hypothesized to play a role in the occurrence of POCD in older adults. In our previous report we investigated whether suppression of the stress response with total intravenous anesthesia (TIVA) would result in a decrease the incidence of postoperative delirium (PD). Although TIVA did suppress the intraoperative stress response we found that the incidence of PD was unchanged ³ . Because our previous work failed to find an association between anesthetic or stress hormone response and delirium we chose to examine anesthetic depth which has shown some relation to postoperative cognition, and strongly with postoperative delirium.

However, much uncertainty has surrounded the use of processed EEG to optimize postoperative cognition. In this regard it is unclear whether median depth, time spent under a threshold depth, or other parameters are most important. In this study we have used a cohort from a previous study to examine the relationship between median depth, cumulative deep state, burst suppression and incidence of postoperative cognitive dysfunction. Our hypothesis is that median depth of anesthesia alone cannot capture the time dependent effect of depth on postoperative cognition. We also present data regarding anesthetic use, cerebral saturation, hemodynamics, and POCD. While exploratory in nature, this information could provide some insight into the pathophysiology of POCD, and suggest potential avenues for intraoperative intervention.

Methods

The protocol was previously described in our study examining postoperative delirium in the same cohort. Briefly, this IRB approved project recruited patients over age 68 who provided informed consent and were scheduled for major non-cardiac surgery under general anesthesia including general, spine, urologic, or thoracic surgery at the Mount Sinai Hospital NY, NY. Major surgery was defined as planned hospital stay of at least 2 days. Exclusion criteria included intracranial surgery, cardiac surgery, preexisting neuropsychiatric disease, history of CVA with residual deficits, baseline Mini-Mental Status Exam (MMSE) < 20 or unable to consent for study participation, and/or unable to speak English. Eligible patients were identified using the scheduling system and approached at least 24 hours prior to surgery. The medical history was obtained from the patient and confirmed with the anesthesiologist.

Cognitive Assessment

Patients were assessed within 30 days prior to surgery and at 3 months after surgery with a neuropsychological battery from the Uniform Data Set (UDS) of the Alzheimer’s Disease Centers plus the Minimental Status Exam (MMSE), and Confusion Assessment Method (CAM) ^4-6. The UDS battery, sensitive to deficits in for both mild cognitive impairment and dementia, tests 5 domains: attention, processing speed, episodic memory, language, and executive function. Using the analysis of Price as a model, we performed a principal component analysis on the baseline tests which identified 2 groups of tests: executive function/attention/processing speed (Trails A & B, Digit Span Forwards, Digit Span Backwards) and memory/language (Logical Memory Immediate and Delayed, Category Fluency)⁷. A raw score and a raw change score (baseline-3months) was calculated for each subject and each test. Each change score was normalized (raw change score divided by standard deviation of the baseline score) and averaged within each group to create the domain composite score. POCD was defined as a change of greater than .75 SD in either domain; .75 was chosen because used a change score normalize to the cohort baseline (instead of healthy controls) and relatively small changes in a population with baseline impairment may have clinical significance.

Intraoperative data collected included vital signs, processed EEG (Bispectral Index, BIS Covidien), estimated blood loss (EBL), anesthetic, surgery type, and the surgery duration (Table 2). On the day of surgery, patients had blood drawn for serum epinephrine, norepinephrine immediately prior to surgery, 4 hours after surgery started, and at 2 hours after surgery. Patients received either TIVA consisting of propofol infusion or gas (sevoflurane) determined by the clinical anesthesia team based on requirements for the procedure. The anesthesiologist was instructed to avoid benzodiazepines and nitrous oxide but permitted to choose any induction agent, and narcotic, or paralytic. Patients were seen by the research team and assessed with the CAM and MMSE in the recovery room and on each postoperative day.

Table 2.

Comparison of intraoperative variables between patients who did and did not develop POCD at 3 months

Variable		POCD+(N=21)	POCD− (N=56)	p-val
TIVA		12 (27.9%)	31(72.1%)	.32
Sevoflurane		9 (26.4%)	25(73.5%)
Median BIS value		46(16.5)	42(16.3)	0.26
BIS	Time<45	82.8(121.5)	122.5(122.5)	0.01
AVGSR	Time with burst supression	35(88)	96(131)	0.04
Hypotension	# of 5 min epochs MAP<55	5(10)	5(22.5)	0.78
Cerebral Oximetry	Time Under 65%	0(3.3)	.7(21.1)	0.79
	Time Under 60%	0	0(.2)	1.00
	Time Under 55%	0	0(0)	1.00
Surgical Time		165(83)	184.5(133)	0.31
Delirium		3 (14.3%)	8(14.3%)	1.00

TIVA= total intravenous anesthesia, Delirium= postoperative delirium on postoperative days 1-3, Each entry is median(interquartile range) or frequency (percentage);

p-val for continuous variables are based on two-sample two-sided median tests

Processed electroencephalography (EEG) was monitored using the Bispectral Index (BIS) monitor (BIS Complete 4 Channel Monitor System, Covidien, Mansfield, MA). Raw and processed EEG was recorded including time in minutes under a BIS value <45 and time in minutes during which the EEG activity indicated burst suppression. Burst suppression is defined as the period in which the average burst suppression ratio >0 (minutes). To calculate the total time per case that BIS <45, we utilized the Average BIS Value (AVGBIS) which is the average of the BIS over the past minute, and summed the number of minutes from start to finish of the case that fell below a value of 45. In a similar fashion, we used the average burst suppression ratio over the past minute and summed the total number of minutes that this value fell below 0. Cerebral oxygen saturation was measured using the FORESIGHT system (CAS Medical Systems, Inc., Branford, CT). We reported the average reading over one minute from both sides of the brain. The oximetry and BIS monitors were turned away from the anesthesiologist, and they were not instructed to attempt to view or react to them.

Statistical analysis

The Chi Squared test or Fisher’s Exact test was used for categorical variables, and Students t-test or Wilcoxon rank sum test for continuous variables to compare patients with and without POCD. As this was a post-hoc investigation, we were powered to show a difference between TIVA and inhalational anesthetic groups with respect to catecholamine levels, but not a difference between POCD between groups. For the sample size calculation we used the difference in serum norepinephrine levels between groups from the work of Marana et al. [17] , which showed a 300 pg/ml difference between patients who received gas and those who received TIVA. With a sample size of 75 using a 2-sided 0.05 t test, we would have a greater than 80% power to detect a difference of this magnitude.

Results

105 patients consented to participate in the study, 77 (73.3%) patients completed the 3-month cognitive testing. The overall incidence of POCD at 3 months after surgery was 27.3%, most characterized by executive dysfunction (16 patients, 20.8%) and fewer with memory dysfunction (5 patients, 6.5%). No patient had both memory and executive dysfunction.

As planned we examined baseline and intraoperative characteristics of patients with and without POCD (Tables 1 and 2). The mean age of patients who developed POCD was 73.7 (3.5) vs. 74 (5.4) years of age (p=.82) for those who did not develop POCD. Patients who developed POCD were similar to those who did not with respect to education, sex, ASA status, surgical time, presence of diabetes, incidence of postoperative delirium, and prevalence of the ApoE4 genotype. However, patients who went on to develop POCD had significantly lower baseline MMSE scores (27.5 (2) vs. 29 (3), p=.02) than those who did not.

Table 1.

Comparison of baseline characteristics of patients who did and did not develop POCD

Variable		POCD+(N=21)	POCD−(N=56)	P-val
Age		73.7(3.5)	74 .0(5.4)	0.82
Sex	Female	11(52.4%)	26(46.4%)	0.80
ASA	1,2	7(33.3%)	16(28.6%)	0.24
	3	14(66.7%)	33(58.9%)
	4	0(0%)	7(12.5%)
MMSE		27.5(2)	29(3)	0.02
Diabetes		4(19.1%)	14(25%)	0.76
ApoE4		4(20%)	10(17.9%)	1.00
Education		16(6)	16(5)	0.87

Each entry is median(interquartile range) or frequncy (percentage), except for age, which is mean (sd); p-val for MMSE and Education are based on two-sample two-sided median tests, except for age, which is based on two-sample two-sided t-test.

The median BIS value for patients who developed POCD was similar (46 (16.5) vs. 42 (16.3)). However, the amount of time spent in deep anesthetic planes indicated by BIS < 45 was lower in the patients who developed POCD: 82.8 (121.5)vs. 122.5 (122.5) minutes for patients who did not (p=.01). The duration of burst suppression was also less in patients who went on to develop POCD (35 (88) v. 96 (131) minutes p=.04). The duration of intraoperative hypotension was similar between patients with and without POCD (5 -5 minute epochs or 25 minutes with MAP less than or equal to 55 mmHg). Intra- and postoperative serum epinephrine and norepinephrine levels were not significantly different between patients who did and did not develop POCD. Cerebral desaturation at 65%, 60% and 55% thresholds was overall uncommon, and not significantly different between patients who did and did not develop POCD. The incidence of postoperative delirium was the same in patients who did and did not develop POCD.

Discussion

This study found a 27% incidence of POCD at 3 months, higher than Monk et al (12.7%) and closer to Newman et al’s 24%. ^{1, 8}. In our cohort, patients who developed POCD spent less time in burst suppression and in deep states. Currently there is very little information regarding various measures of depth of sedation and POCD. A small observational cohort study which used processed EEG and short term follow up (1 week) found similar median depth between patients with and without POCD, and no association with median deep anesthetic time¹³. Similarly we found that median depth is not associated with POCD, instead, the parameter of burst suppression is protective for POCD. Our finding raises two important questions: which is the most important measure to predict postoperative cognition, and whether deep anesthesia is harmful or protective. This is a topic with much uncertainty: data from an elderly hip fracture population suggests that deep planes of sedation are a risk factor for delirium in after regional and sedation anesthesia; although deep anesthesia initially was found to be associated with excess mortality, a recent large trial did not replicate these findings ^{14, 15}.

Currently there are very few geriatric specific intraoperative anesthetic guidelines, and the best anesthetic depth for the elderly has not yet been defined. Farag et al found that greater anesthetic depth was protective for POCD, which our study confirmed ¹⁶. In contrast, a retrospective secondary analysis of a larger trial, found that greater depth was associated with POCD ¹⁷. There are several factors which may be explain these conflicting results including use of median depth not adjusted for length of surgery, and composite scores for POCD classification which may “wash out” domain specific cognitive changes⁷.

Our choice of .75 SD to define decline was relatively low and consistent with new awareness of sensitivity to early and mild cognitive decline which suggests that even subtle change is important ⁹. Alzheimer’s disease studies suggest that patients with <1 standard deviation decline have similar risk of progression to dementia as patients with greater impairment ¹⁰. In specific, there is emerging evidence to suggest that patients with mild cognitive impairment characterized by executive dysfunction are a unique subtype which carries its own prognosis ^{11, 12}. Of note the POCD group had multiple comorbidity and lower MMSE scores suggesting some previously existing vulnerability to cognitive impairment.

Limitations

Our interpretation of these results is limited by the size of the study and the cohort design. Given the small number of patients who developed POCD our ability to present models adjusted for multiple variables is limited. Future studies may gain greater sensitivity to potential cognitive defect by measuring domain specific deficit rather than global cognitive function. Since the comparison of patients with and without POCD was post-hoc, it is possible that a true difference exists between patients who receive TIVA vs. gas. However, the difference we saw in this dataset (22% vs. 17.7%) is very small, of doubtful clinical significance, and would have required a study with more than 3000 patients to be adequately powered.

We investigated whether the lower baseline MMSE was a confounder of the relationship between POCD and BIS. There is a positive correlation between BIS and AVGSR and baseline MMSE score. This suggests that future studies should investigate the relationship of baseline cognition to BIS parameters, specifically time spent with BIS <40 and AVGSR. We did not create a multivariable model to control for the effect of baseline MMSE on POCD since the study is small, and the # of people who had POCD even smaller. However, we believe our results are valid since POCD is defined as a change score, which takes into account the difference in cognitive baseline between patients who did and did not have POCD.

Future study

Our study supports the potential value of identifying best practice for intraoperative monitoring of processed EEG signals. These results suggest that burst suppression may be an important parameter, and our finding make explain why studies which examined only median BIS values or time spent in deep states may miss critical information.^{17, 18}. Future study is needed to examine POCD by domain since memory and executive function involve different areas in the brain. Furthermore, smaller declines in cognition may be clinically significant and future studies should consider long term prognosis including functional measures for patients with mild cognitive impairment after surgery.

Conclusion

POCD characterized by executive dysfunction is common at 3 months after surgery in elderly patients having general anesthesia. Choice of anesthetic agent does not appear to influence the development of POCD. Time spent in deep planes of anesthesia appears to be protective, and this is not reflected in median BIS values for a surgery but rather in parameters such as burst suppression. Future studies should explore POCD by domain and long term follow up to clearly identify modifiable risk factors and prognosis of short term and mild POCD.