Inflammation and frailty in the elderly: A systematic review and meta-analysis

The pathogenesis of frailty and the role of inflammation is poorly understood. We examined the evidence considering the relationship between inflammation and frailty through a systematic review and meta-analysis. A systematic literature search of papers providing data on inflammatory biomarkers and frailty was carried out in major electronic databases from inception until May 2016. From 1856 initial hits, 35 studies (32 cross-sectional studies n=3232 frail, n=11,483 pre-frail and n=8522 robust, and 563 pre-frail+robust; 3 longitudinal studies n=3402 participants without frailty at baseline) were meta-analyzed. Cross-sectional studies reported that compared to 6757 robust participants, both 1698 frail (SMD=1.00, 95%CI: 0.40-1.61) and 8568 pre-frail (SMD=0.33, 95%CI: 0.04-0.62) participants had significantly higher levels of C-reactive protein (CRP). Frailty (n=1057; SMD=1.12, 95%CI: 0.27-2.13) and pre-frailty (n=4467; SMD=0.56, 95%CI: 0.00-1.11) were associated with higher serum levels of interleukin-6 compared to people who were robust (n=2392). Frailty and pre-frailty were also significantly associated with elevated white blood cell and fibrinogen levels. In three longitudinal studies, higher serum CRP (OR=1.06, 95%CI: 0.78-1.44,) and IL-6 (OR=1.19, 95%CI: 0.87-1.62) were not associated with frailty. In conclusion, frailty and pre-frailty are associated with higher inflammatory parameters and in particular CRP and IL-6. Further longitudinal studies are needed.


a b s t r a c t
The pathogenesis of frailty and the role of inflammation is poorly understood. We examined the evidence considering the relationship between inflammation and frailty through a systematic review and meta-analysis. A systematic literature search of papers providing data on inflammatory biomarkers and frailty was carried out in major electronic databases from inception until May 2016. From 1856 initial hits, 35 studies (32 cross-sectional studies n = 3232 frail, n = 11,483 pre-frail and n = 8522 robust, and 563 pre-frail + robust; 3 longitudinal studies n = 3402 participants without frailty at baseline) were meta-analyzed. Cross-sectional studies reported that compared to 6757 robust participants, both 1698 frail (SMD = 1.00, 95%CI: 0.40-1.61) and 8568 pre-frail (SMD = 0.33, 95%CI: 0.04-0.62) participants had significantly higher levels of C-reactive protein (CRP). Frailty (n = 1057; SMD = 1.12, 95%CI: 0.27-2.13) and pre-frailty (n = 4467; SMD = 0.56, 95%CI: 0.00-1.11) were associated with higher serum levels of interleukin-6 compared to people who were robust (n = 2392). Frailty and pre-frailty were also significantly associated with elevated white blood cell and fibrinogen levels. In three longitudinal studies, higher

Introduction
Age-associated decline in reserve and function may result in a reduced ability to cope with acute or external stressors faced every day, which is typically defined as frailty (Clegg et al., 2013). Frailty is a relevant issue in geriatric medicine, since frailty is associated with a higher risk of poor outcomes such as falls, depression, disability, and mortality . Frailty is becoming one of the most significant clinical conditions affecting older people, with a prevalence of 10% for those older than 65 years and 30% for those older than 80 years .
Despite an increase in interest in frailty, the pathophysiological changes underlying and preceding frailty are not clearly known. Inflammation is one such potential pathophysiological change which may be closely linked with frailty (Chen et al., 2014). Proinflammatory cytokines may influence frailty either directly by promoting protein degradation, or indirectly by affecting important metabolic pathways (Lang et al., 2009). A direct association between frailty and elevated levels of inflammation, as marked by elevated interleukin-6 (IL-6), C-reactive protein (CRP), fibrinogen, and factor VIII, independent of common chronic disease states has been observed . Conversely, other studies have found that these markers are not predictive of incident frailty in the elderly (Yao et al., 2011). Thus, there is a lack of clarity considering the role and status of inflammation in frailty and to the best of our knowledge, no meta-analysis has attempted to synthesize the available data on this topic. Therefore, we conducted a systematic review and meta-analysis comparing the inflammatory profile of frail and pre-frail with and robust subjects in cross sectional studies. In addition, we investigated whether or not any inflammatory parameters at baseline could predict the onset of frailty in prospective studies. Our hypothesis was that both frailty and pre-frailty were associated with higher pro-inflammatory cytokines levels.

Materials and methods
This systematic review was conducted according to the Strengthening the Reporting of Observational Studies in Epidemiology [STROBE] criteria (von Elm et al., 2008) and the recommendations in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [PRISMA] statement (Liberati et al., 2009). This work followed a pre-determined, but unpublished protocol available upon request.

Search strategy
Three independent authors (PS, BS and NV)

Study selection
Included studies were those that were published quantitative studies of a cross sectional or longitudinal design that (1) reported on serum levels of inflammatory parameters, (2) used a validated and standardized method for assessing frailty, (3) included a control group (pre-frail and robust as separated entities or together): (4) used serum inflammatory parameters as predictors of frailty (longitudinal design). Studies were excluded if they (1) did not use a clear diagnostic criteria for frailty or used only one item for its diagnosis (e.g. low gait speed), (2) measured only in vitro parameters, (3) did not measure or did not report quantitative cytokine levels.

Data extraction
To be included in the quantitative analyses, we required data on serum inflammatory parameters in frail, pre-frail and robust participants expressed as means ± standard deviation (SD) or median with range (or interquartile range).
Two authors (PS, NV) independently extracted data from the selected studies in a standardized Microsoft Excel spreadsheet. Any disagreement was resolved by through discussion with a third author (BS). The following information was extracted: (i) characteristics of the study population (e.g. sample size, demographics, country in which the study was performed); (ii) setting in which the study was performed; (iii) diagnostic criteria for frailty; (iv) inflammatory parameters assessed with correspondent method of measurement; (v) demographic characteristics (mean age and percentage of women) and mean body mass index (BMI) by frailty status; (v) categorization used for dividing the sample in groups by serum inflammatory levels (for longitudinal studies); (vi) type and number of adjustments in the multivariate analyses (for longitudinal studies); (vi) follow-up period (only for longitudinal studies).
When information on frailty and/or serum inflammatory parameters was missing, study authors were contacted to obtain unpublished data at least 4 times in a one month period.

Outcomes
The primary outcomes were the serum levels of inflammatory cytokines and parameters in frail vs. pre-frail and robust (as separate entities) or vs. pre-frail/robust (as only one group). For incident frailty, the odds ratios (ORs) adjusted for the highest number of covariates available, were considered to assess the association between serum inflammatory parameters and frailty and considered as secondary outcome of our work.

Assessment of study quality
Study quality was assessed by two investigators (PS, PL), whilst a third reviewer was available for mediation (NV). For cross-sectional and longitudinal studies, the Newcastle-Ottawa Scale (NOS) (Wells et al., 2012) was used to assess study quality. The NOS assigns a maximum of 9 points based on three quality parameters: selection, comparability, and outcome (Wells et al., 2012).
In primary analyses, standardized mean differences (SMDs) and 95% confidence intervals (CI) were calculated. In the secondary analyses, the most adjusted pooled HRs were calculated for longitudinal analyses. When combining studies, the random effects model was used to account for anticipated heterogeneity (DerSimonian and Laird, 1986).
Heterogeneity was measured using the chi-squared and Isquared statistics, assuming that a p ≤ 0.10 for the former and a value ≥50% for the latter indicated a significant heterogeneity (Higgins and Thompson, 2002). Given significant heterogeneity, a meta-regression analysis was performed using differences in mean age, body mass index (BMI) and percentage of females among groups (frail, pre-frail, robust) as moderators. Moreover, a sensitivity analysis was conducted stratified by continent in which the study was performed (North America, Asia, Europe), setting (community-dwelling vs. hospital), and definition of frailty (Fried's criteria vs. other definitions).
Publication bias was assessed by visually inspecting funnel plots and using the Begg-Mazumdar Kendall tau (Begg and Mazumdar, 1994) and the Egger bias test (Egger et al., 1997). To account for publication bias, we used the trim-and-fill method, based on the assumption that the effect sizes of all the studies are normally distributed around the center of a funnel plot; in the event of asymmetries, it adjusts for the potential effect of unpublished studies (Egger et al., 1997). Finally, the fail safe number of negative studies that would be required to nullify (i.e. make p > 0.05) the effect size was calculated (Rosenthal, 1979).
All of the included studies used a modified version of Fried et al. (2001) definition for frailty, except for five defined by Frail Scale, Balducci Score, Modified Physical Performance Test Score, Identification of Seniors At Risk, MPI (Addison et al., 2014;Brouwers et al., 2015;Chao et al., 2015;Comptè et al., 2013;Fontana et al., 2013).
Similar results were evident regarding white blood cells and fibrinogen levels, while no differences emerged for TNF-alpha (Table 1).

Meta-regression and sensitivity analyses
Since the differences in inflammatory parameters between frailty, pre-frailty and robustness were characterized by a high heterogeneity (I 2 ≥ 50%, p < 0.05), where a sufficient number of studies were available (at least 4 for each outcome), we ran a metaregression analysis to seek potential moderators.
Supplementary Table 3 shows the results of cross-sectional studies according to some strata, namely country in which the study was performed, the setting and the definition of frailty. Overall, these moderators seem not to significantly affect our findings. In particular, we found that frailty and pre-frailty were associated with significantly elevated CRP and IL-6 levels across all geographical settings and among community and institutionalized participants (see Supplementary Table 3).

Longitudinal meta-analysis findings
Three studies Puts et al., 2005;Reiner et al., 2009) followed-up 3402 older participants without frailty at baseline for a median of 3 years (range: 3-10) (Supplementary Table  4). Table 1 Meta-analysis of studies comparing frail vs. pre-frail or robust older participants with publication bias assessment.

Analysis
Number of studies The bold values are to highlight the significant differences between groups. Abbreviations:, CI = confidence interval, SMD = standardized mean difference.

Fig. 1.
Relationship between serum C-reactive protein levels (a) and interleukin 6 (b) at baseline and incident frailty, adjusted for potential confounders.

Publication bias
Judging from a visual inspection of funnel plots and using Egger's test (Table 1 for cross-sectional studies), no publication bias was evident for all the outcomes included. A similar absence of publication bias was present for longitudinal studies, although limited by the number of studies included.

Discussion
In this meta-analysis including 32 cross-sectional studies and 23,910 older subjects, we observed that frailty and pre-frailty were associated with significantly higher serum inflammatory parameters compared to robust participants. In particular, we found a large increase in CRP and IL-6 in frail and pre-frail participants versus robust participants, with very high fail safe number of studies required to nullify these results. The elevated CRP and IL-6 were consistent across geographical regions and both in community and hospital settings. We also found evidence of elevated white bloods cells and fibrinogen. On the contrary, three large prospective studies failed to find any association between higher inflammatory levels at baseline and incident frailty. Meta-regression analyses of the cross-sectional data suggest that age and BMI moderate the relationship between CRP and frailty.
The relationship between inflammation and frailty is complex since both linearly increase with advancing old age. Both higher inflammatory levels and frailty are associated with several negative outcomes in the elderly, like higher mortality, hospitalization rate and co-morbidity onset (Piggott et al., 2015;Sergi et al., 2015;Zunszain et al., 2013). In cross-sectional studies, the association of frailty with higher inflammation appears to be consistent since both frail and pre-frail participants showed significant higher serum levels of CRP, TNF-a, IL-6, white blood cells and fibrinogen. Several reasons could explain these results. First, frail and pre-frail participants have a higher presence of concomitant factors like disability, medical conditions that could increase the inflammatory parameters. Second, as confirmed by our analyses, frail and pre-frail people (particularly if community-dwelling) are generally more obese than robust participants, and obesity significantly increases inflammatory parameters (Greenberg and Obin, 2006;Solmi et al., 2015;. This hypothesis is also in line with the increase in adiposity observed in frail subjects that seems to affect also muscular mass (Addison et al., 2014) and indirectly confirmed by our meta-regression analysis showing that higher differences in BMI between frail or pre-frail vs. robust participants moderated the results regarding serum CRP concentrations. Finally, frail people seem to have a significant reduction in innate immune system, Tcell activity, antibodies production and increase in mitochondrial activity with an increase in oxidative stress products, ultimately leading to an increase in serum inflammatory levels (Hubbard and Woodhouse, 2010;Li et al., 2011).
Conversely, the analyses from our longitudinal studies did not show any association between higher inflammatory levels and the onset of frailty. The lack of relationship might be due to the paucity of data, however, it may well also indicate that these inflammatory markers are not associated with a predisposition for developing frailty and may arise once frailty has set int. Another hypothesis is that the longitudinal studies included in our meta-analysis included a median of 9 baseline potential confounders. Therefore, an over-adjustment of the analyses could not to be excluded. The absence of an univocal operational definition for frailty, in fact, makes the development of measurable biomarkers particularly important (Calvani et al., 2015) and if higher inflammatory parameters are able to predict the onset of frailty could be of importance since they are largely diffused worldwide and since higher inflammatory parameters probably contribute to transitions from frailty to disability and other negative outcomes (Zaslavsky et al., 2013). A possible hypothesis for this lack of association is that frail people are very sensitive to acute and sub-acute diseases that might increase inflammatory parameters during follow-up period and none of the studies included adjusted their analyses for inherent changes in these markers. Clearly, future studies including these adjustments are needed to better investigate the potential role of inflammatory parameters in predicting frailty in the elderly.
The meta-regression analyses identified some characteristics of frail and pre-frail participants compared to robust subjects could explain the differences in inflammatory parameters seen. In particular, differences in age and BMI moderated the results in the comparison between frail vs. pre-frail/robust regarding CRP and IL-6 suggesting an important role of these factors in the higher inflammatory levels found in these subjects compared to robust ones. On the contrary, the definition of frailty did not affect our findings suggest that, independently from the definition used, frail people are characterized by a metabolic signature in which inflammation plays a relevant role (Fontana et al., 2013).
The findings of our study should be interpreted within its limitations. First, we encountered moderate-high heterogeneity in most of the cross-sectional analyses. Whilst the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines state heterogeneity is to be expected when analyzing observational data, we were only partially able to explain the heterogeneity with meta regression analysis. Second, almost all the studies used the criteria proposed by Fried et al., but this definition includes only physical frailty and does not consider other aspects (for instance cognitive status), which is of relevance in frailty (Zaslavsky et al., 2013). The focus on physical frailty is not suitable for people with advance cognitive decline (e.g. dementia) who may also be highly susceptible to frailty. Therefore, future research is also required to understand frailty (considering cognitive status) and inflammation among people with dementia. Moreover, the frailty phenotype did not allow to a better understanding of the underlying contributing factors to frailty and inflammation. Third, the categorization of people as robust/pre-frail/frail other factors (e.g. differences in comorbidities) may explain some of the relationship between physical frailty and inflammation. Therefore, future studies using other definitions of frailty should attempt to match up frail/pre frail and robust participants and account for between group differences in important comorbidities/factors which may also influence inflammation. Fourth, all the studies investigating frailty according to the definition of Fried et al. and did not use the original versions and instead adapted the frailty criteria which can influence the quality of the composite score and potentially introduce bias (Theou et al., 2015). Agreement between modified criteria with the primary frailty phenotype, in fact, is generally low-moderate. However, the impact of this on our results is unclear. Finally, only three longitudinal studies were eligible, with a lack of clarity concerning the results. Therefore, future longitudinal research is required to disentangle the directionality of the inflammation and frailty relationships we observed in our comprehensive cross sectional analyses. Nonetheless, our meta-analysis is a first and included a large number of studies included and inflammatory outcomes. The data from our cross sectional results were also free from publication bias and the results indicating greatly increased levels of CRP and IL-6 had very large fail safe number of studies (both >1000).
In conclusion, frailty and pre-frailty are associated with higher inflammatory parameters levels, in particular CRP and IL-6. However, longitudinal studies did not confirm these findings, suggesting that other studies are needed to better understand if these inflammatory markers could be used as potential biomarkers of frailty in the elderly.

Conflict of interest
None.