Role of inflammatory markers in the diagnosis of vascular contributions to cognitive impairment and dementia: a systematic review and meta-analysis

Vascular contribution to cognitive impairment and dementia (VCID) is a clinical label encompassing a wide range of cognitive disorders progressing from mild to major vascular cognitive impairment (VCI), which is also defined as vascular dementia (VaD). VaD diagnosis is mainly based on clinical and imaging findings. Earlier biomarkers are needed to identify subjects at risk to develop mild VCI and VaD. In the present meta-analysis, we comprehensively evaluated the role of inflammatory biomarkers in differential diagnosis between VaD and Alzheimer’s disease (AD), and assessed their prognostic value on predicting VaD incidence. We collected literature until January 31, 2021, assessing three inflammatory markers [interleukin(IL)-6, C-reactive protein (CRP), tumor necrosis factor (TNF)-α] from blood or cerebrospinal fluid (CSF) samples. Thirteen cross-sectional and seven prospective studies were included. Blood IL-6 levels were cross-sectionally significantly higher in people with VaD compared to AD patients (SMD: 0.40, 95% CI: 0.18 to 0.62) with low heterogeneity (I2: 41%, p = 0.13). Higher IL-6 levels were also associated to higher risk of incident VaD (relative risk: 1.28, 95% CI: 1.03 to 1.59, I2: 0%). IL-6 in CSF was significantly higher in people with VaD compared to healthy subjects (SMD: 0.77, 95% CI: 0.17 to 1.37, I2: 70%), and not compared to AD patients, but due to limited evidence and high inconsistency across studies, we could not draw definite conclusion. Higher blood IL-6 levels might represent a useful biomarker able to differentiate people with VaD from those with AD and might be correlated with higher risk of future VaD. Supplementary Information The online version contains supplementary material available at 10.1007/s11357-022-00556-w.


Introduction
Vascular contributions to cognitive impairment and dementia (VCID) are conditions arising from vascular diseases or abnormalities that result in a wide range of cognitive disorders progressing from mild to major vascular cognitive impairment (VCI), which is also defined as vascular dementia (VaD) [1,2]. Among the different forms of dementia, VaD is considered the second most common cause after Alzheimer's disease (AD), accounting approximately the 20% of dementia cases [2]. Its prevalence is estimated to be 0.6-2.1% in subjects aged over 65 years [3], and it increases with age, up to 4.8% in those over 85 years [4]. Cerebrovascular disease is the main etiological feature of VCID, independent of the underlying mechanism (e.g., multiple or single territorial or small infarcts, strategic infarcts) and the occurrence of stroke symptoms [5]. However, it is becoming clear that white matter damage and cognitive impairment occur also in absence of stroke symptoms, suggesting that often there is a silent and slow progression of the disease due to involvement of cerebral small vessels [6]. For example, VCID can be detected in subjects suffering from arrhythmias (e.g., atrial fibrillation) or other vascular risk factors (e.g., tobacco use, hypertension, obesity, hyperlipidemia, diabetes, hyper-coagulation), but without stroke history. In addition, vascular manifestations in older adults often fluctuate over time, resulting in diagnostic delay and ineffective treatments [7].
The main cerebrovascular signs of VCID are brain atrophy, white matter hyperintensities (WMH) lesions, infarctions, and hemorrhages. Indeed, imaging techniques (e.g., magnetic resonance imaging or computed tomography) represent an essential step in the diagnosis and evaluation of disease progression. However, evidence at neuroimaging is already a sign of advanced stage of disease and non-reversible brain damage. As for the definition of AD with introduction of amyloid, tau, neurodegeneration (AT[N]) system [8], also for VCID, there is a need to change the paradigm traditionally based on clinical history and signs and symptoms of the disease, toward a biological framework founded on early biomarkers able to predict development of VCID. This approach could allow the recognition of subjects at risk in a preclinical phase and timely the implementation of potential preventive strategies.
To date, no specific circulating biomarker is available in the diagnosis of VCID [9]. Over the last several years, there has been growing interest in addressing the relationship between inflammation, cardiovascular diseases, and cognitive dysfunction [10]. It is becoming clear that inflammation may have a role in the pathogenesis of dementia. As already reviewed, promising perspectives came from inflammatory biomarkers in predicting risk of overall dementia [11]. Previous meta-analyses showed that increased circulating interleukin(IL)-6 and C-reactive protein (CRP) levels were associated to higher risk of dementia from all causes, but not to AD [12,13]. These studies did not test the role of inflammatory biomarkers in differential diagnosis between VaD and AD, and did not include inflammatory markers from cerebrospinal fluid (CSF). We hypothesized that inflammatory biomarkers may be increased to a greater extent in VCID compared to AD; thus, the goal of this review was to examine the diagnostic and predictive power of selected inflammatory biomarkers for VCID.

Search strategy and study selection
The present systematic literature review and metaanalysis followed the requirements of the PRISMA statement [14]. An a priori protocol was established and registered on PROSPERO, an international prospective register of systematic reviews (http:// www. crd. york. ac. uk/ PROSP ERO; registration number: CRD42021268548).
Two study authors (C.C. and A.C.) independently conducted a systematic search of the databases MED-LINE, PubMed, Scopus, Web of Science, and Google Scholar until January 31, 2021. Search terms included combinations of the following keywords: ("C-reactive protein" OR "interleukin-6" OR "tumor necrosis factor-α" OR "inflammation" OR "inflammatory marker") AND (("blood" AND "vessels") OR "blood vessels" OR "vascular")) AND ("cognitive impairment" OR "dementia"). To be included in the present meta-analysis, studies had to be observational with either cross-sectional or longitudinal design. Studies were required to meet the following inclusion criteria: (1) conducted in humans; (2) (4) prospective studies including subjects with dementia at baseline; (5) autoptic studies. Articles were initially screened based on title and abstract by two study authors (C.C., and A.C.), with the full text sought if the abstract did not provide sufficient information. Reference lists of the articles were reviewed to identify additional relevant articles. Disagreement was resolved by discussion or in consultation with a senior author (V.S.). We contacted the authors of primary studies to obtain any missing information.

Data extraction
The following details were extracted from each study: first author's name, publication year and country, sample size, details of study population (mean age, health status), study duration, study design, assessed inflammatory markers, definition of dementia diagnosis, potential confounders that were considered in the analysis and main results. For prospective studies, we extracted the reported effect estimates (relative risk (RR) or hazard ratio (HR)) and the corresponding 95% confidence interval (CI) derived from the most fully adjusted model for potential confounders if studies reported several multivariable-adjusted RRs.

Risk of bias assessment
The quality of the studies was assessed using appropriate tool for observational studies: the Newcastle-Ottawa Quality Assessment Scale (NOS) [15]. Two study authors (A.C. and D.G.) assigned a rating, using stars, based on three domains: selection of study population (0-4 stars), comparability of study groups (0-2 stars), ascertainment of outcome (for cohort studies) or exposure (for case-control studies) (0-3 stars). The final NOS score for each study ranges from 0 stars (lowest quality) to 9 stars (highest quality), with studies scoring 0-3 stars judged as low quality, those between 4 and 6 as medium quality, and those between 7 and 9 considered to be of high quality. Discrepancies in the evaluation were solved by discussion. The reliability of assessment was ensured by revision and consultation with a senior author (V.S.).

Statistical analysis
For studies with cross-sectional design, each study's effect size, or standardized mean difference (SMD), was calculated by comparing mean and standard deviation of inflammatory biomarkers, between VaD and control or AD groups [16]. If the data were reported as median and interquartile range, the correspondent mean and standard deviation were estimated using the method developed by Wan and colleagues [17].
In accordance with convention, effect sizes were classified as small (0.2), moderate (0.5), and large (0.8) [18]. For studies with longitudinal design, study-specific risk estimates were extracted from each article, and log risk estimates were weighted by the inverse of their variances to obtain a pooled risk estimate. The primary analyses combined ln RR associated with one-unit change in inflammatory markers. Studies were combined using the DerSimonian and Laird random-effects model, which considers both within-and between-study variations [16]. Heterogeneity across studies was estimated by I 2 statistic. It measures percentage of variation that is caused by heterogeneity between studies, and is larger when heterogeneity increases [16]. Sensitivity analyses were performed to investigate the influence of each individual study on the overall meta-analysis summary estimate and the validity of the effect size. Further sensitivity analysis was performed to determine the robustness of findings by excluding studies with poor-quality assessment (NOS score < 7). Funnel plots and Egger's tests were utilized to detect bias in meta-analyses [16]. Statistical analyses were performed in RevMan 5.4 (The Cochrane Collaboration, Oxford, England) and STATA 14.0 software (StataCorp LP, College Station, TX, USA). Each p value is based on two-sided alternative hypothesis, and a level of 0.05 or below was considered statistically significant.

Quality assessment
We evaluated distribution of the risk of bias across the 20 studies included in the quantitative synthesis. The quality of the included studies ranged from medium to high, with comparability for case-control study and ascertainment of outcome for cohort studies as the major concerns for potential sources of bias (Supplementary Table 1).

Interleukin-6 in cerebrospinal fluid and vascular dementia
For quantitative synthesis, only three studies were eligible which compared IL-6 levels in the CSF between 82 patients with VaD, 99 with AD, and 81 healthy subjects [23,24,28]. IL-6 was significantly higher in people with VaD compared to healthy subjects (SMD: 0.73, 95% CI: 0.12 to 1.34) (Fig. 5A), but not compared to AD patients (SMD: 0.14, 95% CI: − 0.65  (Fig. 5A and B, respectively), together with poor overall quality of the studies, limited the reliability of these findings.

Discussion
In the present systematic review and meta-analysis, we investigated the usefulness of blood and CSF inflammatory biomarkers for VaD diagnosis. We found that, compared to healthy subjects, a moderate to large elevation of both blood IL-6 and TNF-α levels was associated with VaD diagnosis. However, only blood IL-6 concentrations significantly differed between VaD and AD subjects such that patients with VaD had small to moderate elevation of IL-6 compared to those with AD. Moreover, we found that each unit increase of IL-6 levels predicted 28% higher risk of VaD. In the CSF of VaD patients, IL-6 levels were significantly higher than in healthy subjects, but no difference was detected compared to AD patients. Data from CSF should be taken with caution due to high inconsistency related to the still limited number of studies with relatively small sample size. Present findings might suggest that among inflammatory markers, circulating IL-6 levels could be a useful biomarker able to differentiate across healthy, VaD, and AD subjects. A recent meta-analysis by Ng and colleagues showed that blood inflammatory markers, including IL-6, were not significantly different between AD patients and controls [39]. However, evidence from both cross-sectional and prospective studies highlight that higher IL-6 levels are related with poorer cognitive performance [40,41] and faster cognitive decline [42,43]. The relationship between cognitive impairment and inflammation in VCID is partly explained by the existence of a clear association between inflammatory status, atherosclerosis, and prothrombotic conditions [44]. Compared to previous meta-analytic findings that did not evidence any GeroScience significant association between circulating CRP and IL-6 levels and future risk of AD [12,13], we found a positive linear relationship between blood IL-6 and risk of incident VaD.
In the present meta-analysis, among CSF inflammatory biomarkers, only IL-6 had enough studies to be included in quantitative synthesis. However, other inflammatory markers in CSF are under investigation. For example, few reports showed that TNF-α levels in the CSF were higher than in sera among subjects with dementia, suggesting an intrathecal synthesis of this cytokine [28]. Assessment of the soluble forms of TNF-α receptors (sTNFR1 and 2) which may provide more accurate information about activation of the TNF-α system revealed that patients with mild cognitive impairment (MCI) who converted to VaD had higher concentrations of these biomarkers compared to those who converted in AD [45]. Biomarkers of microglial activation in CSF, which are related to neuroinflammation (i.e., YKL-40 and calcium binding protein B), were not able to differentiate between AD and VaD patients [46]. However, Olsson and colleagues showed that in subjects with MCI followed over 5.7 years, higher levels of YKL-40 and sCD14 in CSF predicted conversion to VaD but not to AD [47]. Further well-conducted studies are warranted to draw conclusion on reliability of inflammatory markers from CSF in VaD diagnosis.
Our findings might suggest that systemic inflammation contributes to VCID. Studies on brain biopsies showed controversial results on the contribution of inflammatory mechanisms in the pathogenesis of VaD [48][49][50]. It has been hypothesized that different types of cerebral small vessel disease (SVD) might be mechanistically linked to different forms of inflammation [51]. Cerebral SVD represents one of the most common neuropathological features of VCID [52]. It has been shown that biomarkers of systemic inflammation, like IL-6, may be associated with a specific form of SVD, the cerebral amyloid angiopathy (CAA) also known as type 2 SVD, which involves lobar regions and the centrum semiovale [51]. Conversely, sustained elevation over time of biomarkers of systemic inflammation is longitudinally associated with SVD progression [51].
Among different inflammatory biomarkers, we found a preeminent role of IL-6 in the diagnosis of VaD. Preclinical and clinical studies have demonstrated that during aging, in endothelial and smooth muscle cells, there occurs an overexpression of genes coding for inflammatory cytokines, chemokines, adhesion molecules, and other proinflammatory mediators, leading to the development of a proinflammatory microenvironment that promotes vascular dysfunction [53,54]. Moreover, higher inflammatory markers may underlie a damage of neurovascular unit [55]. Indeed, inflammatory and oxidative injuries may alter neurons and white matter function by interfering with neurovascular coupling [56]. This process exacerbates tissue hypoxia, by contrasting proliferation, migration, and differentiation of oligodendrocyte stem cells and by compromising mechanisms of reparation of damage in the white matter [57]. In addition, the activation of leukocytes and the release of inflammatory cytokines and cell adhesion molecules, which have been observed in patients with hypertension, may induce a dysregulation of the signaling of angiotensin II [58]. This dysregulation may lead to the impairment of the modulation of cerebral perfusion in response to blood pressure variations [59]. Specifically, IL-6 is involved in atherosclerosis through a large variety of pathways leading to plaque formation, from the stimulation of the acute-phase reactants and coagulation factors synthesis in the liver to the promotion of proliferation and differentiation of leukocytes and the activation of endothelial cells [60]. The latter respond to the IL-6 stimuli by releasing chemokines and increasing the expression of cellular adhesion molecules as the intercellular adhesion molecule 1 (ICAM-1), which is involved in the adhesion and transmigration of circulating leukocytes [61]. Promising perspectives come from other blood proinflammatory biomarkers as midregional proenkephalin A (MR-PENK A), mainly associated with pain sensation, cardiac function, and immunity, which has been positively associated with increased risk of VaD [62].
Despite data from randomized controlled trials are still scarce, targeting proinflammatory pathways may be a promising approach for the prevention of cardiovascular diseases and potentially VCID. Among eligible pharmacological strategies geared toward systemic inflammation, the inhibition of TNF-α signaling or the treatment with the IL-6 inhibitor tocilizumab determined an improvement of endothelial function assessed by means of flow-mediated dilatation [63,64] GeroScience infarction [65]. On the other hand, the administration of low-dose methotrexate did not result in fewer cardiovascular events compared to placebo [66]. Great interest has been aroused by the effect of a therapeutic monoclonal antibody targeting IL-1β, canakinumab, whose administration led, in a large cohort of patients with previous myocardial infarction, to a significantly lower rate of recurrent cardiovascular events [67]. Nevertheless, canakinumab was not approved for cardiovascular disease prevention, due to increased risk of fatal infections. Also, the statins, beyond their lipid-lowering effect, have well-characterized antiinflammatory properties including the inhibition of the formation of isoprenoids and proinflammatory mediators, and the subsequent reduction of asymmetrical dimethylarginine, implicated in endothelial dysfunction [68,69]. Noteworthy, several randomized controlled trials demonstrated that patients taking statins had a significant reduction of CRP levels [70,71], but evidence of a protective role of statins against VCID are still insufficient [72]. A few preclinical studies explored the effectiveness of other compounds with anti-inflammatory properties for VCID prevention. The angiotensin-(1-7) glycosylated mas receptor agonist demonstrated the ability to restore visual-spatial memory in a murine model of VCID [73]. Another molecule, the N-palmitoylethanolamide-oxazoline, reduced in mice the histological alterations typical of VCID and improved behavioral disorders through neuroprotective and anti-inflammatory activity [74]. Furthermore, treatment with resveratrol which has wellknown anti-inflammatory and antioxidant properties was associated, in a rodent model of VaD, to better vascular reactivity and reduction of cognitive decline [75]. Future preclinical and clinical studies should test if strategies targeting chronic inflammation and, in particular, blood IL-6 could have a role in reducing incidence of VCID or slow down its progression.
To the best of our knowledge, this is the first metaanalysis exploring the reliability of few well-accepted inflammatory biomarkers (IL-6, CRP, and TNF-α) for differential diagnosis between VaD and AD. This distinguishes our findings from those of other previous systematic reviews and meta-analyses that assessed only the association between inflammation and overall dementia or AD [12,13]. The present study has also some limitations. First, despite the strict inclusion/ exclusion criteria, there is wide heterogeneity observed across the studies, due to potential several reasons: (a) different measurement platforms, (b) small sample size per each study, (c) different case adjudication methods, (d) presence of subjects in different VCID stages in cross-sectional studies, or (e) different lengths of follow-up in longitudinal studies. Second, several studies had relatively small sample sizes that could potentially lead to overestimation of effects. Nevertheless, we performed sensitivity analysis excluding the studies at higher risk of publication bias, and we did not detect any significant change in the results. Third, for most of the studies, the assessment of inflammatory state was based only on a single value of the biomarker which could lead to a misclassification of exposure. Fourth, although we included only studies in which VaD and AD diagnosis were based on internationally validated criteria, misclassification of outcome should be accounted given the different methods used for diagnosis and the few studies including a confirmation of diagnosis by imaging techniques. In this regard, it is worthy of mention that whenever specified, subjects with mixed dementia were excluded. Fifth, the included studies adjusted the analysis for different factors; therefore, there could be unmeasured confounders associated with inflammation and dementia. Sixth, results on CSF biomarkers should be considered with caution due to limited number of included studies. Finally, the assays for biochemical measurements of serum or plasma IL-6, CRP, and TNF-α varied across the studies.