Resolution of elevated interleukin-6 after surgery is associated with return of normal cognitive function

Abstract

Background

Unresolved surgical inflammation might induce chronic cognitive decline in older adults. Although inflammatory biomarkers have been correlated with perioperative cognitive impairment and delirium, the effects of prolonged inflammation on cognition are not well studied. This prospective cohort study investigated 1-yr dynamics in plasma interleukin-6 levels and executive function.

Methods

Patients undergoing major surgery (n=170) aged ≥65 yr completed Trail Making Test B and other neuropsychological assessments with plasma interleukin-6 levels collected on postoperative days 1–9 and 90, and at 1-yr. Mixed-effects analyses were conducted for Trail Making Test B (and other assessments), including interleukin-6 levels, time, and additional confounders (fixed effects), and a random effect for participant.

Results

Changes in interleukin-6 levels were associated with changes in Trail Making Test B over 1 yr in a generalised additive model (β=0.074, P<0.001) supporting that unresolved inflammation impaired executive function. This result was robust to confounders, outlier rejection, and fitting to non-linear models. Changes in interleukin-6 levels also correlated with changes in Trail Making Test A and Controlled Oral Word Association Test. Sensitivity analyses conducted on binary definitions of cognitive decline (>1, >1.5, or >2 standard deviations from baseline) were also associated with interleukin-6 changes.

Conclusions

Delayed resolution of inflammation is associated with cognitive impairment after surgery. Monitoring interleukin-6 might provide an opportunity to intervene with anti-inflammatory therapies in vulnerable patients.

Clinical trial registration

NCT01980511, NCT03124303.

Editor's key points.

• •Surgical trauma-induced neuroinflammation might lead to postoperative cognitive decline in older adults.
• •Data from a prospective cohort study including participants undergoing major noncardiac, non-neurological surgery were analysed for changes in the inflammatory biomarker interleukin-6 and cognitive function tests.
• •Changes in plasma interleukin-6 concentration were associated with changes in the Trail Making Test B over 1 yr after surgery.
• •These findings of an association between delayed resolution of inflammation with cognitive impairment after surgery suggest that monitoring interleukin-6 as a postoperative biomarker of inflammation could provide an opportunity to intervene with anti-inflammatory therapies in vulnerable patients.

Major surgery is associated with significant longitudinal changes in cognitive trajectory and the potential onset of dementia.^1,2 Up to 65% of people express fears of permanent postoperative cognitive deficits.³ It has been proposed that the delayed resolution of systemic inflammation after surgery leads to postoperative neurocognitive disorder especially in older surgical patients.4, 5, 6 Although short-term perioperative changes in inflammation and potentially related changes in β-amyloid, tau, and biomarkers of neurodegeneration have been correlated with acute cognitive impairment and delirium7, 8, 9, 10, 11, 12 (but also see¹³), the effects of prolonged postoperative inflammation on cognition, especially after elective surgery, are not well characterised.

Surgical trauma provokes neuroinflammation by increasing blood–brain barrier permeability to immunocytes and pro-inflammatory cytokines including interleukin (IL)-6.^7,9 Mechanistic studies suggest that IL-6 modulates neural plasticity and synaptic activity in the hippocampus and frontal cortex,¹⁴ which are brain regions associated with executive function. In mouse models, IL-6 was both necessary and sufficient to induce postoperative cognitive dysfunction: tocilizumab, a neutralising antibody to the IL-6 receptor, prevented postoperative neurologic dysfunction, while exogenous administration of IL-6 alone reproduced the post-surgical phenotype of cognitive decline.^15,16 Based on these data, we hypothesised that 1-yr cognitive recovery of executive function after surgery requires resolution of IL-6-mediated inflammation.14, 15, 16 However, the association between IL-6 and postoperative cognitive change has only been established in short-term studies limited to less than 1 month of follow-up.^7,8 Moreover, longer-term studies of chronic inflammation and executive function in non-surgical community-dwelling cohorts have yielded inconsistent conclusions.^17,18

We focused on executive function as it might be preferentially impaired by systemic inflammation over other cognitive domains. For example, increased IL-6 is frequently implicated in executive dysfunction, but not memory or attention deficits studied in the same cohorts.^19,20 Although global cognition has also been associated with peripheral inflammation,²¹ it might be less specific, and screening tests such as the Mini-Mental State Examination have a known ceiling effect and limited sensitivity for detecting subtle changes on follow-up.²⁰ Moreover, particular deficits in executive function were identified in patients after delirium related to critical illness.²² Identification of a biomarker associated with executive dysfunction could facilitate immunomodulatory therapies to prevent further cognitive decline and expedite rehabilitation after surgery.²³

Hence, we explored whether delayed resolution of increased plasma IL-6 up to 1 yr after surgery was associated with changes in executive function with a sensitive bedside test: Trail Making Test B (TMTB). Secondary analyses were conducted for other brief neuropsychological assessments, such as the Trial Making Test A (TMTA), Controlled Oral Word Association Test (COWAT), and Mini-Cog assessment. Sensitivity analyses investigated whether binary definitions of significant cognitive decline (>1, >1.5, and >2 standard deviations (sd) from baseline) were associated with increased IL-6 levels.

Methods

Study design

We analysed the ongoing prospective cohort study IPOD-B3 from the University of Wisconsin–Madison (Madison, WI, USA). The University of Wisconsin–Madison Institutional Review Board (2015-0374) provided ethics approval, and the trial was registered on ClinicalTrials.gov (NCT01980511, NCT03124303). Data are reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Deidentified data that support the findings of this study are available upon reasonable request.

Participants

Participants aged ≥65 yr (n=170) were recruited from clinical lists at University of Wisconsin–Madison Health scheduled to undergo major non-intracranial surgery requiring general or neuraxial anaesthesia and anticipated to stay ≥2 days in hospital. In total, 3946 patients were screened as described.⁹ Participants were excluded if they had a documented history of dementia, resided in a nursing home, or were unable to complete neurocognitive testing because of language, vision, or hearing impairments.

Three patients without postoperative plasma IL-6 data were excluded (Fig. 1a). Up to 59 further participants were excluded from mixed-effects analyses as they lacked complete demographic information, or baseline, postoperative cognitive test data, or both. There were 115 participants included in the analysis of the primary outcome: the relationship between postoperative change in IL-6 and TMTB.

Fig 1.

STROBE diagram and methods schematic. (a) STROBE, Strengthening the Reporting of Observational Studies in Epidemiology; pre-post, preoperative, and postoperative. (b) Schematic for the collection of cognitive test and IL-6 plasma concentration data, and calculation of change from baseline for each postoperative time point. COWAT, Controlled Oral Word Association Test; IL-6, interleukin-6; TMTA, Trail Making Test Part A; TMTB, Trail Making Test Part B.

Study procedure

At baseline, participants were assessed with a brief neuropsychological battery: TMTB, TMTA, COWAT (letters F, A, S), and Mini-Cog. The scores for the TMTB and TMTA were the time taken to complete the assessment, with a lower score indicating better performance; the COWAT was assessed by the total number of words produced by the patient in 3 min; and the Mini-Cog assessment (score 0–5) was derived from the sum of a three-word recall exercise (score 0–3) and a clock-drawing test (score 0–2).

Demographic and clinical information was obtained from patients at baseline or perioperatively, respectively, including the National Surgical Quality Improvement Project surgical risk of death score (NSQIP-D). Participants were asked if they ever smoked, experienced a stroke or transient ischaemic attack (TIA), or received a prior diagnosis of hypertension. Educational attainment was also binarised into ≤12 yr or >12 yr for each participant. Operating time (min), blood loss from surgery (ml), and surgery type (cardiovascular or other: ENT, general, orthopaedic, thoracic, or urological) were recorded.

Plasma samples were collected in EDTA-containing tubes and stored at –80°C for subsequent multiplex assay including IL-6 (Eve Technologies, Montreal, QC, Canada). The IL-6 measurements were performed in duplicate. Cognitive tests and plasma collection for IL-6 were conducted at baseline, postoperative days (PODs) 1–9, day 90, and 1 yr after surgery (Fig. 1b).

Data processing and analysis

Plasma IL-6 concentrations (pg ml⁻¹) and neuropsychological test scores were transformed by log₁₀ and normalised to baseline preoperative values by subtracting baseline values for each postoperative time point (Fig. 1b). Logarithmic transformation was favoured as we were interested in relative change from baseline. The Mini-Cog score range was shifted to 1–6 from 0–5 by the addition of one point before logarithmic transformation.

Mixed-effects analyses were conducted for each test with generalised additive mixed models used if non-linear effects were required to fit the data; otherwise, linear random effects were used. Generalised additive models were used to model the trajectories of change in neuropsychological test scores across PODs 1–9, day 90, and 1 yr, treating POD as a continuous variable. Penalised splines were applied for POD by gradually reducing the basis dimension (k) to a minimum such that the model provided the most appropriate fit as per the Akaike information criterion. In all analyses, this was achieved with an upper limit of k–1=2 degrees of freedom. These models were adjusted for potential confounders: age, sex, education, hypertension, stroke or TIA, and ever having smoked. In the unadjusted models, IL-6 was treated as a linear fixed effect, individual participant as a random effect, and a smoothing function for POD only. For the adjusted models, confounders were added as non-smoothed fixed effects. Approximate P-values for non-linear fitting of POD were calculated using the Wald statistic; P-values for linear fixed effects were calculated using Satterthwaite degrees of freedom, with the threshold of statistical significance set at α=0.05.

Sensitivity analyses

Sensitivity analyses were conducted by considering interaction terms between possible confounders (e.g. IL-6×age), outlier rejection, linear fitting for POD, computing the difference between TMTB and TMTA scores, and investigating significant cognitive decline using a 1-sd, 1.5-sd, or 2-sd threshold. The threshold of 1.5-sd was selected as our principal sensitivity analysis to distinguish poor cognitive performance in individuals dissimilar to age-matched peers.²⁴ The thresholds of 1-sd and 2-sd declines in neuropsychological scores were based on the International Perioperative Cognition Nomenclature Working Group recommendations.²⁵ Outliers were validated by Cook's distance, applying a conservative threshold (4×mean). The difference between TMTB and TMTA scores (TMTB–A) was analysed against change in IL-6 concentration over time to isolate the set-switching component of Part B, as it is proposed to be a more sensitive index of executive function than TMTB alone.²⁶

Participants were binarised as cognitive decliners or non-decliners for each neuropsychological test per postoperative time point, based on thresholds of 1-sd, 1.5-sd, or 2-sd declines from mean baseline scores. Mixed-effects analyses were performed using change in IL-6 level as the dependent variable, with binarised cognitive decliner status, POD, and confounders as fixed effects, and random effects for individual participant. We used receiver operating characteristic curve analyses to determine the optimal cut-point value for the relative change in IL-6 that predicted 1.5-sd decline in TMTB, by maximising the Youden's J statistic.

Power analysis

An effect size of β=0.2 was selected a priori as a standard ‘small’ coefficient of effect between plasma IL-6 level and TMTB, the primary cognitive test of interest in this study. This would mean that a 10-fold increase in plasma IL-6 would result in a 58% increase in the time taken to complete the TMTB. Simulations for linear mixed-effects models between TMTB and IL-6 level estimated that 185 observations would provide a power of 90.1–98.9% (95% confidence interval [CI], α=0.05) to detect this effect size. Our data included 191 observations, so the study was adequately powered.

Results

Because of the non-normality of raw baseline plasma IL-6 concentrations (P=0.001, Shapiro–Wilk), participant characteristics were reported in quartiles (Table 1). The mean raw plasma IL-6 concentration across all measurements was 33 pg ml⁻¹ with a coefficient of variation of 1.6 pg ml⁻¹. Raw plasma IL-6 concentration and neuropsychological test scores (Supplementary Table S1), and the number of participants for each measurement per time point Supplementary Table S2) were reported. The median age was 71 (8) yr and 39% of participants were female. Operating time (P=0.021, Kruskal–Wallis) and diabetes mellitus incidence (P=0.028, Kruskal–Wallis) differed between quartiles, but no statistically significant differences were observed in NSQIP-D scores, baseline test scores, or other demographic variables across quartiles.

Table 1.

Characteristics of participants with plasma interleukin-6 concentration at baseline, grouped by quartiles (low, low-mid, mid-high, high). Raw plasma interleukin-6 concentration ranges (pg ml⁻¹) were reported for each quartile. Median (inter-quartile range) was reported for continuous data (because of non-normal distribution) or n (%) was reported for discrete data. COWAT, Controlled Oral Word Association Test; ENT, ear, nose, and throat; na, not applicable; TIA, transient ischaemic attack; TMTA, Trail Making Test Part A; TMTB, Trail Making Test Part B. ∗Kruskal–Wallis rank sum test; Pearson's χ² test; Fisher's exact test. ^†P<0.05. ^‡National Surgical Quality Improvement Program (NSQIP) risk of death (NSQIP-D) by American College of Surgeons. ^¶All surgeries classified as general involved the gastrointestinal tract and were abdominal in location. ^§Values for race also included Asian, Native Hawaiian or other Pacific Islander, Black or African American, and more than one race. However, no participants had these racial backgrounds.

Characteristic	Overall (n=170)	Low (n=43)	Low-mid (n=43)	Mid-high (n=42)	High (n=42)	P-value∗
Interleukin-6 range (pg ml−1)	0.03–113.3	0.03–1.57	1.58–2.73	2.78–4.90	4.97–113.3	na
Age (yr)	71 (8)	70 (8)	71 (8)	72 (8)	72 (6)	0.4
Female	68 (40)	20 (47)	15 (35)	17 (40)	16 (38)	0.7
NSQIP-D‡	1.4 (3.9)	0.7 (3.1)	1.9 (3.7)	2.0 (4.0)	1.6 (3.8)	0.12
Operating time (min)	304 (244)	305 (250)	215 (179)	335 (332)	332 (200)	0.021†
Blood loss (ml)	400 (1375)	400 (1025)	400 (750)	500 (1925)	375 (2900)	0.7
Hypertension	135 (79)	31 (72)	32 (74)	37 (88)	35 (83)	0.2
Stroke/TIA	15 (8.8)	2 (4.7)	5 (12)	5 (12)	3 (7.1)	0.6
Ever smoked	128 (75)	29 (67)	34 (79)	33 (79)	32 (76)	0.6
Diabetes mellitus	37 (22)	4 (9.3)	8 (19)	15 (36)	10 (24)	0.028†
Surgery type						0.4
Cardiac	21 (12)	8 (19)	6 (14)	5 (12)	2 (4.8)
ENT	1 (0.6)	0 (0)	0 (0)	0 (0)	1 (2.4)
General¶	12 (7.1)	2 (4.7)	3 (7)	3 (7.1)	4 (9.5)
Orthopaedic	50 (29)	16 (37)	16 (37)	7 (17)	11 (26)
Thoracic	8 (4.7)	1 (2.3)	4 (9.3)	2 (4.8)	1 (2.4)
Urological	11 (6.5)	1 (2.3)	1 (2.3)	4 (9.5)	5 (12)
Vascular	67 (39)	15 (35)	13 (30)	21 (50)	18 (43)
Baseline test scores
TMTB	77 (46)	73 (42)	81 (40)	77 (58)	78 (45)	0.8
TMTA	35 (16)	32 (14)	36 (14)	37 (15)	34 (20)	0.2
COWAT	32 (17)	39 (22)	32 (16)	31 (16)	32 (16)	0.4
Mini-Cog total	2 (2)	2 (3)	2 (2)	3 (2)	2 (3)	0.8
Mini-Cog word recall	1 (2)	2 (2)	1 (2)	2 (2)	2 (2)	0.7
Education						0.8
≤12 yr	43 (28)	9 (24)	10 (24)	11 (28)	13 (33)
>12 yr	113 (72)	28 (76)	31 (76)	28 (72)	26 (67)
Ethnicity						>0.9
Hispanic or Latino	1 (0.59)	0 (0)	1 (2.3)	0 (0)	0 (0)
Not Hispanic or Latino	157 (92)	37 (86)	41 (95)	39 (93)	40 (95)
Unknown/not reported	12 (7.1)	6 (14)	1 (2.3)	3 (7.1)	2 (4.8)
Race§						0.8
Native American	1 (0.59)	0 (0)	0 (0)	1 (2.4)	0 (0)
White	156 (92)	37 (86)	41 (95)	38 (90)	40 (95)
Unknown/not reported	13 (7.6)	6 (14)	2 (4.7)	3 (7.1)	2 (4.8)

Data on PODs 4–9 were collected as part of a sub-study of the NeuroVISION trial that overlapped with the IPOD-B3 cohort. Participants with TMTB results collected on PODs 4–9 had a longer operating time (median=374, inter-quartile range=257–548 min) compared with 304 min for the overall cohort (P=0.035, Wilcoxon), but did not have significant differences in baseline IL-6 (3.5, 2.2–7.1 pg ml⁻¹), age (71, 68–75 yr), female proportion (44%), NSQIP-D score (1.6, 0.3–4.8), blood loss (600, 275–2250 ml), hypertension (81%), stroke/TIA (7.0%), ever smoked (77%), or diabetes mellitus incidence (23%). As these data are paired with cognitive data and we used generalised additive models in our analysis, it is unlikely this will bias the direct relationship of inflammation and cognition, though more data would increase the certainty of our findings at specific time points.

Primary outcome: Trail Making Test B

Levels of IL-6 and TMTB resolved over time after surgery (Fig 2, Fig 3; Supplementary Figs S1–S4), with IL-6 correlating with TMTB over 1 yr postoperatively (β=0.074, P<0.001, generalised additive model) (Table 2). Collectively, IL-6 and POD explained 20% of the variance observed in TMTB (R²=0.20) (Table 2). This relationship was not confounded by age, sex, education, hypertension, stroke/TIA, or ever having smoked. As an illustration of the raw data, we also plotted the univariate correlation of 1-yr change in log₁₀ TMTB and IL-6 (Spearman's rho [ρ]=0.33, P=0.019) (Fig. 2c, Supplementary Fig. S5). Relative changes in raw IL-6 over 1 yr similarly correlated with changes in non-logarithmically transformed TMTB (β=16, ρ=0.3, P=0.028) (Supplementary Fig. S3) (i.e. each 10-fold increase in IL-6 at 1 yr compared with baseline corresponded to a 16 s increase in TMTB time).

Fig 2.

Time courses of logarithm-transformed Trail Making Test Part B (TMTB) and plasma interleukin-6 concentration over postoperative days 1–9, day 90, and 1 yr; and relationship between 1-yr change in each. (a) Time course of log₁₀-transformed TMTB (s) normalised to baseline, over postoperative days 1–9, day 90, and 1 yr; n=170, observations (obs)=495. Trendline fitted using a generalised additive model with penalised spines for postoperative day. (b) Time course of log₁₀-transformed plasma IL-6 (pg ml⁻¹) normalised to baseline, over postoperative days 1–9, day 90, and 1 yr; n=173, obs=716. Trendline fitted using a generalised additive model with penalised spines for postoperative day. (c) Relationship between 1-yr change in log₁₀-transformed TMTB and 1-yr change in IL-6. n=51, Spearman's rho (ρ)=0.33, gradient (β)=0.087, P=0.019. Four outliers were excluded by Cook's distance using a conservative threshold of 4×mean. IL-6, interleukin-2.

Fig 3.

Time courses of cognitive test score and plasma interleukin-6 concentration over postoperative days 1–9, day 90, and 1 yr, comparing >1 standard deviation cognitive decliners and non-decliners. All values were log₁₀-transformed and normalised to baseline. Trendlines were fitted using generalised additive models with penalised spines for postoperative day. Participants with a >1 standard deviation decline from the baseline mean in an individual cognitive test at any time point are plotted in purple; all other participants are plotted in blue. Time courses of: (a) Trail Making Test Part B (TMTB) (s); n=170, observations (obs)=495. (b) Trail Making Test Part A (TMTA) (s); n=165, obs=456. (c) Controlled Oral Word Association Test (COWAT) (words in 3 min); n=171, obs=474. (d) Mini-Cog total; n=195, obs=944.

Table 2.

Generalised additive models for the relationships between cognitive test performance and interleukin-6 plasma concentration over time (unadjusted), with covariates of age, sex, education, hypertension, stroke or transient ischaemic attack, and ever smoked (adjusted). Continuous predictors were mean-centred and scaled by one standard deviation. P-values were calculated using Satterthwaite degrees of freedom. Penalised splines were generated for postoperative day, using an upper limit of k–1=2 degrees of freedom. Effective degrees of freedom (edf) were reported for postoperative day. The 95% confidence intervals were reported for effect sizes (β). COWAT, Controlled Oral Word Association Test; IL-6, interleukin-6; POD, postoperative day; TIA, transient ischaemic attack; TMTA, Trail Making Test Part A; TMTB, Trail Making Test Part B. ∗P<0.05.

	Dependent variable models
	TMTB	TMTA	COWAT	Mini-cog total	Mini-cog word recall
Unadjusted
(Intercept)	0.003 (−0.033 to 0.038)	0.009 (−0.018 to 0.035)	−0.002 (−0.031 to 0.027)	0.049 (−0.005 to 0.103)	0.126∗ (0.072–0.179)
IL-6	0.074∗ (0.029–0.119)	0.042∗ (0.004–0.080)	−0.058∗ (−0.092 to −0.024)	−0.062∗ (−0.107 to −0.018)	−0.079∗ (−0.121 to −0.037)
POD (edf)	1.599∗	1.786∗	1.613∗	1.000	1.000
Number of observations	191	166	178	407	407
Pseudo-R2 (fixed effects)	0.203	0.208	0.216	0.019	0.027
Adjusted
(Intercept)	−0.409 (−0.877 to 0.058)	−0.114 (−0.494 to 0.266)	0.123 (−0.247 to 0.493)	−0.465 (−1.108 to 0.179)	−0.593 (−1.260 to 0.073)
IL-6	0.075∗ (0.028–0.121)	0.045∗ (0.006–0.085)	−0.074∗ (−0.103 to −0.045)	−0.063∗ (−0.107 to −0.018)	−0.088∗ (−0.132 to −0.043)
POD (edf)	1.579∗	1.755∗	1.315∗	1.000	1.322
Age	0.006 (−0.001 to 0.012)	0.002 (−0.003 to 0.007)	0.000 (−0.005 to 0.004)	0.004 (−0.005 to 0.012)	0.006 (−0.003 to 0.014)
Male sex	−0.026 (−0.088 to 0.035)	−0.012 (−0.062 to 0.038)	0.052∗ (0.005–0.010)	0.081 (−0.002 to 0.163)	0.095∗ (0.009–0.181)
Education	−0.007 (−0.074 to 0.060)	0.002 (−0.054 to 0.058)	−0.048 (−0.101 to 0.006)	0.094∗ (0.007–0.180)	0.123∗ (0.034–0.213)
Hypertension	0.063 (−0.006 to 0.131)	−0.010 (−0.066 to 0.046)	−0.002 (−0.055 to 0.051)	−0.033 (−0.128 to 0.063)	−0.015 (−0.115 to 0.084)
Stroke/TIA	0.016 (−0.114 to 0.147)	0.034 (−0.066 to 0.135)	0.039 (−0.051 to 0.128)	−0.280∗ (−0.433 to −0.126)	−0.195∗ (−0.353 to −0.036)
Ever smoked	−0.017 (−0.080 to 0.046)	0.020 (−0.032 to 0.072)	0.024 (−0.025 to 0.073)	−0.044 (−0.133 to 0.044)	−0.068 (−0.160 to 0.025)
Number of observations	189	165	175	402	402
Pseudo-R2 (fixed effects)	0.214	0.187	0.254	0.075	0.065

Sensitivity analyses for primary outcome

Results were unchanged by rejection of outliers (Supplementary Table S3). Change in TMTB–A scores were also associated with change in IL-6 levels (β=0.119, P=0.007), but not POD. Generalised additive mixed models for absolute TMTB (no change), adjusted for baseline TMTB also showed similar results (Supplementary Table S4). Furthermore, a linear mixed-effects model with linear fitting for POD showed a significant relationship between IL-6 levels and TMTB over 1 yr (Supplementary Table S5).

Secondary outcomes: Trail Making Test A, Controlled Oral Word Association Test, and Mini-Cog

Generalised additive mixed models for change in IL-6 and POD correlated with TMTA (β=0.042, P=0.032) and COWAT (β=−0.058, P<0.001) performance over 1 yr (Table 2). Both the Mini-Cog total (β=−0.062, P=0.007) and Mini-Cog word recall (β=−0.079, P<0.001) scores were significantly associated with IL-6 level but not POD. COWAT and Mini-Cog word recall performances also covaried with sex; Mini-Cog total and word recall scores covaried with education >12 yr and prior history of stroke/TIA. Sensitivity analyses based on outlier rejection and linear mixed-effects models showed similar results (Supplementary Tables S3–S5).

Significant cognitive decliners

A 1-sd decline in any one of five cognitive tests (TMTB, TMTA, COWAT, Mini-Cog total, Mini-Cog word recall) was associated with a slower resolution of IL-6 levels over 1 yr (Supplementary Tables S6 and S7). In addition, participants who had >1-sd cognitive decline in TMTB or Mini-Cog appeared to follow a different cognitive trajectory to the non-decliners (Fig. 3). Together, these suggest that the nature of cognitive resolution may be associated with the rate of IL-6 resolution.

Similarly, a 1.5-sd decline in any individual cognitive test corresponded to an impaired resolution in IL-6 concentration over 1 yr (Table 3, Supplementary Table S8). Education >12 yr was a significant covariate for Mini-Cog total scores (β=0.106, P=0.008). Mini-Cog word recall was excluded from 1.5-sd analysis since the maximum observed change in score was 1.2 sd. A 2-sd decline in TMTA or COWAT was associated with an impaired IL-6 resolution trajectory over 1 yr (Supplementary Tables S9 and S10).

Table 3.

Linear mixed effects models for the relationships between interleukin-6 concentration and 1.5-sd cognitive decline in cognitive tests over time, with covariates of age, sex, education, hypertension, stroke or transient ischaemic attack, and ever smoked. All cognitive test scores were log₁₀-transformed. The difference between each participant's score and the baseline mean was divided by the baseline standard deviation to obtain a z-score per participant, per test, per time point. Participants were binarised for the variable ‘cognitive decline’ as either having a decline in >1.5 sd from the baseline average (coded as 1) or not (coded as 0). The 95% confidence intervals were reported for effect sizes (β). Continuous predictors were mean-centred and scaled by one standard deviation. P-values were calculated using Satterthwaite degrees of freedom. A 1.5-sd change in Mini-Cog total score was equivalent to a 1-sd change because of the limited range of scores. Mini-Cog word recall was excluded from analysis since the maximum observed change in score was 1.2 sd. COWAT, Controlled Oral Word Association Test; IL-6, interleukin-6; POD, postoperative day; sd, standard deviation; TIA, transient ischaemic attack; TMTA, Trail Making Test Part A; TMTB, Trail Making Test Part B. ∗P<0.05.

	Dependent variable models
	TMTB	TMTA	COWAT	Mini-cog total
(Intercept)	0.305∗ (0.071–0.540)	0.219 (−0.032 to 0.470)	0.367∗ (0.110–0.625)	0.904∗ (0.684–1.123)
1.5-sd cognitive decline	0.445∗ (0.158–0.733)	0.415∗ (0.143–0.687)	0.632∗ (0.228–1.036)	0.152∗ (0.002–0.303)
POD	−0.405∗ (−0.497 to −0.313)	−0.352∗ (−0.451 to −0.253)	−0.424∗ (−0.523 to −0.325)	−0.399∗ (−0.449 to −0.343)
Age	−0.066 (−0.157 to 0.026)	−0.080 (−0.179 to 0.018)	−0.017 (−0.114 to 0.080)	−0.012 (−0.092 to 0.068)
Sex	0.153 (−0.038 to 0.344)	0.098 (−0.109 to 0.304)	0.128 (−0.076 to 0.333)	−0.023 (−0.192 to 0.146)
Education	0.065 (−0.026 to 0.156)	0.057 (−0.040 to 0.153)	0.055 (−0.041 to 0.151)	0.106∗ (0.028–0.184)
Hypertension	0.034 (−0.185 to 0.252)	0.061 (−0.173 to 0.295)	−0.007 (−0.236 to 0.223)	−0.064 (−0.258 to 0.130)
Stroke/TIA	0.126 (−0.297 to 0.549)	0.199 (−0.228 to 0.626)	0.074 (−0.342 to 0.490)	0.092 (−0.227 to 0.411)
Ever smoked	0.008 (−0.192 to 0.207)	−0.013 (−0.231 to 0.205)	−0.003 (−0.216 to 0.211)	−0.138 (−0.321 to 0.044)
Number of observations	189	166	175	409
Pseudo-R2 (fixed effects)	0.374	0.348	0.377	0.309

The optimal cut-point for change in log₁₀ IL-6 to predict 1.5-sd decline in TMTB was 0.693, with a sensitivity of 0.750, specificity of 0.683, Youden's index of 0.433, and associated area under the receiver operating characteristic curve of 0.720 (95% CI 0.61–0.82); this corresponds to a 4.93× increase in plasma IL-6 from baseline. Most of these high IL-6 values (313/314) occurred in the first 9 days postoperatively, suggesting that for most participants, short-term anti-inflammatory therapy might be sufficient to hasten cognitive recovery.

Discussion

Delayed postoperative resolution of inflammation as indicated by elevated plasma IL-6 concentration was associated with worse TMTB scores over 1 yr after surgery in older adults. Similar results were observed with the TMTA, COWAT, and Mini-Cog assessments. The relationship between change in TMTB score and change in IL-6 level over 1 yr was robust to confounding variables, outlier rejection, and fitting linear and non-linear models. Significant cognitive decline, defined by >1-sd or >1.5-sd reduction from mean baseline in any individual neuropsychological test (TMTB, TMTA, COWAT, or Mini-Cog score), was also associated with change in plasma IL-6 over 1 yr. Cut-point analyses determined that a 4.93-fold increase in plasma IL-6 from baseline on PODs 1–9 correlated with significant executive dysfunction (>1.5-sd decline in TMTB), although further studies are advised to refine its diagnostic utility.

Our findings support the hypothesis that incomplete resolution of inflammation in surgical patients is associated with postoperative neurocognitive decline and impaired executive function. Postoperative neurocognitive disorder has been implicated in possible trajectories towards dementia, with associated changes in Alzheimer's disease biomarkers including β-amyloid and the microtubule-associated protein tau, enhancing plausibility for this link.⁶ Executive function is a potential early predictor of the progression of dementia onset.²⁷ IL-6 is known to be neurotoxic and to exacerbate other neurodegenerative disease processes.²⁸ Elevated cytokine levels, including IL-6, disrupt the blood–brain barrier. In animal models, IL-6 and bone marrow-derived monocytes thereby enter the hippocampus, activating microglia which release more IL-6 in a feed-forward loop.²⁸ This disrupts long-term potentiation in adult mice and promotes trans-signalling in hippocampal CA1 neurones,²⁹ which is a proposed pathway in the development of postoperative neurocognitive disorder in humans.²⁸ Therefore, monitoring inflammatory markers might facilitate institution of immunomodulatory therapy to minimise secondary cognitive injury.

There are scarce prior data exploring the relationship between changes in inflammation and cognition in surgical cohorts. A small cohort study (n=17) of adults aged 50–70 yr found that change in plasma IL-6 predicted performance on the Trail Making Test 7 days after cardiac surgery.⁸ In an observational study (n=34) involving cerebrospinal fluid measurements, postoperative trajectories of inflammatory biomarkers varied by cognitive decline at 3 months after elective orthopaedic surgery.⁷ These studies were limited by small sample size and limited follow-up time. In contrast, some large prospective cohort studies of community-dwelling participants have shown associations between cross-sectional plasma IL-6 levels and cognition,^19,20 but the longitudinal findings are equivocal. For example, the Mayo Clinic Study of Aging (n=1602) with a median of 2.7 yr follow-up did not observe a longitudinal relationship between serum IL-6 levels and cognitive scores.¹⁸ The Taiwan Longitudinal Study of Aging (n=596) conducted over 11 yr found that higher IL-6 concentrations were associated with lower baseline cognitive scores, but not change over time.³⁰ These studies were subject to confounding factors related to peripheral inflammation (e.g. surgery, chronic illness, hospital admission, and lifestyle factors), and not directly linked in time to an inflammatory stressor, as in this study. Cognitive outcomes might be impacted by the progression of subclinical dementia in older patients.³¹ Hence, we focused on patients undergoing planned elective surgery. If plasma IL-6 were a sensitive surrogate of executive dysfunction, its routine collection could trigger preoperative initiation of immunomodulatory therapy and guide its tapering and cessation when its side-effect profile outweighed the potential cognitive benefit. This threshold could be determined by evaluating the relative change in IL-6 sufficient to induce cognitive decline, as we have explored here.

This study has several strengths. Whereas some previous studies with similar rationale were limited by their cross-sectional design, our longitudinal approach measuring relative changes from baseline allowed for comparisons of inflammation and executive function over 1 yr. We accounted for several confounders, including age and lower education level, which are associated with impaired cognitive reserve and vulnerability to perioperative neurocognitive disorder.⁶ Despite known associations with cognition, our analyses suggested that these factors were not necessarily predictive of the speed of resolution of executive impairment after surgery. Sensitivity analyses with TMTB–A scores were performed to control for processing speed deficits examined by both Parts A and B, and to isolate the set-shifting component of Part B. IL-6 remained robust with TMTB–A, implying a fronto-subcortical pattern of impairment, although further studies are necessary to confirm this.

There are also limitations to our study. Firstly, these data are observational, and causality between the variables cannot be established, though our hypothesis has a solid biological foundation. Despite this, our multiple sensitivity analyses suggest that the associations found are robust. In our interpretation, we do not comment on when resolution of inflammation should occur. To define this with certainty, even greater sampling at different time points would be required, which is an important future direction for research. The single-centre study design and relative lack of racial and ethnic diversity could limit generalisability. Participants' abilities to perform cognitive tests, such as the TMTA and TMTB, could have been impacted by pain in the immediate postoperative period. While we controlled for education, we did not measure premorbid intelligence quotient or examine memory using a sensitive measure with extended delay. Future studies could explore mediating effects of cognitive reserve, intelligence quotient and memory, which might be relevant given the links between IL-6 and hippocampal integrity.³² Nonetheless our analyses included assessments of the cognitive tests before surgery to enhance detection of short-term changes in cognition and inflammation. Our conclusions were limited to 1 yr postoperative changes, and future investigations should determine whether these impairments persist longer. Several underlying physiological and pathological processes aside from the resolution of the postoperative inflammatory response could also impact plasma IL-6 levels, and these effects might introduce bias.

Other limitations include the absence of an independent replication cohort and the lack of analyses on subjective cognitive complaints or instrumental activities of daily living. Although we have data on activities of daily living, we only started collecting data on subjective memory complaints after the release of the postoperative neurocognitive disorder guidelines.²⁵ Furthermore, few patients had a >1.5-sd or >2-sd decline in cognitive tests. Nevertheless, our primary objective was to quantify how IL-6 changes in the surgical population ‘on average’, with sensitivity analyses for significant cognitive decliners. Although it is reassuring that a >1.5-sd decline was relatively rare in our population, the fact that inflammation could be a modifiable risk factor for delayed cognitive recovery is important and offers an opportunity for enhanced cognitive rehabilitation postoperatively. This also provides a translational bridge for animal studies, showing the relevance of inflammation to unresolved cognitive deficits.

Conclusions

Change in plasma IL-6 predicts 1-yr postoperative change in executive function, as quantified by Trail Making Test B and other neuropsychological tests. Substantial cognitive decline (>1-sd, >1.5-sd or, >2-sd) also correlated with postoperative plasma IL-6 concentration. These findings are consistent with a model in which impaired resolution of surgically-induced peripheral inflammation leads to neuropathological changes and postoperative neurocognitive disorder.⁵ Whereas further investigations are required with a larger sample size and extended follow-up, this research avenue could help target susceptible patients with interventional immunomodulatory strategies.

Authors’ contributions

Designed the research: RDS, RL, MM, SN, RAP

Collected data: MP, CC, DK, CR

Ran the analyses with input from RDS: JGW, JT

Wrote the manuscript with input from all authors: JGW, RDS

Declaration of interest

RDS is an editor of the British Journal of Anaesthesia. The other authors declare that they have no conflicts of interest.

Funding

US National Institutes of Health Grants (K23 AG055700, R01 AG063849-01) to RDS.

Handling editor: Hugh C Hemmings Jr

Appendix A. Supplementary data

The following is the Supplementary data to this article: