We searched the published literature in two electronic databases including MEDLINE, PubMed, Embase, and Web of Science and Scopus up to December 2021. Search terms were as follows:

Studies were included if they met the following criteria: (i) the study had a cross sectional or longitudinal prospective cohort design; (ii) the cross sectional studies should have reported data of AD, MCI with respect to controls, while in longitudinal studies, cognition performance was used at baseline and follow-up; in (iii) levels of cytokines were measured in blood; (iv) the longitudinal study measured the association of cytokine level and cognitive decline; (v) the article was available in English. Exclusion criteria included: (i) participants with dementia or cognitive impairment were included at baseline; (ii) the association between baseline cytokine level and cognitive decline was not reported; (iii) if the concentration of cytokine markers were measured in post-mortem samples. (iv) Small sample size less than 5 was used.

The quality assessment was performed by two authors independently according to the Newcastle–Ottawa Scale, and disputes were resolved by discussion. The quality score was calculated based on three major components of cohort studies: quality of selection (zero to four stars), comparability (zero to two stars), and exposure and outcome of study participants (zero to three stars). A higher score represents better methodological quality. Studies were defined as high (greater than seven stars), medium (six to seven stars), or low quality (less than six stars).

Each type of dataset was extracted independently by two authors. Discrepancies were reconciled through full discussion and consensus among the reviewers. For longitudinal studies, the extracted data involved the following: (I) summary and baseline of patients included in our study including study ID (name of the author, year, and setting of the publication), study design, subjects at baseline, the proportion of females at baseline, mean age at baseline (years), mean follow up assessment of global cognition in months, and the conclusion of each study; (II) risk of bias (ROB) domains including three major components of cohort studies: quality of selection (zero to four stars), comparability (zero to two stars), and exposure and outcome of study participants (zero to three stars); and (III) the outcome measures. The outcome measures were extracted as odds ratio (OR) and 95% confidence intervals (CIs) for the adjusted model, and confounders were adjusted for in the regression analysis. For cross sectional studies, the sample sizes and mean (±SD) concentrations of markers were extracted and Hedge’s g was used for effect size (ES) for meta-analysis.

Statistical analyses were performed using Open Meta Analyst (AHRQ, CEBM; Brown University, Providence, RI, USA) and STATA version 16.0 (StataCorp LLC, College Station, TX, USA). We ultimately employed the random-effects model with the DerSimonian and Liard method (DerSimonian and Laird, 1986). From cross sectional studies, the sample sizes and mean (±SD) concentrations of markers were extracted and Hedges’s g was used for ES for meta-analysis. For longitudinal studies, all data were dichotomous (events and no events) and were pooled as weighted proportions and risk ratios (RRs) with 95% CI (Cumpston et al., 2019). Pooled rates of proportions were calculated through the Freeman–Tukey transformation meta-analysis of proportions using MedCalc (Version 15.0; MedCalc Software, Ostend, Belgium).

Heterogeneity between studies was examined visually and statistically through Chi-square and I2 tests: a Q statistic with P < 0.1 indicated heterogeneity, whereas I2 values of 0, 25, 50, and 75% represented no, low, moderate, and high heterogeneity, respectively (Higgins et al., 2003). When detecting considerable heterogeneity, we performed sensitivity analyses to ascertain the source of heterogeneity by excluding one study at a time in addition to and subgroup analyses. Publication bias was visually examined through funnel plot symmetry as well as mathematically through Egger’s regression test, Begg’s test, and Duval’s non-parametric trim-and-fill analysis (Begg and Mazumdar, 1994; Irwig et al., 1998; Duval and Tweedie, 2000).

Our search extracted 25,513 unique citations after searching electronic databases. Following title and abstract screening, 619 full-text articles were retrieved and screened for eligibility. Of them, 540 articles were excluded, and 79 studies were reviewed in detail and included in this meta-analysis (PRISMA flow diagram; Figure 1).

The bibliography of the included randomized control trials (RCTs) was manually searched but added no further records. All studies were conducted between 2002 and 2021. Table 1 summarizes the characteristics of included patients and cross-sectional studies, while Table 2 represents longitudinal studies.

Following the Newcastle Ottawa Scale, the quality of the included studies ranged from moderate to high. The main concern was absent control groups. A summary of quality assessment domains with authors’ judgments is attached. Funnel plots of the inverse of the standard error vs. the ES demonstrated asymmetry. However, Egger’s test (P = 0.08) and Begg’s test (P = 0.13) indicated no small-study effects. Also, we employed the trim-and-fill approach to verify the robustness of the results, which exhibited no significant changes to the results when imputing three missing studies.

A total of 46 studies were included in this analysis. Cross-sectional studies have demonstrated an increased level of CRP (Hedges’s g 0.35, p < 0.05) (Figure 2), IL-1β (0.94, p < 0.05) (Figure 3), IL-6 (0.46, p < 0.005) (Figure 4), TNF alpha (0.22, p < 0.05) (Figure 5), and sTNFR-1 (0.74, p < 0.05) (Figure 6) in AD compared to controls. Similarly, higher levels of IL-1β (0.17, p < 0.05) (Figure 7), IL-6 (0.13, p < 0.005) (Figure 8), TNF alpha (0.28, p < 0.05) (Figure 9), and CRP (0.21, p < 0.05) (Figure 10) was also observed in MCI vs. control samples.

One hypothesis to explain the effects of increased cytokines in MCI and AD as compared to healthy controls is the impact of systemic inflammation on brain plasticity and function. Systemic inflammation raises pro-inflammatory cytokines such as IL-6, TNF-, and CRP, which may interact with the CNS in three ways: (1) pro-inflammatory cytokine transport proteins facilitate active trafficking across the blood brain barrier (BBB) for central activity (Dantzer et al., 2008; Fung et al., 2012). (2) Systemically generated cytokines may excite afferent nerves (e.g., the vagal nerve), which send inflammation to the brain stem. Vagal nerve projects to the solitary tract nucleus and higher brain areas (McCusker and Kelley, 2013). (3) Circulating cytokines reach outside-BBB organs. There, cells expressing toll-like receptors react to the increased inflammation by releasing pro-inflammatory cytokines, which may reach the brain by volume diffusion (Vitkovic et al., 2000b; McCusker and Kelley, 2013; Sankowski et al., 2015). When triggered peripherally, these three routes activate brain microglia and astrocytes to create pro-inflammatory cytokines, spreading the signal into the neuronal environment (Dantzer et al., 2008; Sankowski et al., 2015).

The blood-brain barrier, often known as the BBB, is an important component in both the preservation of the CNS’s highly specialized milieu and the facilitation of communication with the systemic compartment. Alterations to the BBB may be seen in a variety of CNS diseases, including AD. Some non-disruptive BBB alterations are mediated directly by cytokines (Ericsson et al., 1995). The cerebral endothelium expresses several cytokines, including IL-1, IL-6, and TNF- receptors. Activation of the endothelium is caused by IL-1 before that of neighboring brain regions, which suggests that activation of the BBB is an intermediary stage (Herkenham et al., 1998). It is interesting to note that IFN- decreases the transmigration of type 1 T helper lymphocytes without having any effect on the diffusibility of albumin. This suggests that the effect is not caused by a change in the tight junctions but rather by cytokine-induced non-disruptive changes that discourage neuroinflammation (Prat et al., 2005).

Systemic inflammation is also known to play an important role in altering the hormonal levels. There are a number of cytokines, including TNF-alpha, interleukin-1 beta (IL-1 beta), and IL-6 that work as feedback loops to inhibit the immunological response when the hypothalamic-pituitary-adrenal (HPA) axis is stimulated by the stress of trauma or exercise (Brenner et al., 1997; Licinio and Wong, 1997; Teblick et al., 2019). However, persistent increase of cytokines might potentially inhibit the HPA axis, which can lead to decreased levels of glucocorticoids, growth hormone, and adrenocorticotropic hormone (Teblick et al., 2019).

Preclinical and clinical research have demonstrated that infection-related peripheral inflammation contributes to the development and progression of CNS disorders such AD, personality disorders (PD), Multiple Sclerosis (MS), and stroke. Patients with AD have increased levels of Aβ in their brains due to peripheral inflammation (Le Page et al., 2018). In amyloid precursor protein (APP) transgenic mice, peripheral lipopolysaccharide (LPS) injection enhanced BBB permeability, enabling proinflammatory factors such as IL-6 and TNF- to infiltrate the brain and promote disease development (Jaeger et al., 2009; Takeda et al., 2013). The increased level of cytokines in included studies suggests that patients might have CNS inflammation.

Results from longitudinal studies also showed high levels of IL-6 significantly increased the risk of cognitive decline. However, intermediate levels of IL-6 had no significant effect on the final endpoint. Likewise, neither CRP, ACT, Albumin, or TNF alpha showed a significant impact on cognitive decline.

The neuronal and glial cell function is modulated by a highly controlled network of cytokines and soluble cytokine receptors (Benveniste, 1998; Vitkovic et al., 2000a). This is related to their capacity to modulate neurotransmission. Increases in noradrenergic, dopaminergic, and serotonergic metabolism in the hypothalamus, hippocampus, and nucleus accumbens can be caused by both systemic and central cytokine administration (Mohankumar et al., 1991; Shintani et al., 1993; Linthorst et al., 1995; Merali et al., 1997). IFN- stimulates neuronal differentiation, while IL-1, IL-6, and TNF- have trophic effects on developing neurons and glia (Jonakait, 1997; Zhao and Schwartz, 1998). In the developing brain, IL-1 may also play a role in regulating the synaptic plasticity that underpins learning and memory (Zhao and Schwartz, 1998).

Jordanova et al. (2007), Singh-Manoux et al. (2014) reported that raised IL-6 but not CRP predicted cognitive decline in this population. Inflammatory changes associated with cognitive decline may be specific to particular causal pathways which is consistent with our findings. Weaver et al. (2002), Sasayama et al. (2012), Adriaensen et al. (2014) found a strong association between the level of IL-6 and short-term identification or evaluation of global functional status in the old. They reported that elevated IL-6 and soluble IL-6R levels in Ala carriers may have a negative impact on acquiring verbal cognitive ability requiring long-term memory. Mooijaart et al. (2011) found that plasma CRP concentrations associate with cognitive performance in part through pathways independent of cardiovascular disease. However, lifelong exposure to higher CRP levels does not associate with poorer cognitive performance in old age.

On the other hand, Yaffe et al. (2003), Chen et al. (2014) documented an association between CRP elevation and cognitive impairment which is inconsistent with our meta results. This may be because of the limited number of participants in these studies compared to our meta-analysis. Ashraf-Ganjouei et al. (2020) found that healthy subjects with higher levels of CRP exhibit poorer performance in verbal learning memory and general wakefulness domains of cognition.

Some other interleukins have a role in cognitive impairment other than IL-6. Goldstein et al. (2015) recorded strong relation between IL-8 and cognitive performance in African Americans than Caucasians. However, this relationship should be further examined in larger samples that are followed over time. Conversely, Shi et al. (2020) reported IL-35 polymorphisms were not associated with cognitive decline in coronary heart disease patients over 2 years.

Shen et al. (2019) reported that inflammatory marker levels were found to be significantly different in AD and control groups, supporting the idea that AD is accompanied by inflammatory responses in the peripheral and cerebro spinal fluid (CSF). In a previous review, Wilson et al. (2002) reported that cytokine-mediated inflammation in neurodegenerative disorders such as AD and vascular dementia is increasingly appreciated. Cytokines are an important part of stress activation in the hypothalamic-hypo physiologic-adrenal axis.

Our study is the first meta-analysis to our knowledge to differentiate between several inflammatory cytokines and their relation individually to cognitive impairment. In spite of the large sample size of our study and the strength of meta-analysis, these results are limited by the study design of included studies and the insufficient data about other inflammatory markers other than IL-6. Only two studies reported the effects of ACT, Albumin, and TNF alpha on cognitive outcomes. We need more controlled studies that have the least confounders to assure the strength of results.

Although this meta-analysis only provides additional evidence for the association between some inflammatory markers and cognitive decline, it is worthwhile to discuss the potential impact for the development of novel treatments for AD. There has been an important debate on the current therapeutic target for AD, the amyloid β-protein. However, studies have not demonstrated that anti-amyloid therapies have induced significant therapeutic effects. In this context, some inflammatory markers, especially if targeted before the onset of cognitive deterioration, may be an interesting marker to target. For instance, it has been suggested that some diets, such as the Mediterranean-DASH Intervention for Neurodegenerative Delay (MIND) diets, can improve cognition and also have an anti-inflammatory effect (van den Brink et al., 2019). Also, non-invasive brain stimulation can also improve cognition (Martins et al., 2017) and reduce inflammation in animal models (Laste et al., 2012; de Oliveira et al., 2019; Toledo et al., 2021). One of the limitations of the current review is not including studies related to comorbid vascular dementia, which is connected with AD and MCI especially in elderly patients.