Postoperative delirium and changes in the blood–brain barrier, neuroinflammation, and cerebrospinal fluid lactate: a prospective cohort study

Abstract

Background

Case-control studies have associated delirium with blood–brain barrier (BBB) permeability. However, this approach cannot determine whether delirium is attributable to high pre-existing permeability or to perioperative changes. We tested whether perioperative changes in cerebrospinal fluid/plasma albumin ratio (CPAR) and plasma S100B were associated with delirium severity.

Methods

Participants were recruited to two prospective cohort studies of non-intracranial surgery (NCT01980511, NCT03124303, and NCT02926417). Delirium severity was assessed using the Delirium Rating Scale-98. Delirium incidence was diagnosed with the 3D-Confusion Assessment Method (3D-CAM) or CAM-ICU (CAM for the ICU). CSF samples from 25 patients and plasma from 78 patients were analysed for albumin and S100B. We tested associations between change in CPAR (n=11) and S100B (n=61) and delirium, blood loss, CSF interleukin-6 (IL-6), and CSF lactate.

Results

The perioperative increase in CPAR and S100B correlated with delirium severity (CPAR ρ=0.78, P=0.01; S100B ρ=0.41, P<0.001), delirium incidence (CPAR P=0.012; S100B P<0.001) and CSF IL-6 (CPAR ρ=0.66 P=0.04; S100B ρ=0.75, P=0.025). Linear mixed-effect analysis also showed that decreased levels of S100B predicted recovery from delirium symptoms (P=0.001). Linear regression demonstrated that change in plasma S100B was independently associated with surgical risk, cardiovascular surgery, blood loss, and hypotension. Blood loss also correlated with CPAR (ρ=0.64, P=0.04), S100B (ρ=0.70, P<0.001), CSF lactate (R=0.81, P=0.01), and peak delirium severity (ρ=0.36, P=0.01).

Conclusion

Postoperative delirium is associated with a breakdown in the BBB. This increased permeability is dynamic and associated with a neuroinflammatory and lactate response. Strategies to mitigate blood loss may protect the BBB.

Editor's key points.

• •Postoperative delirium is common in older patients and likely involves neuroinflammation and breakdown of the blood–brain barrier (BBB).
• •Participants undergoing non-intracranial surgery were assessed for delirium severity and CSF and plasma biomarkers for BBB integrity, inflammation, and neurological injury.
• •Perioperative increase in cerebrospinal fluid/plasma albumin ratio and plasma S100B correlated with delirium severity, incidence, and recovery, and change in S100B was independently associated with surgical risk, surgery type, blood loss, and hypotension.
• •This association of a perioperative increase in BBB permeability with delirium incidence and severity provides a potential therapeutic target for postoperative delirium.

Delirium is a disturbance in attention, cognition, and consciousness, and is often an acute physiological consequence of medical events, such as hospital admission, surgery, sepsis, and pharmacological intervention.¹ It is characterised by a sudden onset and fluctuating course not otherwise explained by a pre-existing neurological or medical conditions. Common in older adults, it affects up to 50% of those in hospital older than 65 yr with a healthcare burden estimated at $152 billion per annum in the USA.² Perioperative delirium has been associated with increased mortality,³ comorbidity,⁴ functional and neurocognitive decline,⁵ hospital readmission,⁶ and institutionalisation.⁷ A recent meta-analysis (71 studies) found that older inpatients with delirium experienced a mortality risk three times that of those without delirium.³

The pathogenesis of postoperative delirium likely involves systemic inflammation⁸ and the breakdown of the blood–brain barrier (BBB).⁹ It is argued that these two processes are linked,¹⁰ but perioperative steroids can reduce systemic inflammation without preventing damage to the BBB during cardiac surgery.¹¹ Understanding the mechanisms of BBB injury may be key to preventing delirium as animal studies suggest that the influx of inflammatory cells and mediators into the CNS likely drives changes in cognition.¹²

The BBB is a series of physical and chemical boundaries, or interfaces, between the CNS and other body systems.¹³ The BBB is mostly made up of vascular endothelial cells, joined by adhesion molecules known as ‘tight junctions’, and ependymal cells lining the ventricles and spinal cord in direct contact with CSF.¹⁴ However, the heavy metabolic demands of the CNS require the BBB to be dynamic and permeable to direct blood flow, influx and store energy, and efflux waste via neurovascular coupling.¹⁵ As demonstrated in animal and human studies, the BBB is selective and adaptive to specific brain regions, physiological functions such as sleep, age, genetic factors, and environmental factors, such as stress, illness, and injury, while maintaining the primary CNS functions of cellular signalling and glucose metabolism.¹⁴

BBB deterioration or dysfunction has been implicated in a number of neurological conditions such as Alzheimer's disease and other dementias.¹⁶ Case-control studies have demonstrated that delirium is associated with a breakdown in the BBB.^17,18 To our knowledge, there are no prior studies of temporal change in the BBB and delirium.¹⁹ Although inflammation is known to be associated with BBB disruption,²⁰ recent data strongly argue that plasminogen activation leads to BBB breakdown though both direct effects on astrocytes²¹ and exacerbation of inflammation.²² Given that blood loss and transfusion are associated with delirium,23, 24, 25, 26 we sought to understand if blood loss (with the ensuing fibrinolytic response) was associated with BBB breakdown.

Cerebrospinal fluid/plasma albumin ratio (CPAR) is the gold standard measure of BBB permeability, directly measuring the physical breakdown of the blood–CSF barrier.²⁷ We conducted parallel investigations using CPAR and S100-beta (S100B), a well-validated surrogate biomarker for BBB that is a calcium-binding protein in glial cells with a role in astrocytic cytoskeleton morphology. Elevated S100B in biofluids is a proposed marker of astrocytic activation/injury and BBB disruption. Increases in S100B have been demonstrated in a number of neurological conditions, including traumatic brain injury,²⁸ dementia,¹⁷ and delirium.²⁹ We investigated these two complementary biomarkers¹: CPAR is the gold standard but difficult to obtain, and (2) S100B is a surrogate but can be measured in plasma and hence can easily be accessed in the perioperative period facilitating statistical power for analysis of consecutive participants. A priori we decided that convergent associations on delirium and delirium severity would be required to be confident in our findings and reduce the chance of a false positive. We also considered that S100B and CPAR should change in a similar direction of effects for results to be biologically interpretable and credible.

Recent case-control studies have also suggested that CSF lactate is associated with delirium.^30,31 However, it is unknown if there are within-participant changes in CSF lactate paralleling the onset of delirium. Such a demonstration would be required to establish a mechanistic association that is worthy of further interrogation. Increases in lactate are consistent with the metabolic insufficiency hypothesis, which proposes that delirium results from impaired oxygen and glucose delivery.^32,33 We propose an alternative interpretation: that CSF lactate is a marker of neuroinflammatory inhibition whereby aerobic glycolysis generates lactate to induce localised immunosuppression via hydroxycarboxylic acid (HCA) receptor 2.³⁴ As an exploratory analysis we investigated the secondary working hypothesis that monocytes infiltrating through a permeable BBB provoke a neuroinflammatory response but CSF lactate biofeedback is induced to contain this inflammation. The neuroprotective somnogen, prostaglandin D2 (PGD2), is also produced as a by-product of this process.³⁵ We have concomitantly implicated PGD2 in the EEG changes of delirium in a recent mouse study, enhancing the plausibility of this relationship.³⁶

To investigate our hypotheses of increased permeability of the BBB with delirium and systemic inflammation, we address the following research questions:

• 1.Change in CSF to plasma albumin ratio, as a fluid biomarker of BBB permeability, is associated with delirium incidence and severity.
• 2.Change in S100B, a plasma biomarker of astrocytic injury/activation, is associated with delirium incidence and severity.
• 3.These changes are consistent with change in CSF interleukin-6 (IL-6), a key biomarker of delirium.³⁷
• 4.Change in CSF lactate is associated with inflammation and delirium severity.

Methods

Study design

This is an analysis of two ongoing observational, cohort studies of interventions for perioperative delirium (IPOD-B2 and IPOD-B3) at the University of Wisconsin (Madison, WI, USA). Recruitment commenced in July 2015. Participant consent was obtained in accordance with the Declaration of Helsinki. Ethical approvals were obtained from the University of Wisconsin–Madison (UWM) Institutional Review Board (2015-0374 and 2015-0960) and the trials were registered on clinicaltrials.gov (NCT01980511, NCT03124303, and NCT02926417). Data are reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Participants

Participants, recruited from clinical lists or information provided in clinics at UW Health, were eligible if aged 21+ yr (IPOD-B2) and 65+ yr (IPOD-B3), and undergoing elective, non-intracranial surgery with a minimum anticipated hospital stay of two days. Participants were excluded if they had a documented history of dementia, resided in a nursing home, or were unable to complete neurocognitive testing owing to language, vision, or hearing impairment. Complete inclusion/exclusion criteria are reported in Supplementary Table S1.

Study procedure

Prospects were screened by telephone and, if eligible, met in-person with researchers to ensure written, informed consent before hospital admission. At baseline, participants provided demographic and clinical information, a blood sample, and underwent cognitive assessments. At surgery, researchers collected updated procedure details from the medical record, performed delirium assessments, and obtained blood and CSF. CSF was only collected if a spinal drain was placed for clinical reasons for vascular surgery. During hospitalisation days 1–4, researchers obtained blood and performed further delirium assessments. If a participant was delirious the afternoon of postoperative day 4, they were followed up with cognitive testing (no biofluids) until their delirium resolved.

Measurements

For participants in this analysis, researchers assessed delirium postoperatively on hospitalisation days 0–4 using the 3-min Diagnostic interview for Confusion Assessment Method-defined delirium (3D-CAM) or CAM for the Intensive Care Unit (CAM-ICU) and the Delirium Rating Scale Revised (DRS-R-98). The DRS has similarly demonstrated validity and specificity for delirium severity, and maximum (peak) DRS is the primary outcome used in this analysis. Researchers also collected the results of the National Surgical Quality Improvement Program risk of death (NSQIP-D) assessment with procedure details from the medical record. Cognitive assessments Trail-making Test B (TMTB), Montreal Cognitive Assessment (MoCA), and the Controlled Oral Word Association Test (COWAT) were also conducted pre-postoperatively.

Researchers collected biofluid samples using standard venepuncture or an indwelling line, obtaining a maximum of 80 ml (8×10 ml) of blood and 25 ml (5×5 ml) of CSF during the study in EDTA (ethylenediaminetetraacetic acid) tubes as described.³⁸ Samples were de-identified, anonymised, labelled, processed, and stored at –80°C in a laboratory at the Clinical Sciences Center, UWM.

Data processing and analysis

Consecutive participant plasma and all available CSF biofluids were analysed by techniques such as enzyme-linked immunosorbent assay (ELISA; Eve Technology, Calgary, AB, Canada) as described.^38,39 Albumin concentrations were measured by immunoturbidimetry on a Cobas instrument (Roche Diagnostics, Penzberg, Germany), and neurofilament light (NfL) was analysed using a single-molecule array method as described previously.³⁸ Blood pressure was calculated using area under the curve (AUC) for intraoperative mean arterial pressures <10% of preoperative levels.

Data analysis was conducted in R (R Foundation for Statistical Computing, Vienna, Austria). Data were inspected and cleaned, and data sources were consolidated by participant identifier, time point, and consistent batch where available. Postoperative day one samples were used unless unavailable, then the next available in-patient sample was used. All biomarkers were normalised by logarithm (base 10). Unless otherwise specified, biomarker units are expressed in pg ml⁻¹.

Analyses included summary statistics, plots, and Shapiro–Wilk's test for normality. A Mann–Whitney/Wilcoxon rank sum test with continuity correction was used to differentiate delirious and non-delirious groups in univariate analysis. Correlations used Pearson's method or Spearman's method where data were not normally distributed. Outliers were validated by Cook's distance, using a conservative (4× mean) threshold. Continuous data are reported as mean (m) and standard deviation (sd), or median (M) and inter-quartile range (IQR) where not normally distributed. Biomarker analysis used preoperative-to-postoperative change rather than postoperative values. Categorical data are reported as counts (n) and percentage.

Missing data were not imputed and models are based on complete cases. We conducted power analysis on plasma S100B as we could collect plasma samples unrestricted by clinical events. Consistent with previous studies,⁴⁰ we determined a power threshold of 56 participants (α=0.05), based on a 300 (400) (m [sd]) ng difference in S100B between those with and those without delirium. We performed linear regression modelling for delirium severity with maximum (peak) DRS and a Poisson distribution family as confirmed by histogram, Shapiro–Wilks’ test (P<0.001), and skewness (1.32). We also conducted a linear mixed model for delirium severity (peak DRS), fit with random-effects maximum likelihood (REML) for individual participant and time. P values were calculated using Satterthwaite degrees of freedom.

Results

From the IPODB2 cohort of 32 participants, samples for CPAR (n=25) were analysed. From the combined IPODB2 and IPODB3 cohort of 179 participants, plasma S100B (n=78) were analysed. These consecutive samples were sent as part of a second batch of samples for cytokine analysis from our cohort study. Missing data were attributable to clinical restrictions on the collection of CSF or were planned as part of the timing of batched cytokine assays. As seen in the STROBE diagram (Fig. 1), two (2/25, 8%) participants were excluded from CPAR baseline and three (3/78, 4%) from plasma S100B, of whom 4/5 had surgery cancelled or abandoned and hence delirium data are not available. One patient was unable to be assessed for delirium because of Richmond Agitation Sedation Scale (RASS) of –4 to –5 at the time of assessment. One further participant was excluded because DRS could not be adequately assessed while intubated. This resulted in 22 participants with CPAR at baseline, of which 11 (11/22, 50%) had preoperative and postoperative samples. Seventy-four participants had plasma S100B samples at baseline. Of these, 63 (63/74, 85%) participants had preoperative and postoperative samples with a further two excluded following Cook's distance outlier analysis. Cohort demographic information is summarised in Table 1.

Fig 1.

STROBE diagram. CPAR, CSF/plasma albumin ratio; STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.

Table 1.

Characteristics of participants with CPAR and plasma S100B at baseline. ∗Median (M) and inter-quartile range (IQR) are reported as variables are not normally distributed. ^†National Surgical Quality Improvement Program (NSQIP-D) is used to assess surgical risk. ^‡Blood loss data excludes two outliers identified by Cook's distance. ^¶Surgery type is grouped into Cardiac or Vascular (CV), and Other (Thoracic, General, Orthopaedic, ENT, and Urological). ^§Arterial pressure is reported as baseline area under the curve (AUC) for intraoperative mean arterial pressures <10%. ^||TMTB and MoCA exclude participants with baseline testing on the day of surgery. COWAT, Controlled Oral Word Association Test; CPAR, cerebrospinal fluid/plasma albumin ratio; MoCA, Montreal Cognitive Assessment; TMTB, Trail-making Test B.

Variable	Participants with baseline CPAR					Participants with baseline plasma S100B
	Total with baseline CPAR (n=22)	Participants with baseline CPAR with delirium (n=15)		Participants with baseline CPAR without delirium (n=7)		Total with baseline plasma S100B (n=74)	Participants with baseline plasma S100B with delirium (n=29)		Participants with baseline plasma S100B without delirium (n =45)
	M (IQR)∗ (range)	n (%)	M (IQR)∗ (range)	n (%)	M (IQR)∗ (range)	M (IQR)∗ (range)	n (%)	M (IQR)∗ (range)	n (%)	M (IQR)∗ (range)
Age (yr)	66 (11) (38–82)	15 (68)	66 (10) (38–78)	7 (32)	72 (10) (59–82)	71 (8) (38–85)	29 (39)	69 (7) (38–77)	45 (61)	73 (7) (59–85)
Female	70 (11) (38–82)	8 (80)	70 (13) (38–78)	2 (20)	75 (8) (67–82)	71 (6) (38–83)	14 (52)	69 (4) (38–77)	13 (48)	72 (7) (65–83)
Male	65 (8) (47–78)	7 (58)	63 (7) (47–67)	5 (42)	72 (9) (59–78)	71 (8) (47–85)	15 (32)	67 (6) (47–77)	32 (68)	73 (6) (59–85)
NSQIP-D†	4 (5) (0–33)	14 (70)	5 (8) (1–33)	6 (30)	1 (4) (0–8)	3 (4) (0–15)	29 (40)	4 (5) (0–15)	43 (60)	2 (3) (0–8)
Operating time (min)	467 (194) (180–938)	15 (68)	539 (144) (185–938)	7 (32)	295 (156) (180–531)	296 (258) (90–938)	29 (39)	442 (326) (185–938)	45 (61)	296 (258) (90–557)
Blood loss (ml)‡
Surgery type¶
Cardiac or vascular	5000 (6250) (7–12 000)	14 (67)	6000 (3125) (7–12 000)	7 (33)	200 (275) (50–5000)	300 (2288) (0–12 000)	21 (41)	2500 (6400) (0–12 000)	30 (59)	150 (350) (0–5000)
Other	–	–	–	–	–	400 (580) (5–3000)	7 (32)	500 (550) (250–1100)	15 (68)	250 (535) (5–3000)
Arterial pressure§ (log10 AUC 10%)	5 (1) (4–6)	15 (68)	5 (1) (4–6)	7 (32)	5 (0) (4–6)	5 (1) (3–6)	28 (38)	5 (1) (4–6)	45 (62)	5 (1) (3–6)
Baseline TMTB||	61 (36) (40–144)	6 (67)	76 (32) (50–144)	3 (33)	54 (23) (40–85)	70 (35) (36–43)	23 (32)	72 (30) (36–143)	42 (68)	67 (36) (31–89)
Baseline MoCA	24 (1) (20–26)	5 (62)	23 (1) (23–26)	3 (38)	24 (0) (20–24)	23 (4) (19–27)	20 (32)	23 (3) (19–27)	42 (68)	23 (4) (13–29)
Baseline COWAT	35 (16) (24–51)	7 (64)	39 (15) (24–49)	4 (36)	33 (9) (26–51)	37 (20) (13–91)	23 (35)	40 (21) (17–58)	42 (65)	36 (19) (13–91)

Primary outcome

There were no between-group differences in preoperative CPAR or plasma S100B levels between participants with postoperative delirium vs those without (Wilcoxon P>0.05). As shown in Figure 2a and b, there were greater increases in CPAR and plasma S100B in those with delirium compared with those without postoperatively over time. Importantly, in our analysis of postoperative change to baseline, Figure 2c and d boxplots show an association of delirium incidence with changes in both CPAR (Wilcoxon P=0.012) and plasma S100B (Wilcoxon P=0.001). Similarly, Figure 2e and f show that peak delirium severity is correlated with change in CPAR (ρ=0.78, P=0.005) and plasma S100B (ρ=0.41, P<0.001). Figure 2g and h show that an established biomarker of delirium, CSF interleukin (IL)-6, also correlates with CPAR and plasma S100B, linking BBB permeability to neuroinflammation.

Fig 2.

Perioperative time course of CPAR and plasma S100B, and pre-postoperative change in delirium incidence, and peak DRS. (a) Time course of change in log10 CPAR normalised to baseline. (b) Time course of change in log10 plasma S100B normalised to baseline. (c) Boxplot of change in log10 CPAR by delirium incidence (delirium ever over postoperative days 1–4). (d) Boxplot of change in log10 plasma S100B by delirium incidence. (e) Correlation plot of change in log10 CPAR to peak DRS-R-98. (f) Correlation plot of change in log10 of plasma S100B to peak DRS. (g) Correlation plot of change in log10 CSF IL-6 to change in log10 CPAR. (h) Change in log10 CSF interleukin-6 (IL-6) to change in log10 plasma S100B. CPAR and plasma S100B were normalised by log10 transformation and subtracting baseline from postoperative values. Plasma S100B excludes two outliers, and CSF IL-6 excludes one, based on Cook's distance (>4∗mean). Delirium incidence is measured by the 3 min Diagnostic Cognitive Assessment Method (3D-CAM). Spearman's correlations were used given peak DRS is not normally distributed (Shapiro–Wilks’ normality test, P<0.001). The original unit used for biomarkers was pg ml⁻¹. Chg, change; CPAR, CSF/plasma albumin ratio; DRS-98, Delirium Rating Scale-98; Norm., normalised.

Observing increased plasma S100B proportionately with delirium severity, we tested whether this effect was robust to confounding. In a Poisson regression model, adjusting for age, sex, and baseline cognition (Trail Making Test B), change in plasma S100B predicted peak delirium severity (Supplementary Table S2; z=2.85, P=0.004). Using a linear mixed-effects model, with random effects for participant and time, S100B predicted delirium severity over time (Supplementary Table S3; β=4.36, P=0.001). In summary, S100B changes correlated with both increases in postoperative delirium severity from baseline and recovery.

Secondary outcome: associations with blood loss

As blood loss is associated with delirium, we hypothesised that plasmin activation is associated with exacerbated inflammation and BBB breakdown.²¹ We tested whether blood loss was associated with markers of (1) BBB breakdown and (2) systemic inflammation. Change in CPAR (ρ=0.64, P=0.044), plasma S100B (ρ=0.70, P<0.001) and delirium severity (ρ=0.36, P=0.008) correlated with blood loss (Fig. 3). Blood loss similarly correlated with CSF IL-6 (ρ=0.81, P=0.008) and plasma IL-8 (ρ=0.30, P=0.028), a cytokine marker of peripheral inflammation associated with delirium.^38,41

Fig 3.

Correlation plots of preoperative–postoperative change in CPAR, plasma S100B, and other biomarkers with blood loss. (a) Correlation of change in log10 CPAR with log10 blood loss. (b) Correlation of change in log10 CSF IL-6 with log10 blood loss. (c) Correlation of change in log10 plasma S100B with log10 blood loss. (d) Correlation of change in log10 plasma IL-8 with log10 blood loss. (e) Correlation of log10 blood loss with peak DRS. Delirium incidence is measured by the 3-min Diagnostic Cognitive Assessment Method (3D-CAM). Outliers of blood loss were identified by Cook's distance (>4∗mean) and excluded as indicated. Spearman's correlation method used because blood loss is not normally distributed (Shapiro–Wilks’ normality test, P<0.001). The original unit of biomarkers is pg ml⁻¹ unless otherwise specified, and the original unit of blood loss is ml. Chg, change; CPAR, CSF/plasma albumin ratio; DRS, Delirium Rating Scale.

We also investigated clinical predictors of BBB breakdown with change in plasma S100B as a surrogate, using a linear regression model with the following factors: age, female sex, blood loss, NSQIP-D, stroke/transient ischaemic attack (TIA), and surgery type. Blood loss (P=0.001), low blood pressure (P<0.001), age (P=0.042), surgical risk (NSQIP-D, P=0.001), and cardiovascular surgery (P=0.001) were associated with change in plasma S100B (adjusted R²=0.578, P<0.001; Supplementary Table S4).

Secondary outcome: associations with CSF lactate

We hypothesised that CSF lactate is a biomarker of a neuroinflammatory response and hence is associated with delirium, inflammation, and BBB metrics (and also blood loss). Change in CSF lactate was associated with delirium severity (peak DRS, ρ=0.66, P=0.038), blood loss (R=0.81, P=0.005), BBB permeability indicated by CPAR (R=0.78, P=0.008) and neuroinflammation indicated by CSF IL-10 (R=0.77, P=0.010; Fig. 4). It was also associated with peripheral inflammation indicated by plasma IL-8 (ρ=0.87, P=0.003) and plasma IL-10 (ρ=0.76, P=0.016). CSF lactate demonstrated a trend towards an association with CSF IL-6 but did not correlate with plasma S100B or CSF neurofilament light in this small sample.

Fig 4.

Correlation plots of change in log10 CSF lactate and biomarkers. Delirium incidence is measured by the 3-min Diagnostic Cognitive Assessment Method (3D-CAM). There was one CSF lactate outlier excluded based on Cook's distance and another excluded for plasma S100B. Pearson correlation method indicated by R. Spearman correlation method indicated by ρ (rho). Chg, change; CPAR, CSF/plasma albumin ratio; DRS, Delirium Rating Scale; NfL, neurofilament light.

Discussion

We observed a dynamic breakdown in the BBB associated with delirium and delirium severity using both the gold standard (CPAR) and surrogate (S100B) measures. This dual approach was undertaken to reduce the possibility of false-positive results from a single biomarker. Delirium severity also resolved in parallel with reducing levels of S100B in our mixed-effects regression further supporting causal associations between BBB injury and delirium.

Our data are consistent with a model in which surgical trauma induces a peripheral inflammatory response and blood loss exacerbates that response to induce delirium (Fig. 5). These data are consistent with multiple univariate and multivariable regression analyses. They are also convergent with many animal studies suggesting that surgery-induced cognitive dysfunction is associated with an influx of immune cells through a damaged BBB^42,43 and that BBB breakdown is associated with increases in CSF IL-6 (an established biomarker of delirium).⁴⁴ A recent study demonstrated proof of principle for the neuroinflammatory hypothesis contributing to postoperative neurocognitive disorder in humans,¹⁹ and we extend this concept directly to delirium.

Fig 5.

Blood–brain barrier permeability hypothesis. We propose a mechanistic pathway by which peripheral inflammation, exacerbated by plasmin activation, leads to breakdown of the blood–brain barrier. Subsequent inflammatory changes lead to a central anti-inflammatory response, including lactate-induced activation of HCA receptors. The downstream effects on PGD2 lead to delirium though synaptic suppression. This results in reduced cerebral activity evident in EEG slowing. HCA, hydroxycarbolic acid; PGD2, prostaglandin D2.

Our mechanistic link to blood loss highlights a potential therapeutic opportunity for preventing fibrinolysis–plasmin activation in reducing postoperative delirium. For example, tranexamic acid is reported to be anti-inflammatory and reduce BBB breakdown in animal models⁴⁵ and hence may reduce postoperative delirium. Tranexamic acid should be tested in clinical trials, as proposed by Whitlock and Behrends,²³ given the proposed causal link between delirium and BBB breakdown. Considering the prior RCT of steroids in cardiac surgery,¹¹ steroids alone may be ineffective in protecting the BBB, especially in the setting of blood loss. Future studies should consider targeting both inflammation and blood loss to reduce the risk of delirium.

We extended these findings based on analysis of CSF lactate. We demonstrated that CSF lactate changes with delirium; this is critical as prior data were based on case-control studies and could not exclude pre-delirium differences in lactate. We propose an alternative hypothesis concerning the increase in CSF lactate: that it moderates the immune response. This idea is consistent with both inflammatory models of delirium and with prostaglandin models, which have gained substantial weight recently.^36,46,47 Increases in prostaglandins have been linked to both cognitive impairment⁴⁷ and EEG slowing³⁶ in animal models of inflammation. PGD2, a powerful somnogen, is the likely prostaglandin effector that drives EEG slowing.^36,46 Given that PGD2 is also neuroprotective after a stroke,³⁵ the possibility that suppression of brain activity contributes to this effect needs to be clarified. Our proposed pathway links these different concepts, providing hypotheses that can be tested in animal models (Fig. 5). Our interpretation implies that lactate and prostaglandin release may represent endogenous protective mechanisms, and altering them may lead to exacerbated inflammation and loss of neuroprotective effects.

By extension, this work suggests a re-appraisal of inflammatory, metabolic insufficiency and cognitive disintegration⁴⁸ hypotheses of delirium. Notably, the anti-inflammatory cytokine IL-10, not neuronal injury, correlated with changes in CSF lactate. If metabolic insufficiency was occurring sufficient to induce neuronal injury, we would have expected there to be a correlation of neurofilament light and CSF lactate. Although we cannot exclude that we are underpowered to identify the latter relationship, the closer relationship of IL-10 with CSF lactate highlights the known biological mechanism of lactate as an anti-inflammatory effector.^49,50 Further studies are required to understand whether lactate in this context reflects metabolic failure or an active response to dampen inflammation and neuronal energetic requirements, possibly via PGD2.

There are several limitations to our study. These data are observational, and causality cannot be ascribed. However, we sought multiple ways to test our hypotheses and identified convergence across two assays and between endpoints of delirium severity and incidence, giving us confidence in our findings. Future studies of S100B may wish to focus on serum rather than plasma given that EDTA might affect S100B levels. There is also heterogeneity in the types of surgery in this sample, including cardiac, vascular, and orthopaedic surgery, which may be associated with different mechanisms of injury. Indeed, surgical risk predicted S100B levels and future cohorts should consider focusing on specific surgeries. The variation in precipitating factors such as surgical risk, and associated variance in blood loss, may also explain why certain predisposing risk factors, such as age,¹ were not significantly associated with delirium severity. We conclude that, in our cohort that focuses on major surgery, the majority of risk of delirium is determined by the precipitating factors. This strengthens our ability to study dynamic changes that drive delirium but may weaken our ability to study predisposing risk factors.

Biological gradients for the effects we observed support causality,⁵¹ and we have proposed an intervention through which we can directly test the hypothesis (tranexamic acid, e.g. as a sub-study of NCT04192435). We have also adjusted for confounders where possible. Nonetheless, our sample and event sizes are small and we would caution about extrapolating our findings to other clinical situations, in particular where blood loss may be limited. However, evidence from multiple sources suggests that inflammation in the absence of blood loss may induce similar effects.⁴³ Moreover, the lack of association between CSF lactate and CSF neurofilament light may be a type II error reflecting the small sample rather than a robust lack of association. This analysis confirms a known association between systemic inflammation, blood loss, and delirium supported by factors such as surgical risk and operating times.⁵² It is consistent with the one available study on temporal changes in the BBB in response to perioperative inflammation followed by postoperative cognitive decline.¹⁹ Our analysis adds to this work by demonstrating that a perioperative increase in BBB permeability is associated with delirium incidence and severity. Our analyses highlight possible opportunities and pitfalls for delirium science and demonstrate that mechanistic research in delirium is urgently required.

Authors' contributions

Original principal investigator: RDS.

Current principal investigator: RAP.

Study design: RDS, RAP, in consultation with co-investigator RCL.

Recruitment of participants, and data collection and processing: MP, CPC, DK, CR.

Assays and management of biofluids analysis: HZ and KB.

Processing of EEG data: ST.

Data analysis and drafting of the manuscript: RDS, JT.

All authors provided critical feedback on the manuscript.

Declarations of interest

All authors declare no competing interests that may be relevant to the submitted work. HZ has served at scientific advisory boards, as a consultant, or both for Abbvie, Alector, Eisai, Denali, Roche, Wave, Samumed, Siemens Healthineers, Pinteon Therapeutics, Nervgen, AZTherapies, CogRx, and Red Abbey Labs, has given lectures in symposia sponsored by Cellectricon, Fujirebio, Alzecure and Biogen, and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS). KB has served as a consultant or at advisory boards for Alector, Alzheon, CogRx, Biogen, Lilly, Novartis, and Roche Diagnostics, and is a co-founder of BBS, all unrelated to the work presented here.

Funding

US National Institutes of Health R01 grant (AG063849-01) to RDS, RCL, and RAP, and K23 grant (AG055700) to RDS. Study data are available to qualified investigators upon reasonable request.

Handling editor: Hugh C Hemmings Jr

Appendix A. Supplementary data

The following is the Supplementary data to this article: