Evidence of an association between brain cellular injury and cognitive decline after non-cardiac surgery

Abstract

Background. Postoperative cognitive dysfunction (POCD) is common after non-cardiac surgery, but the mechanism is unclear. We hypothesized that decrements in cognition 1 month after non-cardiac surgery would be associated with evidence of brain injury detected by elevation of plasma concentrations of S100β, neuron-specific enolase (NSE), and/or the brain-specific protein glial fibrillary acid protein (GFAP).

Methods. One hundred and forty-nine patients undergoing shoulder surgery underwent neuropsychological testing before and then 1 month after surgery. Plasma was collected before and after anaesthesia. We determined the relationship between plasma biomarker concentrations and individual neuropsychological test results and a composite cognitive functioning score (mean Z-score).

Results. POCD (≥−1.5 sd decrement in Z-score from baseline) was present in 10.1% of patients 1 month after surgery. There was a negative relationship between higher plasma GFAP concentrations and lower postoperative composite Z-scores {estimated slope=−0.14 [95% confidence interval (CI) −0.24 to −0.04], P=0.005} and change from baseline in postoperative scores on the Rey Complex Figure Test copy trial (P=0.021), delayed recall trial (P=0.010), and the Symbol Digit Modalities Test (P=0.004) after adjustment for age, sex, history of hypertension and diabetes. A similar relationship was not observed with S100β or NSE concentrations.

Conclusions. Decline in cognition 1 month after shoulder surgery is associated with brain cellular injury as demonstrated by elevated plasma GFAP concentrations.

Editor's key points.

• The aetiology of postoperative cognitive function (POCD) is uncertain.

• Biomarkers for brain injury include S100β, NSE, and GFAP.

• In patients undergoing shoulder surgery, the authors measured postoperative levels of S100β, NSE, and GFAP.

• Higher levels of GFAP were associated with POCD 1 month after surgery.

Postoperative cognitive dysfunction (POCD) is common after non-cardiac surgery, particularly in the elderly. It affects between 26 and 41% of patients >60 yr of age at hospital discharge and 10–13% of patients 3 months after surgery.^1,2 Because development of POCD is associated with loss of independence, impaired quality of life, and risk for mortality, this complication has substantial public health implications.^2,3 Although several risk factors for POCD have been identified, the mechanisms underlying this condition remain undefined.^1,2

Debate is ongoing as to whether POCD is a manifestation of perioperative brain injury, whether it results from reversible consequences of anaesthetic or other centrally acting drugs, or whether it is merely a marker of underlying cerebrovascular disease.⁴ Patients are exposed to many events during surgery that could potentially lead to brain injury, including hypotension and tissue injury, which have been experimentally associated with neuroinflammation and impaired cognition.⁵ Preclinical data also suggest that volatile anaesthetic agents possess direct neurotoxic effects.⁶

Studies that have sought a brain structural basis for POCD have been limited mostly to patients undergoing cardiac surgery. However, such studies have been inconclusive on whether POCD is associated with ischaemic injury as detected with sensitive magnetic resonance imaging (MRI), for example.^7,8 Other studies have reported an inconsistent relationship between POCD after cardiac surgery and plasma concentrations of brain injury biomarkers such as S100β and neuron-specific enolase (NSE), likely due to the low specificity of these proteins for the brain.⁹

Glial fibrillary acid protein (GFAP) is an astrocyte protein with high specificity for the brain. It has been shown to be released into the plasma of adults with traumatic brain injury or stroke.^10–12 Determining whether a relationship exists between this sensitive and specific plasma biomarker of brain injury and POCD could provide important mechanistic insight into POCD aetiology. Further, because the diagnosis of POCD is often delayed (days to weeks after surgery) and requires detailed neuropsychological testing, the discovery of a reliable plasma biomarker that could aid in the prediction of this condition could have broad utility. The identification of patients at the time of surgery who are susceptible to POCD after surgery could allow the development of strategies and implementation of supports to reduce the consequences of this condition.⁷ In this study, we hypothesized that decrements in cognition 1 month after non-cardiac surgery would be associated with evidence of structural brain injury, as detected by plasma concentrations of the brain injury biomarkers S100β, NSE, and/or GFAP.

Methods

All study procedures were approved by the Institutional Review Board of the Johns Hopkins Medical Institutions and were performed after receiving written informed consent from participants. The patients were enrolled in a prospective observational study of the effects of beach chair positioning on cerebral autoregulation during shoulder surgery, as previously described.¹³ In that study, postoperative cognitive function did not differ between patients positioned supine and those positioned in the beach chair position (in which the head was elevated 30°–60° above the horizontal).

The patients received routine institutional care that included combined regional and general anaesthesia. Details of patient care have been described previously.¹³

Neuropsychological evaluation

Neuropsychological tests were used to evaluate the patients in the clinic before surgery and during the patients' 1-month postoperative clinic visit.¹³ The test battery included the Rey Auditory Verbal Learning Test (RAVLT), a test of verbal learning and memory; Brief Visuospatial Memory Test–Revised, an assessment of non-verbal learning and memory; Controlled Oral Word Association Test, a test of verbal fluency; Rey Complex Figure Test, a test of planning, constructional skill, and visuospatial memory; Symbol Digit Modalities Test, a test of psychomotor speed and attentional control; and Trail Making Test, which assesses visual scanning, psychomotor speed, attention, and executive function. The tests were administered by experienced research personnel, trained and supervised by a senior neuropsychologist.

Eight cognitive scores (average Rey Auditory Verbal Learning Test I–V and VIII, Controlled Oral Word Association score, Symbol Digit correct score, Complex Figure copy and delayed recall score, Trail Making part A time and part B time) were combined into a composite cognitive score by averaging the Z-scores of each test from the patients' preoperative assessments. Thus, by definition, the mean preoperative score is 0 and the standard deviation is 1. In this calculation, timed test scores were inverted to be consistent with non-timed tests, such that higher values represent better performance for all tests.

Plasma biomarker measurement

Venous blood was collected into glass tubes containing ethylenediaminetetraacetic acid before anaesthesia induction and after surgery in the post-anaesthesia care unit. It was used to measure plasma concentrations of NSE (R&D Systems, Minneapolis, MN, USA), S100β (Sigma-Aldrich, St Louis, MO and Genway, San Diego, CA, USA), and GFAP (Covance, Vienna, VA and Dako, Carpenteria, CA, USA). The samples were processed within 2 h of collection by centrifugation at 1500 g for 8 min; plasma was separated into aliquots and stored in cryotubes at −70°C. Assays were performed in duplicate using an electrochemiluminescent sandwich immunoassay platform [MesoScale Discovery (MSD), Gaithersburg, MD, USA] and were analysed on a Sector Imager 2400 (MSD) according to the manufacturer's protocol.^14,15 The lower limits of quantification (LLOQ) for the assays were GFAP, 0.011 ng ml⁻¹; S100β, 0.6 ng ml⁻¹; and NSE, 0.65 ng ml⁻¹. The interassay variance at the LLOQ for all assays was <10%.

Statistical analysis

The sample size of this study was calculated for determination of a difference in a metric of cerebral autoregulation between patients undergoing shoulder surgery in the beach chair vs the lateral decubitus position.¹³ A total of 240 patients were available for analysis; of those, 91 withdrew from the study prior to postoperative cognitive testing. We compared demographic, medical history, and preoperative biomarker data between patients with (n=149) and without (n=91) postoperative cognitive data to assess for systematic differences related to missing outcome data. In addition, up to 22% of patients with postoperative cognitive data had missing biomarker data. Three analytic approaches were used to handle missing data. First, we used complete case analysis that included only patients who had postoperative cognitive data and biomarker data. Next, we imputed missing biomarker data for patients with non-missing outcome data (n=149) using a multiple imputation method with chained equations with or without inclusion of cognitive data in the imputation model.¹⁶ Finally, all missing data, including the cognitive outcomes, were imputed. Forty datasets were created, and estimated β coefficients and their standard errors were obtained by using multiple-imputation combining rules.¹⁷

Continuous data are presented as mean (sd) unless otherwise specified; categorical data are presented as frequencies and percentages. Paired t-tests were used to assess changes in neurocognitive measures. P-values were adjusted for multiplicity by using the false discovery rate (FDR) approach.¹⁸ To assess the relationship between postoperative plasma biomarkers and change in postoperative cognitive scores from baseline, we fit linear regression models with change in neurocognitive score as the dependent variable and postoperative plasma biomarker as the main independent variable. Other variables in the model were age, patient sex, preoperative composite score, diabetes, and hypertension. Analysis of residuals was used to evaluate the assumptions of the linear regression model and bootstrapped 95% CIs are reported to account for deviations from normality.

Results

The demographic and medical characteristics of the patients with postoperative cognitive testing are shown in Table 1. There were no differences in the listed variables between this group and those missing postoperative cognitive data, except that this group had a higher frequency of use of anti-lipidemic and beta-blocker medications. The neuropsychological test results measured before and 1 month after surgery are shown in Table 2. For the entire sample, mean test scores improved after surgery for all cognitive tests except the Symbol Digit test and the Complex Figure copy score. Higher test scores are likely due to a combination of true cognitive improvement and test practice effects. The degree of improvement was variable, however, and some patients declined on multiple test measures.

Table 1.

Medical characteristics for patients undergoing shoulder surgery. ACEi, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; COPD, chronic obstructive pulmonary disease; CVA, cerebrovascular accident; TIA, transient ischaemic attack

Characteristic	Number of patients (%) (n=149)
Mean age in years (sd)	57 (15)
Male sex	82 (55.0)
Race
Caucasian	131 (88.0)
African American	14 (9.4)
Asian	3 (2.0)
Hispanic	1 (0.7)
Prior CVA	1 (0.7)
Prior TIA	6 (4.0)
COPD	4 (2.7)
Asthma	18 (12.1)
Current smoker	17 (11.4)
Previous smoker	50 (33.6)
Coronary artery disease	6 (4.0)
Peripheral vascular disease	2 (1.3)
Hypertension	67 (45.0)
Diabetes	15 (10.1)
Congestive heart failure	3 (2.0)
Atrial fibrillation	2 (1.3)
Myocardial infarction	3 (2.0)
Cardiac valvular disease	8 (5.4)
Medications
Aspirin	50 (33.6)
HMG-CoA reductase inhibitors	1 (0.7)
Beta-blockers	27 (18.1)
Calcium channel blockers	17 (11.4)
ACEi/ARB	26 (17.3)
Diuretics	33 (22.1)

Table 2.

Neuropsychological testing results measured before and 1 month after surgery. RAVLT, Rey Auditory Verbal Learning Test. ^†Data are presented as mean (sd) or median (25th–75th percentile). *n=225 includes all of available baseline data. **n=149 is the subsample with non-missing postoperative data. ^‡Reported are P-values adjusted for multiplicity using the False Discovery Rate (FDR)

	Preoperative baseline cohort (n=225)*†	Preoperative baseline for patients with postoperative testing (n=149)**,†	1 month after surgery (n=149)†	P-value (paired t-test)‡
RAVLT Trials I–V (average score)	8.30 (2.10)	8.47 (2.19)	10.17 (2.61)	<0.0001
RAVLT Trial VIII Correct	8.04 (3.28)	8.26 (3.33)	10.10 (3.39)	<0.0001
Complex Figure copy	32.17 (6.10)	32.79 (5.58)	30.77 (10.05)	0.011
Complex Figure delayed recall	14.49 (6.46)	14.57 (6.20)	16.52 (7.79)	<0.0001
Symbol Digit number correct	44.09 (13.03)	44.66 (13.64)	42.78 (19.01)	0.163
Trail Making A time (sec)	30 (24–37.5)	30 (24–36)	27 (19.5–35)	<0.0001
Trail Making B time (sec)	67 (56–89)	66 (53–87)	58 (45–76)	0.002
Controlled Word Association, sum of three trials	38.23 (12.00)	39.79 (12.05)	42.03 (13.26)	0.004
Composite Z-score	0.00 (1.0)	0.1 (0.98)	0.44 (1.33)	<0.0001

Plasma GFAP, S100β, and NSE concentrations are shown in Table 3. There was no significant relationship between any of the plasma biomarkers and neuropsychological test results at baseline. The relationship between postoperative plasma biomarker concentrations and change from baseline in neuropsychological test results 1 month after surgery are shown in Table 4. After adjusting for patient age, sex, history of hypertension, and diabetes, there were significant relationships between higher plasma GFAP concentrations and change from baseline scores on Complex Figure copy (P=0.021) and delayed recall (P=0.010) tests and the Symbol Digit number correct test (P=0.004). There was no relationship between higher plasma S100β or NSE concentrations and change in any neuropsychological test results 1 month after surgery.

Table 3.

Plasma glial fibrillary acid protein (GFAP), S100β, and neuron-specific enolase (NSE) concentrations (ng ml⁻¹) measured before and immediately after surgery. The data are presented as median (25th–75th percentile)

Biomarker	Preoperative	Postoperative
GFAP	0.001 (0.001–0.042)	0.001 (0.001–0.04)
S100	0.001 (0.001–3.67)	0.16 (0.001–3.53)
NSE	38 (23.6–77.2)	43.35 (21.95–69.95)

Table 4.

Results of linear regression analysis between plasma glial fibrillary acid protein (GFAP), S100β, and neuron-specific enolase concentrations obtained immediately after surgery and change in neuropsychological test results 1 month after surgery from baseline. Results are adjusted for age, sex, history of hypertension, and diabetes. RAVLT, Rey Auditory Verbal Learning Test

Marker/test	β-coefficient	95% confidence interval	P-value
GFAP
RAVLT (I–V)	−0.04	−0.19, 0.11	0.594
RAVLT VIII	0.10	−0.12, 0.32	0.369
Controlled Word Association	−0.02	−0.86, 0.82	0.961
Complex Figure copy	−1.49	−2.76, −0.23	0.021
Complex Figure delayed recall	−0.94	−1.65, −0.22	0.010
Symbol Digit number correct	−2.98	−5.01, −0.95	0.004
Trail Making A time	−1.03	−2.40, 0.34	0.140
Trail Making B time	−4.03	−8.46, 0.39	0.074
Postoperative Z-score	−0.14	−0.24, −0.04	0.005
S100β
RAVLT (I–V)	0.04	−0.02, 0.11	0.204
RAVLT VIII	0.05	−0.05, 0.14	0.341
Controlled Word Association	−0.05	−0.43, 0.33	0.796
Complex Figure copy	0.18	−0.19, 0.55	0.341
Complex Figure delayed recall	−0.01	−0.29, 0.27	0.950
Symbol Digit number correct	0.30	−0.37, 0.98	0.380
Trail Making A time	−0.38	−0.89, 0.13	0.147
Trail Making B time	0.09	−1.11, 1.29	0.886
Postoperative Z-score	0.02	−0.01, 0.05	0.284
Neuron-specific enolase
RAVLT (I–V)	−0.09	−0.37, 0.19	0.538
RAVLT VIII	0.18	−0.17, 0.53	0.314
Controlled Word Association	−0.03	−1.17, 1.12	0.965
Complex Figure copy	−0.29	−1.61, 1.02	0.661
Complex Figure delayed recall	0.71	−0.32, 1.74	0.176
Symbol Digit number correct	−0.15	−1.67, 1.38	0.852
Trail Making A time	1.09	−0.60, 2.77	0.208
Trail Making B time	0.93	−2.91, 4.76	0.635
Postoperative Z-score	−0.01	−0.10, 0.09	0.857

We examined the relationship between plasma biomarker concentrations and change in the neuropsychological composite Z-score after surgery. The distribution of postoperative composite Z-scores is shown in Fig. 1. Fifteen (10.1%) of the 149 patients met the definition of POCD based on a postoperative composite Z-score <−1.5. There was a negative linear relationship between higher plasma GFAP concentrations measured immediately after surgery and change in postoperative composite Z-score 1 month after surgery [slope −0.14 (95% CI −0.24 to −0.04), P=0.005] after adjusting for age, patient sex, hypertension, and diabetes. The slope estimate was very similar across the three analytic methods that dealt differently with missing data. There was no significant relationship between S100β or NSE and change from baseline in the postoperative cognitive Z-score after adjusting for preoperative score, age, sex, and history of hypertension or diabetes (P=0.284 and 0.857, respectively).

Fig 1.

The percentage of patients vs the distribution in postoperative Z-score (n=149).

Discussion

In this study we found that cognitive decline 1 month after shoulder surgery, assessed as the change from baseline in individual results from each neuropsychological test and as a cognitive composite score (mean of individual test Z-scores) was associated with higher plasma concentrations of the brain-specific biomarker GFAP. Plasma GFAP concentrations obtained immediately after surgery were able to predict diminished cognitive performance subsequently assessed 1 month after surgery with high accuracy. In contrast, the traditional biomarkers S100β and NSE had no relationship with decrements in cognition.

The mechanism for POCD is unknown, and whether this condition represents a manifestation of brain injury or a manifestation of underlying cerebrovascular disease is debated. In a longitudinal study, Selnes and colleagues⁴ found that the rate of cognitive decline over a 6-year follow-up period did not differ between patients who had undergone coronary artery bypass graft surgery and medically treated controls with documented coronary disease. That is, long-term decrements in cognition might be explained by the natural progression of cerebrovascular disease and not necessarily by anaesthesia or cardiac surgery. Studies that have included sensitive brain diffusion-weighted MRI scanning after cardiac surgery have further reported conflicting and inconclusive results on whether POCD is associated with new ischaemic brain injury.^7,8

The use of group-averaged data to assess cognitive status, as done by Selnes and colleagues,⁴ may result in failure to detect decrements in cognition in selected individuals. Further, the absence of new ischaemic injury based on MRI after surgery does not exclude injury that is below the threshold of detection by these methods. Other studies that have attempted to find evidence of brain injury in patients with POCD by measuring plasma S100β or NSE concentrations have reported inconsistent results.⁹ These proteins have been shown to be elevated in patients after cardiac arrest, head trauma, and stroke, and in those with cognitive dysfunction after cardiac surgery.^19–22 However, the diagnostic sensitivity of S100β or NSE is tempered by their non-specificity for brain tissue, which causes a high prevalence of false-positive results.^9,22 Data on the relationship between plasma S100β or NSE and POCD after non-cardiac surgery are limited. A meta-analysis of seven studies in which patients underwent non-cardiac surgery revealed inconsistent and inconclusive results.²³ Most studies find no relationship between plasma NSE concentrations and POCD. We also found no relationship between elevated S100β or NSE and cognitive decrements; however, our analysis was limited to specimens obtained immediately after surgery.

GFAP is a cytoskeleton protein found in astrocytes, the most abundant cell type in the brain. These cells have multiple functions, including participation in cellular proliferation, synaptic plasticity, glutamate re-uptake, neuronal repair after ischaemic injury, and maintaining the functional and structural integrity of the blood–brain barrier.^9,24 Researchers have shown that plasma GFAP concentrations have high diagnostic and prognostic capability in adults with traumatic brain injury or stroke.^10–12,25 GFAP is a marker for astroglial cell injury or necrosis and becomes detectable in circulation within hours, peaking at up to 4 days after an ischaemic stroke in adults and children.^26–28

In this study, we show for the first time that plasma GFAP concentrations are associated with cognition decline 1 month after non-cardiac surgery. The underlying mechanism for this relationship is unclear. One explanation is that plasma GFAP elevations reflect acute neuroinflammation, as demonstrated to occur experimentally. For example, rodents exposed to general anaesthesia and surgical trauma exhibit hippocampal inflammation that most likely results in pro-inflammatory cytokine-mediated microgliosis and astrogliosis.⁵ Increased translation of GFAP in astrocytes was documented for up to 3 days after surgery, with concentrations returning to baseline at 7 days. These changes were associated with cognitive abnormalities in the animals after surgery. Further, in the experimental model, neuroinflammation was associated with increased transcription of neurotoxic proteins, including β-amyloid precursor protein, β-amyloid, and τ protein hyperphosphorylation, all of which are associated with Alzheimer's disease.²⁹ However, the rapid elevation in plasma GFAP that occurred immediately after surgery in our study more likely resulted from astrocyte cytoskeletal breakdown rather than increased protein translation.

We found that mean neuropsychological test results increased between baseline and the 1-month postoperative testing session (Table 2). This finding likely represents the well-appreciated practice effect (i.e., increased familiarity with the tests), as well as the effect of other variables such as improved patient health after surgery (including reduced shoulder pain). However, patients with high postoperative neuropsychological testing performance may mask the test scores of patients with poor performance.

Our study has several limitations, including missing cognitive data from some patients. This resulted from refusal by some patients to continue participation in the study when approached for follow-up testing during their postoperative surgical clinic visit. Further, plasma biomarker concentrations were not measured in all patients enrolled in the primary study due to logistic issues. Our multiple statistical approaches, which included limiting analysis to only those patients with complete data and analysing the entire data set using imputation of missing data, support the validity of our main findings. We only report plasma biomarker concentrations obtained immediately after surgery and not later in the postoperative period because many patients were discharged from the hospital immediately after surgery. However, prior research has shown that plasma GFAP concentrations are elevated within 2–6 h after ischaemic stroke, a time window that may make it suitable for early detection of brain injury after surgery.³⁰ Other limitations of this study include the small sample size that was dictated in part by the number of enrolled patients in the parent study. Further, there was no non-operated control group that underwent cognitive testing at a comparable interval to estimate the magnitude of practice effects on the neurocognitive data. This is less critical in the present study, as our primary goal was to account for within-group variability in postoperative cognitive function. Finally, subjecting the data to multiple comparisons could introduce the possibility of type I error by chance. Nevertheless, the finding of a relationship between lower postoperative cognitive performance and higher plasma GFAP concentrations at a high level of statistical significance for some of the analysis minimizes this possibility of false-positive findings.

In conclusion, these data show that cognition decline detected 1 month after shoulder surgery is associated with elevations in plasma GFAP concentrations measured immediately after surgery. These results suggest that POCD may be related in part to surgery-induced injury to brain cells. Diagnosing this condition early by measuring plasma GFAP concentrations may portend the development of strategies to limit the impact of POCD on patient quality of life and other adverse outcomes associated with the condition.

Author's contributions

Data analysis: T.R., D.H., C.B., J.B., C.D.M., F.S., A.G., G.Y., A.E., C.W.H.

Writing of the manuscript: T.R., A.L., D.H., C.B., J.B., C.D.M., F.S., A.G., G.Y., A.E., C.W.H.

Patient recruitment: A.L.

Data acquisition: A.L., A.E., C.W.H.

Study design: A.E., C.W.H.

Declaration of interest

A.E. receives consulting fees from Immunarray (Richmond, VA, USA). C.W.H. receives research funding from Covidien (Boulder, CO, USA), serves on the advisory board for Ornim (Foxboro, MA, USA) and Grifols (Los Angeles, CA, USA), and has served on the Data and Safety Monitoring Committees for CS Behring (King of Prussia, PA, USA), Merck (Kenilworth, NJ, USA), and The Medicines Company (Parsippany, NJ, USA).

Funding

This study was funded in part by grants to C.W.H. from the Anesthesia Patient Safety Foundation and the National Institutes of Health (R01 092259). D.H. has received funding from the Japan Heart Foundation/Bayer Yakuhin Research Grant Abroad.