1

The association between C-reactive protein, Interleukin-6 and depression among older adults in

the community: a systematic review and meta-analysis

Running title: Inflammation and depression in older adults

Kimberley J. Smith, PhD 1†, Bonnie Au, MSc 2,3†, Lucie Ollis, MSc 1,

Norbert Schmitz, PhD 2,3,4, 5

† Joint first authors

1 Department of Psychological Sciences, School of Psychology, Faculty of Health and Medicine,

University of Surrey, Guildford, UK

2 Department of Epidemiology and Biostatistics, McGill University, Montreal, Quebec, Canada

3 Douglas Mental Health University Institute, Montreal, Quebec, Canada

4 Department of Psychiatry, McGill University, Montreal, Quebec, Canada

5 Montreal Diabetes Research Center, Montreal, Quebec, Canada

Address for correspondence:

Kimberley J Smith

Psychological Sciences

University of Surrey

Stag Hill Campus

Guildford.

GU2 7XH.

E-mail: Kimberley.j.smith@surrey.ac.uk

Wordcount: 8477

2

Abstract

Previous research indicates there may be an association between inflammation and depression in

older adults but results are inconsistent. Therefore, the aim of this review was to determine the cross-

sectional and longitudinal associations of two inflammatory markers C-reactive protein (CRP) and

Interleukin-6 (IL-6) with depression in older adults. We searched five databases for cross-sectional

and longitudinal studies reporting an association between CRP or IL-6 with depression among adults

sampled from the community aged 50 or older. We found 32 studies (23 cross-sectional, 7

longitudinal, and 2 assessing both cross-sectional and longitudinal associations) that met eligibility

criteria. These studies were entered into a random-effects meta-analysis to determine the cross-

sectional association and longitudinal direction of association between both IL-6 and CRP with

depression.

Results indicated a cross-sectional and longitudinal association between both CRP and IL-6 with

depression in older adults, with inflammation leading to depression in longitudinal studies rather than

depression to inflammation. However, there was notable heterogeneity between studies as results

differed based on adjusting for confounders and on how inflammation and depression were measured.

These sources of heterogeneity could explain differences in study results.

Keywords: C-reactive protein, Interleukin-6, depression, older adults, systematic review, meta-

analysis

3

1.0 Introduction

Estimates suggest that 1% to 12% of community-dwelling older adults are affected by major

depression (Copeland et al., 2004, Fiske et al., 2009, Hasin et al., 2005), while depressive symptoms

are present in 10% to 39% of community-dwelling older adults (Blazer, 2003, Djernes, 2006, García-

Peña et al., 2008). Depressive symptoms among older adults can adversely affect quality of life in

terms of functioning and well-being (Penninx et al., 1998) and also increase the risks for morbidity

and mortality (Lee et al., 2001, Fiske et al., 2009, Blazer and Hybels, 2005).

One biological theory that could explain depression in older adults is the inflammatory theory of

depression. This theory posits that increased levels of inflammation can cause depression (Anisman,

2009, Dantzer et al., 2008b, Slavich and Irwin, 2014, Alexopoulos and Morimoto, 2011). It is

possible that this biological theory of depression could be especially important for older adults as they

tend to have higher circulating levels of inflammation than younger adults (Krabbe et al., 2004,

Bruunsgaard et al., 2001, Chung et al., 2011, Chung et al., 2009), and they also have an increased risk

of chronic illness (Wolff et al., 2002) which is associated with increased inflammation.

Two indicators of inflammation commonly examined in relation to depression are the acute-phase

protein C-reactive protein (CRP) and the proinflammatory cytokine Interleukin-6 (IL-6). CRP is a

non-specific marker of inflammation, infection, and tissue damage (De Berardis et al., 2009) shown

to be linked with depression (Haapakoski et al., 2015, Howren et al., 2009, Valkanova et al., 2013).

IL-6 is a broad acting inflammatory cytokine that has a role in stimulating an immune response to

stressors such as fever (Hunter and Jones, 2015). IL-6 has been shown to be increased in the

bloodstream of older adults (Ershler and Keller, 2000, Hager et al., 1994), and has also been linked

with depression (Haapakoski et al., 2015, Howren et al., 2009).

4

There have been three previous meta-analyses that have investigated the association of depression

with both CRP and IL-6. The first review conducted by Howren et al. (2009) found a positive

association between both CRP and IL-6 with depression. However, the collection of studies were

restricted to cross-sectional designs and the meta-analysis consisted of all age groups with no

subgroup analysis based on age. A second meta-analysis published by Haapakoski et al. (2015)

found a moderate association between both CRP and IL-6 with major depression as diagnosed by

structured clinical interviews across twenty and thirty one studies, respectively. However, most

studies were conducted in clinical populations meaning results might not be applicable to a

community sample, and no subgroup analysis based on age was provided. A third meta-analysis used

only longitudinal studies and found a small association between raised CRP levels with subsequent

development of depressive symptoms and a trend towards significance for IL-6 (Valkanova et al.,

2013). However, this study was based on eight studies and only three studies assessed the association

between CRP and depression specifically among older adults, while only three studies examined IL-6

and depression in total for all age groups. Two additional meta-analyses investigated the association

of IL-6 with depression, and both found a significant positive association (Dowlati et al., 2010, Hiles

et al., 2012). However, neither conducted an age-specific subgroup analysis.

Previous work that has been undertaken in older adults examining the association between

inflammation and depression has found inconsistent results with some studies indicating an

association (Lu et al., 2013, Penninx et al., 2003b) and some indicating no association (Stewart et al.,

2008, Valentine et al., 2011). These differences make interpretation of the literature difficult.

Researchers have pointed to differences in study populations, gender, assessment of depression,

inflammatory marker investigated and adjustment for important confounders such as chronic

conditions or adiposity as being possible reasons for inconsistency in results (Almeida et al., 2007,

5

Au et al., 2015, Das, 2016, Hiles et al., 2015). A previous meta-analysis by Hiles et al (2012)

undertook a comprehensive assessment of sources of heterogeneity in studies that examined

depression and IL-6 in all age groups. They found that results varied by depression assessment,

population setting (i.e., inpatient, outpatient or community) and presence of chronic conditions (Hiles

et al., 2012). Thus, considering sources of heterogeneity across studies may be important for

uncovering why inconsistencies in results may exist. However, we lack work that has systematically

investigated potential sources of heterogeneity in older adults.

Given the paucity of synthesis that has explicitly examined the link between CRP and IL-6 with

depression in older adults, the main aim of this review was to examine the relationship between both

CRP and IL-6 with depression in community-based samples of adults aged 50 or older. Further, due

to the inconsistency in findings from previous work we undertook a comprehensive qualitative and

quantitative assessment of potential sources of heterogeneity that could explain differences in

findings.

2.0 Methods

2.1 Search Strategy

A systematic literature search for studies describing the relationship between CRP or IL-6 with

depression was undertaken between January 2017 and May 2017 (with an update undertaken in

September 2017). No language restriction was implemented, but only English papers were reviewed.

Furthermore, no restriction was placed on date of publication, type of analysis employed or length of

follow-up for longitudinal studies.

Three search themes, “C-reactive protein”, “Interleukin-6” and “depression” were combined using the

Boolean operator “and.” (see Appendix A) in five major databases: MEDLINE via PUBMED (United

States National Library of Medicine, Bethesda, MD, USA), EMBASE (Elsevier, Amsterdam,

6

Netherlands), PsycINFO (American Psychological Association, Washington, DC, USA), ISI Web of

Knowledge (Thomson Reuters, New York, NY, USA), and ProQuest (ProQuest, Ann Arbor, MI,

USA). . In addition, we manually searched the reference lists of all identified relevant published

primary studies and review articles.

2.2 Study Selection

Individual studies were considered for inclusion in the systematic review if they assessed the

relationship between either CRP and depression in a community sample of adults 50 or older or they

assessed the relationship between IL-6 and depression in a community sample of adults 50 or older.

Furthermore, the following inclusion criteria were specified: 1) the authors reported data from an

original, observational peer-reviewed journal article (i.e., not review articles, reports, letters, theses,

posters, published abstracts or comments); 2) the authors provided cross-sectional or longitudinal

associations (or both) between CRP and depression or IL-6 and depression; 3) depression was

measured with a validated scale or determined using a diagnostic framework or clinical interview

(e.g., ICD or DSM) 4) the study population was a cohort of non-institutionalized older adults (age ≥

50 years at baseline); 5) for prospective studies, the study accounted for/controlled for baseline levels

of their outcome. Only studies where participants were sampled from the community were included

because the epidemiological inferences are more easily applicable to the general population. Within

this definition, we included any study where participants were community-dwelling older adults who

have been sampled from primary care. Chronic disease populations (i.e., cancer, diabetes,

cardiovascular disease, uremia, renal disease, or respiratory disease) were excluded as were

intervention studies. For publications using the same study cohort, the higher quality publication was

included for final analysis. When both publications had the same quality rating, the paper that

analyzed the larger number of participants was included.

7

Two authors (L.O. and K.J.S.) firstly screened the titles and abstracts of identified studies (see Figure

1). Full-text articles were then screened by two authors (B.A and K.J.S). Discrepancies were resolved

by consensus, or, when necessary, by a third author (N.S.). Those studies that were eligible for

inclusion following full-text analysis were put forward for data extraction and quality assessment.

2.3 Data Extraction and Quality Assessment

Data extraction and quality assessment of eligible studies were performed independently by two

reviewers (B.A. and K.J.S.). The following information was extracted: study characteristics,

participants’ characteristics at baseline, method of assessment for depression, method of assessment

for CRP and IL-6, analysis strategy, and brief results (least adjusted and most adjusted effect measure

was reported if available). For the purpose of this review, the term ‘depression’ is used as a general

descriptor that incorporates any assessment of depression (i.e., depressive symptoms or clinical

depression). Clinical depression refers to studies where clinical diagnostic criteria were used to

determine depression, while depressive symptoms refer to studies that examined depression using a

symptom scale.

The quality of both cross-sectional and longitudinal studies was assessed using the modified

Newcastle-Ottawa Scale (NOS) for observational studies. The NOS is a comprehensive instrument

that has established content validity and inter-rater reliability (Wells et al., Gariepy et al., 2010). The

interpretation of the scale is based on a “star” system wherein a study is assessed on three broad

categories: the selection of study groups; the comparability of the study groups; and the measurement

of exposure/outcome. Only those studies determined to be of a minimum moderate quality were

selected for the meta-analysis (see Appendix B for quality assessments).

2.4 Statistical Analysis

We analyzed cross-sectional and longitudinal associations separately. Data were combined with

random-effects models as these provide more conservative estimates, which are appropriate for our

8

analyses given the potential methodological and analytical heterogeneity across studies. Where

results were stratified (e.g., by gender) we firstly combined these in order to enter one estimate for

each study. We computed correlation coefficients (r) with their corresponding 95% confidence

intervals (CI) and graphically presented these on Forest plots. The majority of the studies reported

both unadjusted and adjusted findings. Therefore, we separated these results into two groups to make

them more comparable: 1) “least adjusted r” (either unadjusted r or r adjusted for basic covariates age

and sex) and 2) “most adjusted r” (the value from the fully adjusted model). Heterogeneity was

examined by calculating the Cochrane Q statistic and the inconsistency index (I2) statistic. The

Cochrane Q statistic is distributed as 2 and tests whether the individual effects are farther away from

the common effect beyond what is expected by chance (a p-value < 0.10 indicates significant

heterogeneity). The I2 statistic is the percentage of total variability in effect measure that is

attributable to variation across studies (a value of 25%, 50%, and 75% indicates low, medium, and

high heterogeneity respectively). Publication bias was assessed visually using Funnel plot and

statistically with Egger’s regression intercept test. All statistical analyses were conducted using

Comprehensive Meta-Analysis Software (version 3.0 Biostat, Englewood, NJ, USA).

3.0 Results

3.1 Study Selection

A total of 7573 studies were identified in our initial search (see Figure 1). Studies were assessed

using broad screening criteria (e.g., suitability of study population, design, exposure or outcome) and

7428 studies were rejected including duplicates. Following this 145 potentially relevant publications

were put forward for full-text screening with 104 excluded. The primary reasons for rejection were

non-representative sample, below age requirement, or no measure of association between CRP and

depression. This left 41 potential studies for quality assessment, nine of which were excluded due to

9

study sample duplication (whichever publication that was of lower quality or that used a subset of

participants were excluded), inability to extract data and design unsuitability (see Figure 1). We were

left with thirty two studies for the systematic review and meta-analysis. Two studies contained both

cross-sectional and prospective analyses, twenty three studies provided cross-sectional data, and

seven studies provided prospective analyses (see Tables 1-3). Inter-rater agreement for studying

screening was good (kappa = 0.72 ± 0.06).

Table 1 summarizes the characteristics and main findings of the twenty five cross-sectional analyses,

Table 2 describes results of the eight longitudinal studies that examined the direction of association

from CRP to depression and/or IL-6 to depression and Table 3 describes results of four studies that

examined the direction of association from depression to CRP and/or depression to IL-6. The quality

assessment results from the cross-sectional and longitudinal studies are included in Appendix B.

Studies included in this review were published between 1999 and 2016.

There were ten studies from Europe, twelve from North America, five from Asia, and three from

Australia. Sample size ranged from 64 to 5,438 participants, with a total of 42,606 individuals from

cross-sectional analyses and 13,337 individuals from prospective analyses included in this review.

3.2 Study review and meta-analysis of cross-sectional association between CRP and depression

3.2.1 Association between CRP and depression: Main results

Twenty two studies conducted cross-sectional analyses on the association between CRP and

depression. There were 17 studies that examined the least-adjusted relationship between CRP and

depression and 17 studies that examined the most-adjusted relationship between CRP and depression.

Of those 17 studies that examined the least adjusted relationship between CRP and depressive

symptoms, seven of them found a significant positive association (Almeida et al., 2007, Hiles et al.,

2015, Kop et al., 2002, Lu et al., 2013, Nadrowski et al., 2016, Smagula et al., 2014, Song et al.,

10

2015). Of those 17 studies that examined the most-adjusted relationship, only three found a

significant positive association (Ma et al., 2011, Penninx et al., 2003a, Song et al., 2015).

The random effects summary correlation coefficient for those studies that reported least adjusted

estimates showed a positive association between CRP levels and depressive symptoms (Figure 2a).

The effect was small yet highly significant, r = 0.043 (95% CI: 0.031-0.056), p < 0.001. For the

studies that reported fully adjusted results (Figure 2b), the association was attenuated, r = 0.010 (95%

CI: -0.009-0.028), p = 0.308. The I2 was 0% and the Q statistic was 13.32 (p = 0.65) for the least

adjusted analysis and the I2 was 35.36% and the Q statistic was 24.75 (p = 0.07) for the most adjusted

analysis. This suggests statistical heterogeneity was low-moderate for these analyses. The funnel

plots (Appendix C) were generally symmetric, though there was an outlying study in the least

adjusted analysis (likely the extreme non-significant result found by Forti). Finally, Egger’s test

produced a non-significant P-value for the least adjusted (0.40) but not the most adjusted analyses

(0.04). This suggests there may have been publication bias for the most adjusted studies. 3.2.2.1

Heterogeneity: Qualitative synthesis

There was notable heterogeneity between studies in terms of depression assessment, analytic

strategies, population studied and confounders controlled for. Of the twenty two studies only twelve

provided both a least and most-adjusted estimate, the remaining ten studies either provided the least

or most adjusted estimate (see Table 1).

There was heterogeneity observed between studies in the measurement of depression. All twenty two

studies assessed depression using a validated depression rating scale, and three studies provided

additional estimates for the relationship of CRP with clinical depression. However, none of the three

studies that assessed the relationship between CRP and clinical depression reported a significant

relationship for either the least or most adjusted estimates (Bremmer et al., 2008, Forti et al., 2010,

11

Tiemeier et al., 2003b). There were also differences observed in how depressive symptoms were

modelled in the statistical analysis. Some studies modelled depressive symptoms continuously, others

modelled them categorically (see Table 1). Where depressive symptoms were modelled categorically

the cut-offs tended to be based on validated cut-offs, however this was not always consistent between

studies. For example, of the four studies that used the GDS-15 categorically one used a cut-off of ≥5

(Lu et al., 2013), one used ≥6 (Smagula et al., 2014) and two used ≥7 (Almeida et al., 2007,

Nadrowski et al., 2016).

Alongside heterogeneity in how depressive symptoms were modelled, there was also heterogeneity in

how CRP was modelled statistically with some studies examining this variable continuously and

others categorically. When modelled categorically there was notable heterogeneity between studies.

Some studies used clinically defined cut-offs (Almeida et al., 2007, Hamer and Chida, 2009), but

most used data-driven cut-offs such as the median or quartiles (Bremmer et al., 2008, Forti et al.,

2010, Lu et al., 2013, Nadrowski et al., 2016, Pan et al., 2008, Penninx et al., 2003a). This meant that

the range used to define high CRP across studies varied between ≥1.51 mg/L (Pan et al., 2008) and

>9.65g/dl (Lu et al., 2013).

Another source of heterogeneity that was observed between studies was the modelling of the

exposure and outcome. Seven of the included studies examined depression as the exposure and CRP

as the outcome (Kop et al., 2002, Loucks et al., 2006, Ma et al., 2011, Mezuk et al., 2016, Milaneschi

et al., 2009, Stewart et al., 2008, Tiemeier et al., 2003a), whereas the remaining studies examined

inflammation as the exposure and depression as the outcome The majority of studies that found

significance for the least adjusted (six out of seven studies) and most adjusted (two out of three

studies) analysis examined CRP as their exposure and depression as the outcome (see Table 1).

12

In terms of confounders controlled for heterogeneity was also evident, three studies controlled for no

confounders (Lu et al., 2013, Matsushima et al., 2015, Valentine et al., 2011) whereas others

controlled for a variety of factors known to influence depression and inflammation including

sociodemographics, lifestyle, chronic conditions, medication and adiposity (see Table 1). Those

studies that included both a least and most adjusted association tended to find that significant least-

adjusted results were attenuated after adjustment for confounders (Almeida et al., 2007, Kop et al.,

2002, Smagula et al., 2014). However, one study conducted in 2,242 American women became

significant after adjustment for confounders (Ma et al., 2011), and a second study conducted in 569

older Korean’s remained significant in men after adjustment for confounders (Song et al., 2015).

Another source of heterogeneity between studies was the health of the sample. For example, five

studies excluded participants with very high levels of CRP indicative of systemic inflammation (see

Appendix II). Of these five studies, only one conducted in 1,410 Australian older adults found a

significant association between CRP and depression (Hiles et al., 2015). Furthermore, seven studies

excluded people with different chronic conditions, meaning the population studied could have been

healthier than other populations included within the review (Baune et al., 2012, Kop et al., 2002, Lu

et al., 2013, Mezuk et al., 2016, Pan et al., 2008, Smagula et al., 2014, Stewart et al., 2008, Valentine

et al., 2011). However, three of these seven studies did find a significant result, so there is no

indication exclusion of people with chronic conditions systematically influenced results (see Table 1).

There were also sociodemographic differences between the samples, with some studies having a

minimum age of 50 for inclusion and others 70. However, closer examination of significant results

revealed no systematic differences in the association between CRP and depression based on age of

inclusion (see Table 1).

13

Some studies also provided gender stratified estimates or were only conducted in one gender. Of

those studies conducted in men all four of studies that investigated the least adjusted estimate found a

significant association (Almeida et al., 2007, Hiles et al., 2015, Smagula et al., 2014, Song et al.,

2015), however in women only one of three studies found a significant least adjusted association

(Hiles et al., 2015). Notably, after controlling for confounders most of these relationships were

reduced to non-significance.

3.2.2.2 Heterogeneity: Quantitative synthesis

To ascertain whether the observed heterogeneity had an impact on the least or most adjusted estimate,

we ran a series of sensitivity analyses (see Table 4). Our analyses indicated that most results remained

unchanged when stratifying by sociodemographics, statistical/methodological differences and

diagnostic cutpoints. However, in those cross-sectional analyses that examined CRP as the exposure

and depression as the outcome, the fully adjusted estimate was significant (r = 0.032 [95% CI: 0.008-

0.057], p = 0.011). This indicated that the while heterogeneity was indicated in study methodology,

participants, analytic strategies and confounders controlled for that that these differences did not

substantially impact overall meta-analytic estimates.

3.3 Study review and meta-analysis of cross-sectional association between IL-6 and depression

3.3.1 Association between IL-6 and depression: Main results

In total there were seventeen studies that examined the cross-sectional association of IL-6 with

depression, with fourteen examining the least-adjusted association and thirteen examining the most

adjusted association. In studies that examined the association between IL-6 and depression, seven out

of fourteen studies found at least one significant positive association for the least adjusted estimate

(Dentino et al., 1999, Forti et al., 2010, Lu et al., 2013, Nadrowski et al., 2016, Pan et al., 2008,

Smagula et al., 2014, Hiles et al., 2015), and six out of thirteen studies found a significant positive

14

association for the most adjusted estimate (Baune et al., 2012, Dentino et al., 1999, Forti et al., 2010,

Hiles et al., 2015, Nadrowski et al., 2016, Penninx et al., 2003a).

The random effect summary correlation coefficients for IL-6 and depression indicated a significant

association for both the least and most adjusted estimates. The correlation coefficients for the least

and most adjusted estimates were r = 0.050 (95% CI: 0.035-0.066), p <0.001 and r = 0.032 (95% CI:

0.010-0.054), p <0.005, respectively. For the least adjusted analysis the I2 was 3.0% and the Q

statistic was 13.44 (p = 0.41), and for the most adjusted analysis the I2 was 12.57% and the Q statistic

was 13.73 (p = 0.32). This suggests statistical heterogeneity was low for both analyses. The funnel

plots (Appendix C) were generally symmetric and Egger’s test produced a non-significant P-value for

the least adjusted (0.58) and most adjusted analyses (0.76) suggesting no publication bias.

3.3.2.1 Heterogeneity: Qualitative synthesis

There was heterogeneity between those studies that examined the association between IL-6 and

depression for depression assessment, statistical analysis, populations and confounding (see Table 1).

Heterogeneity was observed in the measurement of depression. Most of the included studies

examined depressive symptoms using the CES-D and GDS (see Table 1). However, an additional

four studies examined clinical depression. All fours studies that examined clinical depression and IL-

6 found a significant relationship (Bremmer et al., 2008, Dimopoulos et al., 2008, Forti et al., 2010,

Tiemeier et al., 2003a), which remained significant after controlling for confounders. Whereas results

for depressive symptoms and IL-6 were more variable (see Table 1).

When examining exposures and outcomes, five of the 17 studies modelled depression as the exposure

and IL-6 as the outcome (Dimopoulos et al., 2008, Friedman et al., 2007, Nadrowski et al., 2016,

Stewart et al., 2008, Tiemeier et al., 2003a) with the remaining studies modelling IL-6 as the

exposure and depression as the outcome (see Table 1). Of the five studies that modelled depression as

15

the exposure only one found a significant association (Nadrowski et al., 2016), the remaining

significant associations were in those studies modelling IL-6 as their exposure (see Table 1).

There was also heterogeneity in how studies modelled IL-6 (continuous or categorical) and

depression (continuous or categorical). Like CRP, there was notable heterogeneity in the cut-offs

used for categorising IL-6 with the range used to define high IL-6 ranging from ≥ 1.08 pg/ mL

(Smagula et al., 2014) to >8.67 pg/ mL (Bremmer et al., 2008). However, closer examination of

results indicated no systematic differences in significance of results based on how IL-6 or depression

was modelled statistically or differences in cut-offs used (see Table 1).

As with CRP there was notable heterogeneity in terms of confounders controlled for. There were four

studies that only provided unadjusted estimates (Dimopoulos et al., 2008, Lu et al., 2013, Matsushima

et al., 2015, Valentine et al., 2011), and one only controlled for sociodemographics (Dentino et al.,

1999). However, the majority of included studies controlled for a range of sociodemographic,

lifestyle, health and medication-related variables (see Table 1). Of the six studies that found a most

adjusted significant result only one did not adjust for medication, adiposity and/or chronic conditions

(Dentino et al., 1999). Furthermore, in those nine studies that provided both a least adjusted and most

adjusted estimate the addition of confounders into the statistical model did not change results for six

studies. In the remaining three studies statistically significant results were attenuated by the inclusion

of confounders (Hiles et al., 2015, Pan et al., 2008, Smagula et al., 2014). However, in Hiles et al

(2015) the inclusion of confounders into the statistical model meant that the previously non-

significant least adjusted result for men became borderline significant in the most adjusted result.

When compared on the health of participants there was also heterogeneity between studies. Two

studies excluded participants on the basis of systemic inflammation (Hiles et al., 2015, Stewart et al.,

2008), however as one study found significance and the other did not there is no evidence this would

16

affect estimates (see Table 1). There were seven studies that made exclusions to their populations on

the basis of having chronic conditions (Baune et al., 2012, Dimopoulos et al., 2008, Lu et al., 2013,

Pan et al., 2008, Smagula et al., 2014, Stewart et al., 2008, Valentine et al., 2011). However, as four

of these studies found at least one significant result there is no indication that exclusion of chronic

conditions systematically affected results (see Table 1).

There was also some heterogeneity between studies by age of inclusion, which ranged from 50 to 70

(see Table 1). Of note, five of the six studies that found significance in the most adjusted result had a

minimum age of 65 (Baune et al., 2012, Dentino et al., 1999, Forti et al., 2010, Hiles et al., 2015,

Nadrowski et al., 2016, Penninx et al., 2003a). In terms of gender only one study provided gender

stratified estimates (Hiles et al., 2015) and found a significant association between IL-6 and

depression for the least adjusted analysis in women but not men. However, in the most adjusted

analysis the significant association in women was attenuated and the association for men became

significant (Hiles et al., 2015). One additional study explored the association of IL-6 and depression

in men only but found no significant result (Smagula et al., 2014).

3.3.2.2 Heterogeneity: Quantitative synthesis

To determine whether any of these sources of heterogeneity could impact the estimates we ran a

series of stratified sensitivity analyses (see Table 4). When stratifying for various

statistical/methodological, diagnostic and demographic factors the least adjusted estimate remained

significant except when we only examined those studies that excluded people with chronic conditions

at baseline. The most adjusted estimate was no longer significant when we only included those

studies where they adjusted for chronic conditions or removed people with chronic conditions from

their analysis. It was also non-significant when we only examined those studies that did not exclude

people with systemic inflammation. In studies where depression was examined continuously and in

17

studies where IL-6 was assessed either categorically or continuously, the results become non-

significant after adjustment for confounders (see Table 4). This indicates the relationship between IL-

6 and depression may be influenced by systemic inflammation and chronic conditions.

Finally we performed an analysis to examine the association of clinical depression with IL-6. There

were four studies (Bremmer et al., 2008, Dimopoulos et al., 2008, Forti et al., 2010, Tiemeier et al.,

2003a) that examined this association and the correlation coefficient from the random effects meta-

analysis was significant for both the least and most adjusted estimate: (r = 0.115 [95% CI: 0.072-

0.158], p = <0.001) and ( r = 0.035 [95% CI: 0.035- 0.174], p = 0.003), respectively.

3.4 Study review and meta-analysis of the prospective association between CRP and depression

3.4.1 Prospective association between CRP and depression

A total of eight studies examined the direction of association from CRP to depression (see Table 2),

and four examined the direction of association from depression to CRP (see Table 3). Of these

studies, three examined the bi-directional association between CRP and depression (Au et al., 2015,

Das, 2016, Stewart et al., 2009).

3.4.1.1 CRP to depression: Main results

There were six studies that examined the least adjusted association between CRP to depression. Of

these studies, four found a significant association between CRP at baseline with depression at follow-

up (Au et al., 2015, Luukinen et al., 2010, van den Biggelaar et al., 2007, Zalli et al., 2016). For the

most adjusted analysis, five studies found a significant association between CRP at baseline with

depression at follow-up (Das, 2016, Luukinen et al., 2010, Stewart et al., 2009, van den Biggelaar et

al., 2007, Zalli et al., 2016) (see Table 2).

When the results for the least adjusted estimate were combined using a random effects meta-analysis,

a positive correlation between CRP at baseline and depressive symptoms at follow-up was found

18

(Figure 4a). The effect was small but significant, r = 0.074 (95% CI: 0.006-0.142), p= 0.033. For the

seven studies that reported fully adjusted results (Figure 4b), the strength of the correlation increased

slightly and remained statistically significant, r = 0.074 (95% CI: 0.031-0.116), p =0.001.

The I2 for the least and most adjusted analyses were 64.18% and 35.24% respectively. The Q statistic

for the least adjusted analysis was 16.75 (p = 0.01) and the most adjusted analysis was 10.81 (p =

0.15). This indicated moderate-high and significant heterogeneity for the least adjusted analysis and

moderate and non-significant heterogeneity for the most adjusted analysis. The funnel plot for the

least adjusted analyses (Appendix C) were fairly symmetric and Egger’s test produced a non-

significant P-value (0.9), which represents an absence of publication bias. However, the funnel plot

for the most adjusted analysis and Egger’s test (P-value 0.008) were suggestive of borderline positive

publication bias.

3.4.1.2 Depression to CRP: Main results

Of the four studies that examined the relationship between depression at baseline and CRP at follow-

up, none found significant results for either the least or most adjusted analyses (Au et al., 2015, Das,

2016, Luciano et al., 2012, Stewart et al., 2009). Two studies provided a least adjusted estimate (Au

et al., 2015, Luciano et al., 2012) and four provided a most adjusted estimate (Au et al., 2015, Das,

2016, Luciano et al., 2012, Stewart et al., 2009). When we combined the least adjusted and most

adjusted analyses there was no correlation between depression at baseline and CRP at follow-up: least

adjusted r = 0.047 (95% CI: -0.007-0.100, p=0.09); most adjusted r = 0.005 (95% CI: -0.028-0.037, p

= 0.772). Results from both the least (I2=0.0%, Q statistic=0.64; p=0.43) and most (I2=0.0%, Q

statistic=2.372; p=0.50) adjusted analyses indicated low heterogeneity. The results from funnel plots

and Eggers test for the most adjusted analysis (see Appendix C) indicated no publication bias.

3.4.2 Heterogeneity

19

For studies that examined both directions of association, there were differences in methodology,

analysis, length of follow-up, confounders controlled for and population. However, these sources of

heterogeneity did not seem to affect the depression to CRP relationship as no study found

significance (see Table 3).

3.4.2.1 CRP to depression: Qualitative heterogeneity synthesis

All studies assessed depressive symptoms using rating scales, except for one study that also examined

clinical depression (Forti et al., 2010). However, this study did not find any association between CRP

at baseline with the incidence of major depression over 4 years (Forti et al., 2010).

There was heterogeneity in the measurement of depressive symptoms, scales utilised included the 8

and 20-item CES-D, 15 and 30-item GDS, Zung depression rating scale and Beck Depression

Inventory (see Table 2). Six studies examined depressive symptoms categorically, with most using

validated cut-offs (see Table 2), however the GDS-15 cut-off used by Van den Biggelaar et al (2007)

was lower than other studies ≥ 2. Notably this study was the most significant study in both the least

and most adjusted analysis, possibly because the cut-off used was lower (and thus would have

included more depression cases) than other studies.

There was also heterogeneity in terms of confounders controlled for. Of the included studies, five

adjusted for baseline depression (Au et al., 2015, Das, 2016, Stewart et al., 2009, van den Biggelaar et

al., 2007, Zalli et al., 2016), and three investigated incident depression (Forti et al., 2010, Hiles et al.,

2015, Luukinen et al., 2010). All studies that adjusted for baseline depression found a significant

result for either their least or most adjusted analysis (see Table 2).

There was some heterogeneity also observed for measurement of CRP. Most studies examined CRP

continuously (see Table 2), however three studies employed cut-offs for CRP. Of those studies, two

20

used the same cut-off of ≥3 mg/dl (Au et al., 2015, Luukinen et al., 2010) and the other used quartiles

with a similar cut-off of for the third quartile of ≥3.3 mg/dl (Forti et al., 2010).

There was also heterogeneity between studies for the length of follow-up for studies which varied

from 2.5 years (Luukinen et al., 2010) to 6 years (Stewart et al., 2009). However, there were no

notable differences in results between studies based on length of follow-up (see Table 2). There was

also heterogeneity evident in the population studied with ages of inclusion ranging from 50 to 85.

However, there was no notable difference in study results based upon the age of participants (see

Table 2).

There was some heterogeneity between studies for the health of the sample. Five studies also

excluded people with evidence of systemic inflammation from their analysis (see Appendix II), and a

further study examined this in a sensitivity analysis (Au et al., 2015). However, there were no notable

differences in significance between those studies that did versus did not exclude people for systemic

inflammation.

There was some heterogeneity in terms of confounder control. One study did not control for

confounders for CRP to depression (Hiles et al., 2015), while all other studies provided adjusted

estimates where they controlled for important confounders such as sociodemographics, lifestyle,

medication, adiposity and chronic conditions (see Table 2). There were more studies that found

significance in adjusted studies than non-adjusted studies (see Table 2).

3.4.2.2 CRP to depression: Quantitative heterogeneity synthesisTo ascertain if these sources of

heterogeneity impacted on the study estimates we ran a series of sensitivity analyses (see Table 5).

For the least-adjusted estimate, studies that investigated incidence of depression, used CRP as a

categorical predictor, examined people aged 60 or older, or stratified by gender or systemic

inflammation were all non-significant (see Table 5). For the most-adjusted analysis, there was no

21

significant association for studies that investigated incidence of depression (rather than controlled for

depressive symptoms at baseline), studies that had examined depressive symptoms continuously or

examined CRP categorically (see Table 5). These results indicate that studies that control for baseline

depressive symptoms rather than examine incident depression are more likely to find a significant

association, as are studies that examine CRP continuously.

3.4.2.3 Depression to CRP: Qualitative heterogeneity synthesis

There was less observable heterogeneity for studies that examined the direction of association from

depression to CRP with all four studies either being set in the UK or USA (see Table 3). For the

depression to inflammation analyses, three studies adjusted for baseline inflammation (Au et al.,

2015, Das, 2016, Stewart et al., 2009), with the fourth study assessing change in inflammation as

their outcome (Luciano et al., 2012). The range in follow-up for studies ranged from 3 years

(Luciano et al., 2012) to 6 years (Stewart et al., 2009). Only one study examined their exposure and

outcome categorically (Au et al., 2015) with all other studies modelling exposures and outcomes

continuously. All studies controlled for important confounders, with two studies also providing least

adjusted estimates (Au et al., 2015, Luciano et al., 2012). Three of the four studies had a baseline

study age of less than 60 (Au et al., 2015, Das, 2016, Stewart et al., 2009). All studies that examined

depressive symptoms to CRP found no significant relationship for either the least or most adjusted

estimate (see Table 3). As no statistical difference in results was uncovered and heterogeneity was

low, we did not perform further sensitivity analyses.

3.4.3 Direction of causality

Of the three studies that examined the bi-directional relationship between depressive symptoms and

CRP, two found a significant most-adjusted result from CRP to depression, but not vice versa (Das,

2016, Stewart et al., 2009). The third study also indicated that CRP led to depression rather than

22

depression to CRP in their least adjusted analysis, however the effect was attenuated in their most

adjusted analysis (Au et al., 2015). These results taken in tandem with those discussed above indicate

that CRP may lead to depression in older adults, rather than depression leading to CRP.

3.5 Study review and meta-analysis of the prospective association between IL-6 and depression

3.5.1 Prospective association between IL-6 and depression

In total five studies examined the prospective relationship between IL-6 at baseline to depression at

follow-up, and one study examined the prospective relationship between depression at baseline with

IL-6 at follow-up.

3.5.1.1 IL-6 to depression: Main results

Of the four studies that investigated the least adjusted association between IL-6 at baseline to

depression at follow-up, two found significant associations (Hiles et al., 2015, Zalli et al., 2016). For

the five studies that provided a most-adjusted estimate, only one found a significant association (Zalli

et al., 2016).The random effects summary correlation coefficient indicated a small positive significant

association between IL-6 at baseline and depressive symptoms at follow-up (Figure 5a) for the least

adjusted estimate, r = 0.085 (95% CI: 0.039-0.131), p< 0.001. However, the significant effect was

attenuated in the most adjusted estimate (Figure 5b), r = 0.034 (95% CI: -0.007-0.074), p= 0.103.

There was low heterogeneity indicated from the least (I2=6.01%, Q statistic=4.26; p=0.37) and most

(I2=14.66%, Q statistic=5.86; p=0.32) adjusted analyses. Publication bias was indicated by a

significant Egger test (p=0.05) for the least adjusted analysis (for funnel plots see Appendix C) but

not the most adjusted analysis (Egger test, p=0.16).

3.5.1.2 Depression to IL-6: Main results

23

Only one study investigated the direction of association from depression to IL-6 (Stewart et al.,

2009). This study found that depressive symptoms as measured with the BDI-II were a predictor of 6-

year change in IL-6, even after adjustment for demographic, biomedical, and behavioral factors.

Since only one study investigated this association in a relatively healthy population, limited

inferences can be made.

3.5.2.1IL-6 to depression: Qualitative heterogeneity synthesis

As only one study examined depression to IL-6 the focus of this section is on the five studies that

examined IL-6 to depression. As with previous studies, there was heterogeneity in terms of

methodology, analysis, length of follow-up and confounders controlled for.

There was some heterogeneity evident for the measurement and modelling of depression. All

included studies investigated depressive symptoms using validated scales, except Forti et al (2010)

who also included an estimate for high IL-6 leading to incident clinical depression. This study found

no evidence that the highest two quartiles of IL-6 led to incident clinical depression or high

depressive symptoms. Of those studies that investigated depressive symptoms, two used the CES-D,

two used the GDS and one used the Beck Depression Inventory (see Table 3). Four of the five studies

categorised depressive symptoms into high versus low, with the final study analyzing depressive

symptoms continuously (Stewart et al., 2009). Therefore, the two studies that found a significant

result had categorised depressive symptoms (Hiles et al., 2015, Zalli et al., 2016). There was also

some heterogeneity for the measurement and modelling of IL-6. When examining how IL-6 was

analyzed only one study examined cut-offs for IL-6 (Forti et al., 2010) with all other studies

examining IL-6 continuously. Both studies that found a significant result analyzed IL-6 continuously

(Hiles et al., 2015, Zalli et al., 2016).

24

There was minimal heterogeneity observed in terms of length of follow-up. The range in follow-up

for studies ranged from 3.5 (Hiles et al., 2015) to 6 years (Stewart et al., 2009) and there were no

systematic differences between results based on length of follow-up (see Table 2).

There was minimal heterogeneity in terms of confounders controlled for, as all studies provided a

most-adjusted estimate that controlled for important sociodemographic, lifestyle, chronic conditions

and adiposity factors (see Table 1). However, there was some heterogeneity in terms of how

depressive symptoms were controlled for at baseline. Two of the studies examined incident

depression (Forti et al., 2010, Hiles et al., 2015), and the other three studies controlled for depressive

symptoms at baseline (Stewart et al., 2009, van den Biggelaar et al., 2007, Zalli et al., 2016).

However, the observed differences did not appear to influence statistical results (see Table 2).

There was also heterogeneity in terms of the baseline health of the samples. Three studies controlled

for systemic inflammation (Hiles et al., 2015, Stewart et al., 2009, van den Biggelaar et al., 2007) and

one removed people with some chronic conditions from their analysis (Stewart et al., 2009).

However, there were no systematic differences in statistical significance based on these study

differences.

There were also some differences in the population studied. One study provided gender stratified

analyses which indicated a significant least adjusted association in women but not men (Hiles et al.,

2015), however gender inferences are difficult to make as no other study stratified by this variable.

Furthermore baseline study age ranged from 50 to 85 years old. However, no systematic differences

in results were evident on the basis of age (see Table 2).

3.5.2.2 IL-6 to depression: Quantitative heterogeneity synthesis

We conducted a series of sensitivity analyses to determine whether these differences between studies

could affect our meta-analysis (see Table 5). The least-adjusted estimate was reduced to non-

25

significance in studies that controlled for depression at baseline and in studies where people with

systemic inflammation were removed from the analysis. All estimates for the most-adjusted analysis

remained non-significant. Due to the small number of studies in these sensitivity analyses, results

should be interpreted with caution.

4.0 Discussion

We identified a total of 32 observational studies (23 cross-sectional, 7 longitudinal, and 2 assessing

both cross-sectional and longitudinal associations) that examined the association between CRP and

depression or IL-6 and depression among older adults from the community. Evidence from the least

adjusted cross-sectional analyses indicated a weak but positive association between CRP and

depression, which was attenuated after adjustment for confounding. There was also a weak positive

significant association between IL-6 and depression, which remained significant after adjustment for

confounding. There was no association between CRP and clinical depression, though there was a

significant association between IL-6 and clinical depression even after adjustment for confounders.

Evidence from prospective analyses indicated that both CRP and IL-6 levels at baseline have a small

but significant association with subsequent depressive symptoms, however this result only remained

significant for CRP to depression after adjustment for confounding. There was no evidence for the

direction of association from depression to CRP, and only one study investigated depression to IL-6

thus inferences were difficult to draw.

The results also demonstrated significant heterogeneity between studies in modelling of depression

and inflammation, confounders controlled for, analytic strategies and populations studied. Many

sources of heterogeneity affected our estimates in sensitivity analyses. For example, s, whether the

health of the sample was accounted for. Furthermore, for prospective analyses the analysis of the

exposure as ‘incident’ from baseline, or after adjustment for baseline levels of the exposure was an

26

additional source of heterogeneity. We suggest sources of heterogeneity should be considered in

future work that examine the association of inflammation and depression in older adults. For

example, if depression and inflammation are analyzed categorically it could be worth undertaking a

sensitivity analysis where they are analyzed continuously to determine the impact this has on results.

Previous work has proposed that inflammatory depression should be considered an important subtype

of depression in older adults (Gallagher et al., 2016, Alexopoulos and Morimoto, 2011). This subtype

of depression has been associated with a higher risk of developing diabetes (Au et al., 2014), a greater

chronicity of depression (Gallagher et al., 2016) and is proposed to be involved in the pathogenesis of

dementia (Leonard, 2007). Thus, inflammatory depression in older adults is a serious issue. However,

based on our work it is challenging to conclude the extent to which depression and inflammation are

linked in older adults. The small effect size between both CRP and IL-6 with depressive symptoms

and clinical depression suggests that the involvement of inflammation in depression as a condition is

small.

The observed small effect size could be partly due to the fact we only included community-based

samples. The cross-sectional meta-analysis conducted by Howren et al. (Howren et al., 2009)

observed a much smaller association in community-based samples compared to clinical samples.

Thus it is possible that the association of inflammation with depression is more pronounced in clinical

samples, something we did not investigate in this study.

Our most-adjusted results indicated that IL-6 was linked with both clinical depression and depressive

symptoms cross-sectionally but not longitudinally, whereas CRP was not linked with clinical

depression or depressive symptoms cross-sectionally but were associated with an increased risk of

depressive symptoms longitudinally. These heterogenous findings suggest that different inflammatory

markers could have different associations with depression in older adults, with IL-6 being an

27

indicator of current depressive symptoms and CRP being involved in the etiology of depression in

older adults. However, these inferences are limited by findings from our sub-group analysis and the

difficulty of inferring causality from observational studies. Furthermore, our work indicated there is

more need for research to be done for the direction of association from depression to IL-6, as there is

currently only one study that examines this association (Stewart et al., 2009).

Our results also point to the importance of confounders such as chronic conditions, adiposity and

lifestyle in explaining the cross-sectional association between depression and inflammation in older

adults. The estimate between CRP and depression was attenuated in the most-adjusted estimate, and

the most-adjusted IL-6 and depression estimate was affected in subgroup analyses that accounted for

chronic conditions. Previous research proposes that the association between inflammation and

depression is explained by cardiometabolic abnormalities (Au et al., 2015) and chronic conditions

(Almeida et al., 2007). Furthermore, meta-analyses suggest the relationship between IL-6 and

depression may be attenuated by adjusted for lifestyle and adiposity (Dowlati et al., 2010, Hiles et al.,

2012). However, there are also other variables that could explain this relationship such as frailty

(Soysal et al., 2016, Soysal et al., 2017), physical activity (Schuch et al., 2017, Vancampfort et al.,

2017), physical disability (Cole and Dendukuri, 2003, Reuben et al., 2002), bereavement (Cole and

Dendukuri, 2003, Schultze-Florey et al., 2012) and social isolation (Cacioppo et al., 2010, Shankar et

al., 2011) among many others. There is also evidence that rather than acting as confounders in the

depression-inflammation relationship that some of these variables should instead be modelled as

effect modifiers. Previous research has indicated that BMI and metabolic abnormalities are important

variables that modify the association of depression with inflammation. For example, work from a

Chinese study of adults aged 45 or older shows that the association between depression and CRP was

only significant in those older people who were underweight (Qin et al., 2017). Furthermore, research

28

that stratified by depression and diabetes status showed that those people with both diabetes and

depression had much higher levels of inflammation than those with depression or diabetes alone

(Doyle et al., 2013).

We suggest that consideration of the variables that may confound or modify the relationship between

inflammation and depression in older adults should be carefully investigated in future work so we can

better understand what may be driving these associations.

Another possible effect modifier to consider is the heterogeneity of depression itself, and the fact

inflammation may be linked with particular subtypes of depression (Penninx et al., 2013a) or

particular symptoms of depression (Gallagher et al., 2016). There is some evidence that inflammation

may be associated with the atypical subtype of depression (Lamers et al., 2013, Rudolf et al., 2014)

which is characterized by overeating, weight gain and hypersomnia. Furthermore, there is evidence

that inflammation may be linked primarily with somatic symptoms linked with depression (Duivis et

al., 2013) such as hypersomnia (Irwin et al., 2016), adiposity (Shelton and Miller, 2010) or

psychomotor slowing (Brydon et al., 2008). This could also explain why there was evidence for an

association of depressive symptoms rather than clinical depression with CRP (because of high scores

on somatic depression symptoms that are not necessarily indicative of a clinical depression).

The role of inflammation in depression has long been of interest to researchers, and previous work

has theorised a bi-directional relationship between inflammation with depression (Kiecolt-Glaser et

al., 2015). Our work suggests that in older adults it may be inflammation that leads to depression,

supporting results from a previous meta-analysis of longitudinal studies (Valkanova et al., 2013).

There are various biological explanations posited for this direction of association. Evidence shows

pro-inflammatory cytokines can cross the blood-brain barrier and act directly on emotion-regulating

brain structures involved in depression such as the amygdala (Penninx et al., 2013b, Krishnadas and

29

Cavanagh, 2012). Furthermore, administration of pro-inflammatory cytokines have also been shown

to induce sickness behaviour, a syndrome characterized by vegetative, somatic, and psychological

symptoms that are similar to depression (Dantzer et al., 2008a).

As older adults have higher levels of circulating cytokines and inflammation (Bruunsgaard et al.,

2001, Chung et al., 2011, Chung et al., 2009, Krabbe et al., 2004), it is plausible that increased

inflammation may be a particularly important risk factor for the development of depression in this

population.

As with all reviews there are limitations that should be borne in mind when interpreting results. The

first is the notable between-study heterogeneity observed for both those studies that conducted cross

sectional and prospective analyses.

There is evidence that additional inflammatory markers such as tumor necrosis factor-α, IL-1, IL-8

and IL-10 might also be involved in depression (Köhler et al., 2017). Thus, it will be important for

future work to examine different inflammatory markers.

An additional limitation is that our longitudinal analyses are based on a relatively small number of

prospective studies (especially for the depression to inflammation direction) which may explain why

some sensitivity analyses were reduced to non-significance. Further, most studies in the review

assessed depressive symptoms rather than assessing clinical depression. Although the rating scales

included have well-established psychometric properties and are commonly used in epidemiological

studies they are not clinical diagnoses of depression. It is likely that inclusion criteria for the review

(i.e., exclusion of clinical studies) precluded many studies that may have used diagnostic interviews

for depression. However, our results indicate that it may be important to examine the association of

clinical depression with inflammation and not just depressive symptomatology.

30

There was also some evidence of publication bias though this was minimal. Similarly, language bias

may also be an issue. Although we did not restrict our search by language, only English publications

could be reviewed, which may account for the reason that the majority of our studies were conducted

in Western countries.

Despite our study limitations, our review includes several strengths. We conducted a comprehensive

search from several databases. By including cross-sectional as well as prospective analyses, we were

able to examine not only cross-sectional associations between CRP and IL-6 with depression, but we

were also able to assess directionality via the longitudinal analyses. While the heterogeneity of

studies included could be seen as a limitation we conducted a thorough analysis of these differences

to uncover how they might affect results. Furthermore, by restricting our sample to older adults, we

can more clearly make inferences about how inflammation and depression are linked in older adults.

5.0 Conclusion

Overall, findings from this review suggest that there is a complex cross-sectional and prospective

relationship between CRP and IL-6 levels with depression which has been further complicated by

heterogeneity between studies. We found evidence to indicate a weak but positive correlation

between CRP and depression that was attenuated for the most-adjusted estimate. We also found

evidence for a cross-sectional association between IL-6 and depression that remained significant after

adjustment. However, these results were impacted by sensitivity analyses for chronic conditions.

Prospective analyses indicated a small positive correlation between both CRP and IL-6 levels at

baseline with depressive symptoms at follow-up, which remained significant for CRP after adjusting

for confounders. There was little evidence for the opposite direction of causality, though more

research for this direction is needed especially for IL-6. Results suggest that inflammation as

indicated by CRP and IL-6 are associated with depression in older adults, and it is likely that

31

inflammation leads to depression rather than depression to inflammation. The heterogeneity between

studies and results from sensitivity analyses also highlight the need for future studies to assess the

role of potential confounders and moderators in the inflammation-depression relationship and

carefully consider how variables will be analyzed for this relationship.

References

ALEXOPOULOS, G. S. & MORIMOTO, S. S. 2011. The inflammation hypothesis in geriatric depression. International journal of geriatric psychiatry, 26, 1109-1118.

ALMEIDA, O. P., NORMAN, P., HANKEY, G. J., JAMROZIK, K. & FLICKER, L. 2007. The association between C-reactive protein concentration and depression in later life is due to poor physical health: results from the Health in Men Study (HIMS). Psychol Med, 37, 1775-86.

ANISMAN, H. 2009. Cascading effects of stressors and inflammatory immune system activation: implications for major depressive disorder. Journal of psychiatry & neuroscience: JPN, 34, 4.

AU, B., SMITH, K. J., GARIEPY, G. & SCHMITZ, N. 2015. The longitudinal associations between C-reactive protein and depressive symptoms: evidence from the English Longitudinal Study of Ageing (ELSA). Int J Geriatr Psychiatry, 30, 976-84.

AU, B., SMITH, K. J., GARIÉPY, G. & SCHMITZ, N. 2014. C-reactive protein, depressive symptoms, and risk of diabetes: results from the English Longitudinal Study of Ageing (ELSA). Journal of psychosomatic research, 77, 180-186.

BAUNE, B. T., SMITH, E., REPPERMUND, S., AIR, T., SAMARAS, K., LUX, O., BRODATY, H., SACHDEV, P. & TROLLOR, J. N. 2012. Inflammatory biomarkers predict depressive, but not anxiety symptoms during aging: the prospective Sydney Memory and Aging Study. Psychoneuroendocrinology, 37, 1521-30.

BLAZER, D. G. 2003. Depression in late life: review and commentary. The Journals of Gerontology Series A: Biological Sciences and Medical Sciences, 58, M249-M265.

BLAZER, D. G., 2ND & HYBELS, C. F. 2005. Origins of depression in later life. Psychol Med, 35, 1241-52.BREMMER, M. A., BEEKMAN, A. T., DEEG, D. J., PENNINX, B. W., DIK, M. G., HACK, C. E. & HOOGENDIJK, W. J.

2008. Inflammatory markers in late-life depression: results from a population-based study. J Affect Disord, 106, 249-55.

BRUUNSGAARD, H., PEDERSEN, M. & PEDERSEN, B. K. 2001. Aging and proinflammatory cytokines. Current opinion in hematology, 8, 131-136.

BRYDON, L., HARRISON, N. A., WALKER, C., STEPTOE, A. & CRITCHLEY, H. D. 2008. Peripheral inflammation is associated with altered substantia nigra activity and psychomotor slowing in humans. Biological psychiatry, 63, 1022-1029.

CACIOPPO, J. T., HAWKLEY, L. C. & THISTED, R. A. 2010. Perceived social isolation makes me sad: 5-year cross-lagged analyses of loneliness and depressive symptomatology in the Chicago Health, Aging, and Social Relations Study. Psychology and aging, 25, 453.

CHAPMAN, D. P., PERRY, G. S. & STRINE, T. W. 2005. PEER REVIEWED: The vital link between chronic disease and depressive disorders. Preventing chronic disease, 2.

CHUNG, H., LEE, E., CHOI, Y., KIM, J., KIM, D., ZOU, Y., KIM, C., LEE, J., KIM, H. & KIM, N. 2011. Molecular inflammation as an underlying mechanism of the aging process and age-related diseases. Journal of dental research, 90, 830-840.

32

CHUNG, H. Y., CESARI, M., ANTON, S., MARZETTI, E., GIOVANNINI, S., SEO, A. Y., CARTER, C., YU, B. P. & LEEUWENBURGH, C. 2009. Molecular inflammation: underpinnings of aging and age-related diseases. Ageing research reviews, 8, 18-30.

COLE, M. G. & DENDUKURI, N. 2003. Risk factors for depression among elderly community subjects: a systematic review and meta-analysis. American Journal of Psychiatry, 160, 1147-1156.

COPELAND, J. R., BEEKMAN, A. T., BRAAM, A. W., DEWEY, M. E., DELESPAUL, P., FUHRER, R., HOOIJER, C., LAWLOR, B. A., KIVELA, S.-L. & LOBO, A. 2004. Depression among older people in Europe: the EURODEP studies. World Psychiatry, 3, 45.

DANTZER, R., O'CONNOR, J. C., FREUND, G. G., JOHNSON, R. W. & KELLEY, K. W. 2008a. From inflammation to sickness and depression: when the immune system subjugates the brain. Nat Rev Neurosci, 9, 46-56.

DANTZER, R., O’CONNOR, J. C., FREUND, G. G., JOHNSON, R. W. & KELLEY, K. W. 2008b. From inflammation to sickness and depression: when the immune system subjugates the brain. Nature reviews. Neuroscience, 9, 46.

DAS, A. 2016. Psychosocial distress and inflammation: Which way does causality flow? Social Science and Medicine, 170, 1-8.

DE BERARDIS, D., SERRONI, N., CAVUTO, M., MARTINOTTI, G., CAMPANELLA, D., D'ALBENZIO, A., CARANO, A., ALESSANDRINI, M., FARINA, B., DI LORIO, G., MOSCHETTA, F. S. & DI GIANNANTONIO, M. 2009. The emerging role of C-reactive protein in affective and psychotic disorders. Italian Journal of Psychopathology, 15, 231-242.

DENTINO, A. N., PIEPER, C. F., RAO, K. M. K., CURRIE, M. S., HARRIS, T., BLAZER, D. G. & COHEN, H. J. 1999. Association of interleukin-6 and other biologic variables with depression in older people living in the community CAN'T ACCESS FULL TEXT. Journal of the American Geriatrics Society, 47, 6-11.

DIMOPOULOS, N., PIPERI, C., PSARRA, V., LEA, R. W. & KALOFOUTIS, A. 2008. Increased plasma levels of 8-iso-PGF2α and IL-6 in an elderly population with depression. Psychiatry Research, 161, 59-66.

DJERNES, J. K. 2006. Prevalence and predictors of depression in populations of elderly: a review. Acta Psychiatrica Scandinavica, 113, 372-387.

DOWLATI, Y., HERRMANN, N., SWARDFAGER, W., LIU, H., SHAM, L., REIM, E. K. & LANCTÔT, K. L. 2010. A meta-analysis of cytokines in major depression. Biological psychiatry, 67, 446-457.

DOYLE, T. A., DE GROOT, M., HARRIS, T., SCHWARTZ, F., STROTMEYER, E. S., JOHNSON, K. C. & KANAYA, A. 2013. Diabetes, depressive symptoms, and inflammation in older adults: results from the Health, Aging, and Body Composition Study. J Psychosom Res, 75, 419-24.

DUIVIS, H. E., VOGELZANGS, N., KUPPER, N., DE JONGE, P. & PENNINX, B. W. 2013. Differential association of somatic and cognitive symptoms of depression and anxiety with inflammation: findings from the Netherlands Study of Depression and Anxiety (NESDA). Psychoneuroendocrinology, 38, 1573-1585.

ERSHLER, W. B. & KELLER, E. T. 2000. Age-associated increased interleukin-6 gene expression, late-life diseases, and frailty. Annual review of medicine, 51, 245-270.

FISKE, A., WETHERELL, J. L. & GATZ, M. 2009. Depression in older adults. Annu Rev Clin Psychol, 5, 363-89.FORTI, P., RIETTI, E., PISACANE, N., OLIVELLI, V., MARIANI, E., CHIAPPELLI, M., LICASTRO, F. & RAVAGLIA, G.

2010. Blood inflammatory proteins and risk of incident depression in the elderly. Dement Geriatr Cogn Disord, 29, 11-20.

FRIEDMAN, E. M., HAYNEY, M., LOVE, G. D., SINGER, B. H. & RYFF, C. D. 2007. Plasma interleukin-6 and soluble IL-6 receptors are associated with psychological well-being in aging women. Health Psychology, 26, 305-313.

GALLAGHER, D., KISS, A., LANCTOT, K. & HERRMANN, N. 2016. Depression with inflammation: longitudinal analysis of a proposed depressive subtype in community dwelling older adults. International journal of geriatric psychiatry.

33

GARCÍA-PEÑA, C., WAGNER, F. A., SÁNCHEZ-GARCIA, S., JUÁREZ-CEDILLO, T., ESPINEL-BERMÚDEZ, C., GARCÍA-GONZALEZ, J. J., GALLEGOS-CARRILLO, K., FRANCO-MARINA, F. & GALLO, J. J. 2008. Depressive symptoms among older adults in Mexico City. Journal of general internal medicine, 23, 1973-1980.

GARIEPY, G., NITKA, D. & SCHMITZ, N. 2010. The association between obesity and anxiety disorders in the population: a systematic review and meta-analysis. Int J Obes (Lond), 34, 407-19.

HAAPAKOSKI, R., MATHIEU, J., EBMEIER, K. P., ALENIUS, H. & KIVIMÄKI, M. 2015. Cumulative meta-analysis of interleukins 6 and 1β, tumour necrosis factor α and C-reactive protein in patients with major depressive disorder. Brain, behavior, and immunity, 49, 206-215.

HAGER, K., MACHEIN, U., KRIEGER, S., PLATT, D., SEEFRIED, G. & BAUER, J. 1994. Interleukin-6 and selected plasma proteins in healthy persons of different ages. Neurobiology of aging, 15, 771-772.

HAMER, M. & CHIDA, Y. 2009. Associations of very high C-reactive protein concentration with psychosocial and cardiovascular risk factors in an ageing population. Atherosclerosis, 206, 599-603.

HASIN, D. S., GOODWIN, R. D., STINSON, F. S. & GRANT, B. F. 2005. Epidemiology of major depressive disorder: results from the National Epidemiologic Survey on Alcoholism and Related Conditions. Archives of general psychiatry, 62, 1097-1106.

HILES, S. A., BAKER, A. L., DE MALMANCHE, T. & ATTIA, J. 2012. A meta-analysis of differences in IL-6 and IL-10 between people with and without depression: exploring the causes of heterogeneity. Brain, behavior, and immunity, 26, 1180-1188.

HILES, S. A., BAKER, A. L., DE MALMANCHE, T., MCEVOY, M., BOYLE, M. & ATTIA, J. 2015. Unhealthy lifestyle may increase later depression via inflammation in older women but not men. J Psychiatr Res, 63, 65-74.

HOWREN, M. B., LAMKIN, D. M. & SULS, J. 2009. Associations of depression with C-reactive protein, IL-1, and IL-6: a meta-analysis. Psychosom Med, 71, 171-86.

HUNTER, C. A. & JONES, S. A. 2015. IL-6 as a keystone cytokine in health and disease. Nature immunology, 16, 448-457.

IRWIN, M. R., OLMSTEAD, R. & CARROLL, J. E. 2016. Sleep disturbance, sleep duration, and inflammation: a systematic review and meta-analysis of cohort studies and experimental sleep deprivation. Biological psychiatry, 80, 40-52.

KATON, W. J. 2011. Epidemiology and treatment of depression in patients with chronic medical illness. Dialogues Clin Neurosci, 13, 7-23.

KIECOLT-GLASER, J. K., DERRY, H. M. & FAGUNDES, C. P. 2015. Inflammation: depression fans the flames and feasts on the heat. American Journal of Psychiatry, 172, 1075-1091.

KÖHLER, C., FREITAS, T., MAES, M., ANDRADE, N., LIU, C., FERNANDES, B., STUBBS, B., SOLMI, M., VERONESE, N. & HERRMANN, N. 2017. Peripheral cytokine and chemokine alterations in depression: a meta‐analysis of 82 studies. Acta psychiatrica Scandinavica, 135, 373-387.

KOP, W. J., GOTTDIENER, J. S., TANGEN, C. M., FRIED, L. P. & ET AL. 2002. Inflammation and coagulation factors in persons greater than 65 years of age with symptoms of depression but without evidence of myocardial ischemia. The American Journal of Cardiology, 89, 419-24.

KRABBE, K. S., PEDERSEN, M. & BRUUNSGAARD, H. 2004. Inflammatory mediators in the elderly. Experimental gerontology, 39, 687-699.

KRISHNADAS, R. & CAVANAGH, J. 2012. Depression: An inflammatory illness? Journal of Neurology, Neurosurgery and Psychiatry, 83, 495-502.

LAMERS, F., VOGELZANGS, N., MERIKANGAS, K. R., DE JONGE, P., BEEKMAN, A. T. F. & PENNINX, B. W. J. H. 2013. Evidence for a differential role of HPA-axis function, inflammation and metabolic syndrome in melancholic versus atypical depression. Molecular Psychiatry, 18, 692-699.

LEE, Y., CHOI, K. & LEE, Y. K. 2001. Association of comorbidity with depressive symptoms in community-dwelling older persons. Gerontology, 47, 254-62.

34

LEONARD, B. E. 2007. Inflammation, depression and dementia: are they connected? Neurochemical research, 32, 1749-1756.

LOUCKS, E. B., BERKMAN, L. F., GRUENEWALD, T. L. & SEEMAN, T. E. 2006. Relation of social integration to inflammatory marker concentrations in men and women 70 to 79 years. Am J Cardiol, 97, 1010-6.

LU, Y., FENG, L., FENG, L., NYUNT, M. S., YAP, K. B. & NG, T. P. 2013. Systemic inflammation, depression and obstructive pulmonary function: a population-based study. Respiratory research, 14, 53.

LUCIANO, M., MÕTTUS, R., STARR, J. M., MCNEILL, G., JIA, X., CRAIG, L. C. A. & DEARY, I. J. 2012. Depressive symptoms and diet: Their effects on prospective inflammation levels in the elderly. Brain, Behavior, and Immunity, 26, 717-720.

LUUKINEN, H., JOKELAINEN, J. & HEDBERG, P. 2010. The relationships between high-sensitivity C-reactive protein and incident depressed mood among older adults. Scand J Clin Lab Invest, 70, 75-9.

MA, Y., CHIRIBOGA, D. E., PAGOTO, S. L., ROSAL, M. C., LI, W., MERRIAM, P. A., HÉBERT, J. R., WHITED, M. C. & OCKENE, I. S. 2011. Association between depression and C-reactive protein. Cardiology Research and Practice, 1.

MATSUSHIMA, J., KAWASHIMA, T., NABETA, H., IMAMURA, Y., WATANABE, I., MIZOGUCHI, Y., KOJIMA, N., YAMADA, S. & MONJI, A. 2015. Association of inflammatory biomarkers with depressive symptoms and cognitive decline in a community-dwelling healthy older sample: a 3-year follow-up study. J Affect Disord, 173, 9-14.

MEZUK, B., CHOI, M., DESANTIS, A. S., RAPP, S. R., ROUX, A. V. D. & SEEMAN, T. 2016. Loneliness, Depression, and Inflammation: Evidence from the Multi-Ethnic Study of Atherosclerosis - See Supplementary file for CESD and CRP. Plos One, 11.

MILANESCHI, Y., CORSI, A. M., PENNINX, B. W., BANDINELLI, S., GURALNIK, J. M. & FERRUCCI, L. 2009. Interleukin-1 receptor antagonist and incident depressive symptoms over 6 years in older persons: the InCHIANTI study. Biol Psychiatry, 65, 973-8.

NADROWSKI, P., CHUDEK, J., SKRZYPEK, M., PUZIANOWSKA-KUŹNICKA, M., MOSSAKOWSKA, M., WIĘCEK, A., ZDROJEWSKI, T., GRODZICKI, T. & KOZAKIEWICZ, K. 2016. Associations between cardiovascular disease risk factors and IL-6 and hsCRP levels in the elderly. Experimental Gerontology, 85, 112-117.

PAN, A., YE, X., FANCO, O. H., LI, H., YU, Z., WANG, J., QI, Q., GU, W., PANG, X., LIU, H. & LIN, X. 2008. The association of depressive symptoms with inflammatory factors and adipokines in middle-aged and older Chinese. PLoS ONE, 3.

PENNINX, B., KRITCHEVSKY, S. B., YAFFE, K., NEWMAN, A. B., SIMONSICK, E. M., RUBIN, S., FERRUCCI, L., HARRIS, T. & PAHOR, M. 2003a. Inflammatory markers and depressed mood in older persons: Results from the Health, Aging and Body Composition study. Biological Psychiatry, 54, 566-572.

PENNINX, B. H., GURALNIK, J. M., FERRUCCI, L., SIMONSICK, E. M., DEEG, D. H. & WALLACE, R. B. 1998. DEpressive symptoms and physical decline in community-dwelling older persons. JAMA, 279, 1720-1726.

PENNINX, B. W., KRITCHEVSKY, S. B., YAFFE, K., NEWMAN, A. B., SIMONSICK, E. M., RUBIN, S., FERRUCCI, L., HARRIS, T. & PAHOR, M. 2003b. Inflammatory markers and depressed mood in older persons: results from the Health, Aging and Body Composition study. Biol Psychiatry, 54, 566-72.

PENNINX, B. W., MILANESCHI, Y., LAMERS, F. & VOGELZANGS, N. 2013a. Understanding the somatic consequences of depression: biological mechanisms and the role of depression symptom profile. BMC medicine, 11, 129.

PENNINX, B. W. J. H., MILANESCHI, Y., LAMERS, F. & VOGELZANGS, N. 2013b. Understanding the somatic consequences of depression: Biological mechanisms and the role of depression symptom profile. BMC Medicine, 11.

QIN, T., LIU, W., YIN, M., SHU, C., YAN, M., ZHANG, J. & YIN, P. 2017. Body mass index moderates the relationship between C-reactive protein and depressive symptoms: evidence from the China Health and Retirement Longitudinal Study. Scientific Reports, 7.

35

REUBEN, D. B., CHEH, A. I., HARRIS, T. B., FERRUCCI, L., ROWE, J. W., TRACY, R. P. & SEEMAN, T. E. 2002. Peripheral blood markers of inflammation predict mortality and functional decline in high functioning‐ community dwelling older persons. ‐ Journal of the American Geriatrics Society, 50, 638-644.

RUDOLF, S., GREGGERSEN, W., KAHL, K. G., HÜPPE, M. & SCHWEIGER, U. 2014. Elevated IL-6 levels in patients with atypical depression but not in patients with typical depression. Psychiatry research, 217, 34-38.

SCHUCH, F., VANCAMPFORT, D., FIRTH, J., ROSENBAUM, S., WARD, P., REICHERT, T., BAGATINI, N. C., BGEGINSKI, R. & STUBBS, B. 2017. Physical activity and sedentary behavior in people with major depressive disorder: a systematic review and meta-analysis. Journal of affective disorders, 210, 139-150.

SCHULTZE-FLOREY, C. R., MARTÍNEZ-MAZA, O., MAGPANTAY, L., BREEN, E. C., IRWIN, M. R., GÜNDEL, H. & O’CONNOR, M.-F. 2012. When grief makes you sick: Bereavement induced systemic inflammation is a question of genotype. Brain, behavior, and immunity, 26, 1066-1071.

SHANKAR, A., MCMUNN, A., BANKS, J. & STEPTOE, A. 2011. Loneliness, social isolation, and behavioral and biological health indicators in older adults. Health Psychology, 30, 377.

SHELTON, R. C. & MILLER, A. H. 2010. Eating ourselves to death (and despair): the contribution of adiposity and inflammation to depression. Progress in neurobiology, 91, 275-299.

SLAVICH, G. M. & IRWIN, M. R. 2014. From stress to inflammation and major depressive disorder: A social signal transduction theory of depression. Psychological bulletin, 140, 774.

SMAGULA, S. F., ANCOLI-ISRAEL, S., BARRETT-CONNOR, E., LANE, N. E., REDLINE, S., STONE, K. L. & CAULEY, J. A. 2014. Inflammation, sleep disturbances, and depressed mood among community-dwelling older men. J Psychosom Res, 76, 368-73.

SONG, B. M., LEE, J.-M., CHOI, W., YOUM, Y., CHU, S. H., PARK, Y.-R. & KIM, H. C. 2015. Association between C reactive protein level and depressive symptoms in an elderly Korean population: Korean Social Life, Health and Aging Project. Bmj Open, 5.

SOYSAL, P., STUBBS, B., LUCATO, P., LUCHINI, C., SOLMI, M., PELUSO, R., SERGI, G., ISIK, A. T., MANZATO, E. & MAGGI, S. 2016. Inflammation and frailty in the elderly: A systematic review and meta-analysis. Ageing research reviews, 31, 1-8.

SOYSAL, P., VERONESE, N., THOMPSON, T., KAHL, K. G., FERNANDES, B. S., PRINA, A. M., SOLMI, M., SCHOFIELD, P., KOYANAGI, A. & TSENG, P.-T. 2017. Relationship between depression and frailty in older adults: A systematic review and meta-analysis. Ageing Research Reviews.

STEWART, J. C., JANICKI-DEVERTS, D., MULDOON, M. F. & KAMARCK, T. W. 2008. Depressive symptoms moderate the influence of hostility on serum interleukin-6 and C-reactive protein. Psychosom Med, 70, 197-204.

STEWART, J. C., RAND, K. L., MULDOON, M. F. & KAMARCK, T. W. 2009. A prospective evaluation of the directionality of the depression-inflammation relationship. Brain Behav Immun, 23, 936-44.

TIEMEIER, H., HOFMAN, A., VAN TUIJL, H. R., KILIAAN, A. J., MEIJER, J. & BRETELER, M. M. 2003a. Inflammatory proteins and depression in the elderly. Epidemiology, 14, 103-7.

TIEMEIER, H., HOFMAN, A., VAN TUIJL, H. R., KILIAAN, A. J., MEIJER, J. & BRETELER, M. M. B. 2003b. Inflammatory proteins and depression in the elderly. Epidemiology, 14, 103-107.

UDINA, M., CASTELLVÍ, P., MORENO-ESPAÑA, J., NAVINÉS, R., VALDÉS, M., FORNS, X., LANGOHR, K., SOLÀ, R., VIETA, E. & MARTÍN-SANTOS, R. 2012. Interferon-induced depression in chronic hepatitis C: a systematic review and meta-analysis.

VALENTINE, R. J., WOODS, J. A., MCAULEY, E., DANTZER, R. & EVANS, E. M. 2011. The associations of adiposity, physical activity and inflammation with fatigue in older adults. Brain Behav Immun, 25, 1482-90.

VALKANOVA, V., EBMEIER, K. P. & ALLAN, C. L. 2013. CRP, IL-6 and depression: a systematic review and meta-analysis of longitudinal studies. J Affect Disord, 150, 736-44.

36

VAN DEN BIGGELAAR, A. H., GUSSEKLOO, J., DE CRAEN, A. J., FROLICH, M., STEK, M. L., VAN DER MAST, R. C. & WESTENDORP, R. G. 2007. Inflammation and interleukin-1 signaling network contribute to depressive symptoms but not cognitive decline in old age. Exp Gerontol, 42, 693-701.

VANCAMPFORT, D., FIRTH, J., SCHUCH, F. B., ROSENBAUM, S., MUGISHA, J., HALLGREN, M., PROBST, M., WARD, P. B., GAUGHRAN, F. & DE HERT, M. 2017. Sedentary behavior and physical activity levels in people with schizophrenia, bipolar disorder and major depressive disorder: a global systematic review and meta analysis. ‐ World Psychiatry, 16, 308-315.

WELLS, G., SHEA, B., O'CONNELL, D., PETERSON, P., WELCH, V., LOSOS, M. & TUGWELL, P. The Newcastle-Ottawa Scale (NOS) for Assessing the Quality of Nonrandomised Studies in Meta-Analyses [Online]. Ottawa Health Research Institute: Ottawa. Available: http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp [Accessed].

WOLFF, J. L., STARFIELD, B. & ANDERSON, G. 2002. Prevalence, expenditures, and complications of multiple chronic conditions in the elderly. Archives of internal medicine, 162, 2269-2276.

ZALLI, A., JOVANOVA, O., HOOGENDIJK, W. J. G., TIEMEIER, H. & CARVALHO, L. A. 2016. Low-grade inflammation predicts persistence of depressive symptoms. Psychopharmacology, 233, 1669-1678.