A systematic review and meta-analysis of the stability of peripheral immune markers in healthy adults

Abstract

Peripheral immune markers are widely used to predict risk for inflammatory disease. However, whether single assessments of inflammatory biomarkers represent stable individual differences remains unclear. We reviewed 50 studies (N = 48,674; 57 % male; mean age 54 (range 13–79) years) that assessed markers of inflammation on >1 occasion, with time between measures ranging from 24 h to 7+ years. Separate random effects meta-analyses were conducted for each inflammatory marker and time interval. Markers that had broad coverage across most time intervals included C-reactive protein (CRP; k = 37), interleukin (IL)-6 (k = 22), TNF-α (k = 10), and fibrinogen (Fg; k = 9). For CRP, IL-6, and TNF-α, stability estimates generally decreased with time, with strong to moderate stability over intervals <6 months (r’s = 0.80-0.61), modest to moderate stability over 6 months – 3 years (r’s = 0.60-0.51), and low stability for >3 years (r’s = 0.39-0.30). Estimates were less reliable for Fg for time intervals ≤ 3 years although they generally followed the same pattern; more reliable findings suggested greater stability for Fg than other markers for intervals >3 years (r = 0.53). These findings suggest that single measures of inflammatory biomarkers may be an adequate index of stable individual differences in the short term (<6 months), with repeated measures of inflammatory biomarkers recommended over intervals ≥ 6 months to 3 years, and absolutely necessary over intervals >3 years to reliably identify stable individual differences in health risk. These findings are consistent with stability estimates and clinical recommendations for repeated measurement of other cardiovascular measures of risk (e.g., blood lipids, blood pressure).

1. Introduction

Increasing evidence suggests that some otherwise healthy individuals have chronically elevated levels of inflammatory markers in peripheral circulation that are independent of antigen exposure or injury (Rohleder, 2014). These markers measured at a single point in time are widely used in the psychoneuroimmunology (PNI) literature, and clinically, as a predictor of risk for common inflammatory diseases of aging. For example, studies demonstrate associations between circulating levels of the inflammatory markers C-reactive protein (CRP), interleukin (IL)-6, and fibrinogen and the development of coronary heart disease (CHD) up to 12 years later (Danesh et al., 2004; Danesh et al., 2008; Folsom et al., 2002; Kannel et al., 1987; Kaptoge et al., 2014; Zhang et al., 2018), and future risk for frailty, neurocognitive decline, and all-cause mortality (Engelhart et al., 2004; Singh and Newman, 2011; Schmidt et al., 2002; Walker et al., 2019). Similarly, CRP, IL-6, and tumor necrosis factor (TNF)-α¹ predict cancer incidence and cancer-related mortality (Dibaba et al., 2019; Emerging Risk Factors Collaboration, 2010; Guo et al., 2013). These associations are largely independent of traditional risk factors and often contribute to the prediction of risk at a magnitude similar to adiposity, cigarette smoking, and diabetes (Danesh et al., 2004; Folsom et al., 2002; Kannel et al., 1987). Thus, convergent evidence supports the assessment of markers of inflammation (e.g., CRP, IL-6, fibrinogen, TNF-α) among otherwise healthy individuals in the prediction of general health and risk for common diseases of aging.

For economic and practical reasons, PNI researchers and clinicians often use measurements of inflammatory markers taken at a single point in time to predict disease risk among otherwise healthy individuals. However, there is some evidence to suggest that single assessments of inflammatory markers may not provide a reliable index of stable individual differences and thus may underestimate health risk. For example, one large prospective study found that accounting for year-to-year variation in IL-6 increased prediction of CHD risk by almost 50 % over a 12-year period (i.e., OR: 1.46 → 2.14; Danesh et al., 2008). Similarly, accounting for variability in CRP over time has been shown to increase prediction of risk for CHD over an 8-year period (i.e., HR:1.57 → 2.62; Koenig et al., 2003). Thus, understanding the stability of inflammatory factors over time has broad implications for estimating their contribution to the prediction of risk for inflammatory diseases.

To date, the temporal stability of these peripheral markers is poorly understood and has not been widely investigated or accounted for in measurements of risk. In the current review, we examine the temporal stability of the most frequently assessed circulating markers of inflammation. We define stability as the correlation between measurements over time (i.e., test-retest reliability; Segerstrom, 2020). In the context of physiometrics, intraclass correlations of 0.48 between physiological measures have been considered moderately stable, 0.62-0.73 highly stable, and 0.90-0.92 very highly stable (Gloger, Smith, and Segerstrom, 2021). In the psychometrics literature, ICCs <0.40 are generally considered poor, 0.40-0.59 fair, 0.60-0.74 good, and estimates above 0.75 have generally been considered to have excellent temporal stability (Moriarity and Alloy, 2021). Thus, it seems reasonable to define stability coefficients of <0.50 as low, 0.50-0.60 as modest, 0.60-0.75 as moder-ate, and >0.75 as strong.

Many contemporaneous factors can influence circulating levels of inflammatory mediators, including environmental, individual, and measurement considerations. Environmental factors that associate with increased circulating levels of inflammatory markers include exposure to pollution (Baja et al., 2013; Green et al., 2016) and both acute and chronic psychosocial stress (Marsland et al., 2017; Segerstrom and Miller, 2004). Circulating levels of some inflammatory mediators also show diurnal and seasonal variation, with higher levels in afternoon/evening and in the winter months (Maes et al., 1994; Nilsonne et al., 2016; Rudnicka et al., 2007). Individual factors that can influence circulating levels of inflammatory mediators include diet or fasting state of the participant, recent physical activity, acute or chronic inflammatory-related illness, psychiatric illness, and/or medication use (Zhou et al., 2010; Pearson et al., 2003). Here, the presence of acute or chronic illness, psychiatric illness, bouts of physical activity, or recent digestion of meals high in dietary fats may increase circulating levels of inflammatory mediators (Zhou et al., 2010; Pearson et al., 2003), whereas most anti-inflammatory medications and some psychiatric medications may decrease them (Zhou et al., 2010; Pearson et al., 2003). There is also evidence for an age-related increase in mean concentration of many inflammatory markers (Lustig et al., 2017). In addition to these environmental and individual factors, measurement and sampling factors may influence levels of circulating inflammatory mediators, including the type and precision of the biological assay (e.g., enzyme-linked immunosorbent assay (ELISA) for a single analyte vs multiplex assay for multiple analytes; Pang et al., 2005; Zhou et al., 2010) and differences in the matrix from which the analyte is measured (e.g., blood, plasma, serum; Zhou et al., 2010). Taken together, these environmental, individual, and measurement factors can influence the level of circulating inflammatory mediators as assessed at a single point in time and contribute noise to the reliable assessment of stable individual differences. To obtain measures that more reliably characterize long-term risk for disease, repeated measures may be required (Pearson et al., 2003).

Although evidence suggests that single measures of inflammation are impacted by multiple factors and thus may underestimate disease risk, individual studies provide mixed evidence for stability in markers of inflammation over time, reporting anywhere from modest to strong stability coefficients for CRP, IL-6, and TNF-α over multiple years (Al-Delaimy et al., 2006; Clendenen et al., 2010; Danesh et al., 2004; Engelberger et al., 2015; Epstein et al., 2013; Gu et al., 2009; Koelman et al., 2019; Hofmann et al., 2011). Other studies suggest that fibrinogen demonstrates modest stability over a period of months to years but low stability over a period of several weeks (De Bacquer et al., 1997; Rosenson et al., 1994; Sakkinen et al., 1999). With regard to potential moderators, the temporal stability of inflammatory markers across the lifespan is largely unknown (Nash et al., 2013; Wu et al., 2012; Koelman et al., 2019). Similarly, there is inconsistent evidence that sex may influence the temporal stability of inflammatory markers over time (Fontes et al., 2013; Koelman et al., 2019; Nash et al., 2013; Wu et al., 2012). It is notable that many of these studies are limited by small sample size or truncated age range of the population. In theory, stability of measurements should be greater over shorter time intervals (e.g., days, weeks) because environmental, individual, and measurement factors are less variable across measurements. However, to date, there are no known systematic reviews investigating evidence for stability of peripheral markers of inflammation among healthy samples across the lifespan.

In the current investigation, we aimed to systematically assess the temporal stability of peripheral markers of inflammation among healthy individuals. We also sought to determine whether the stability of inflammatory biomarkers was moderated by age, sex, or sampling characteristics. With regard to sampling factors, we were interested in whether stability of inflammatory biomarkers was affected by method of measurement (e.g., ELISA vs Multiplex assay; Pang et al., 2005), sampling matrix (e.g., plasma, serum), time of day for blood draw (morning/afternoon), or fasting status (i.e., fasting or not; Zhou et al., 2010). Our primary hypothesis was that most inflammatory markers would demonstrate modest to moderate stability over time. We posited that temporal stability would decrease over time, with higher concordance when measured over weeks/months compared to years. Given the sparse and potentially equivocal findings reported in the literature to date, we had no a-priori hypotheses about how the stability of inflammatory markers would differ by age, sex, method of analytic measurement, sampling matrix, time of day for blood draw, or fasting status.

2. Methods

The Preferred Reporting Items for Systematic Reviews and meta-Analyses (PRISMA) guidelines (Moher et al., 2009) were used as a guide in performing this review. The search strategy, strategy for data collection, and data analysis plan were registered on the Open Science Framework prior to inspection and analysis of the data (https://osf.io/ebjzx/). Minor modifications from the initial registration are listed in the Supplement (Fig S9).

2.1. Literature search strategy

Our electronic search strategy used combinations of the following search terms: humans, chronic stress, psychology, biomarker, immune, cytokine, inflammatory markers; cytokine levels; immune markers, CRP, IL-6, IL-1β, TNF-α, fibrinogen, population, prospective, longitudinal, stability, reliability, ICC, intraclass correlation, repeated, reproducibility, variability, intraindividual variability, and long-term variability. We restricted our search to peer-reviewed studies published in English after 1971 (the origin of the ELISA) in humans (Aydin, 2015). The exact search terms used in each database is included in the supplementary material (Fig. S1).

2.2. Inclusion/exclusion criteria

Studies were included if they met all of the following criteria: a) peer reviewed empirical papers published in English; b) measurement of at least one inflammatory biomarker in peripheral circulation; c) inflammatory biomarker was measured at two time points at least 24 h apart. Studies were excluded if they met any of the following criteria: 1) case studies, reviews, or dissertations; 2) studies with an eligible group sample size < 5; 3) mean age of sample < 1 year old. Studies were also excluded if they were limited to only clinical populations (e.g., hypertension, obesity, metabolic syndrome, symptoms of depression or anxiety, mild cognitive impairment, or women who were pregnant or < 12 months post-partum). Intervention studies were also excluded unless they included a healthy no treatment control group; in these studies, only data from the no treatment group were included in the meta-analysis (See Figure S2 for more detailed exclusion criteria).

2.3. Study selection

Once articles were identified (N = 2525), duplicates were removed (Fig S3, PRISMA Flow Diagram). Titles and abstracts were each screened for eligibility by two independent reviewers, with percent agreement between pairs ranging from 80 to 86 %. Mismatched decisions were settled by consensus of the study team. 155 full-text articles and an additional 34 articles identified from their references were assessed for eligibility and 151 retained. See Fig. S2 for primary reasons for excluding studies. If identified articles did not provide enough data to estimate an effect size (k = 143 out of 151), we contacted the study’s authors up to three times via email. We received data from 42 of the 143 studies (no response k = 74; data no longer available k = 27), resulting a final sample of 50 studies for inclusion in the review.

The initial literature search was conducted in 11/2017. Articles were screened for eligibility and data was extracted from 12/2017 – 05/2019.Authors were contacted between 06/2019–01/2020. A final search of the literature for any newly published articles that contained the necessary information in the text was performed from 07/2021 – 09/2021; no additional papers were identified for inclusion.

2.4. Data abstraction and study coding

See Fig. S4 for a detailed description of extracted variables. Data extraction and coding was performed by 3 independent reviewers (CW, BN, KM). All data was double entered. Based on the distribution of time intervals between assessments across studies, five different time intervals were examined for CRP and four intervals were examined for all other inflammatory markers (Table 1). These intervals ranged from 24 h to 7+ years. Age of the study sample was coded both as a continuous variable, and as a categorical variable for IL-6 and CRP (IL-6: 0 = <70 years, 1 = ≥ 70 years; CRP: 0 = <45 years, 1 = ≥ 45 years; Nash et al., 2013; Wu et al., 2012). Sex was coded as [0 = male; 1 = female].

Table 1.

Stability Estimates and Confidence Intervals for the Measurement of Inflammatory Markers Across Time Intervals.

Biomarker	Interval
	1 day – 3 months	6 months – 1 year	3 years	5 years	>7 years
CRP	r = 0.789	r = 0.505	r = 0.530	r = 0.355	r = 0.347
	[0.646, 0.878]	[0.339, 0.641]	[0.399, 0.639]	[0.285, 0.420]	[0.234, 0.450]
	k = 8	k = 9	k = 14	k = 17	N = 11
IL-6	r = 0.607	r = 0.599	r = 0.574	r = 0.385
	[0.405, 0.753]	[0.419, 0.734]	[0.458, 0.670]	[0.284, 0.479]
	k = 8	k = 9	k = 9	k = 11
TNF-α	r = 0.802	r = 0.600	0.816*	r = 0.292*
	[0.341, 0.952]	[0.288, 0.797]	[0.033, 0.978]	[0.106, 0.458]
	k = 5	k = 5	k = 4	k = 2
Fibrinogen	r = 0.883*	r = 0.614*	r = 0.586*	r = 0.533
	[0.313, 0.985]	[0.512, 0.699]	[0.472, 0.680]	[0.496, 0.568]
	k = 2	k = 4	k = 2	k = 6

Notes: CRP = C-reactive protein; IL = interleukin; TNF = tumor necrosis factor; r = correlation between measurements or stability in these measurements across this interval; [ ] = 95 % confidence interval; k = # of independent study samples.

^*Estimate is based on<5 studies and thus warrants less confidence (Lipsey and Wilson, 2001).

For sample processing variables, time of day of blood draw was coded [0 = morning; 1 = afternoon], fasting status was coded [0 = no; 1 = yes], and the sample matrix in which the immune marker was measured was coded [0 = plasma; 1 = serum; 2 = whole blood]. The method of measurement for all cytokines was coded [0 = ELISA; 1 = multiplex; (Pang et al., 2005)]. For CRP, measurement methods were coded [0 = single analyte ELISA; 1 = multiplex ELISA; 2 = immunoturbidometric assay; 3 = latex aggregation/latex-enhanced immunoturbidometric assay; 4 = immunonephelometric assay; 5 = radial immunodiffusion assay; 6 = non-latex bead-based assay; (Aziz et al., 2003; Eda et al., 1998; Molina-Bolívar and Galisteo-González, 2005; Otsuji et al., 1982)]. For fibrinogen (Fg), measurement methods were coded [0 = functional/clotting assays (Clauss assays, prothrombin time (PT)-Fg tests, other clottable protein assays); 1 = immunoassays (ELISA, radial immunodiffusion, electrophoretic techniques, immunonephelometry/immunoturbidimetry); (Mackie et al., 2003; Rumley et al., 2003)].

2.5. Assessment of study quality

To assess study quality, we used a modified version of the “NHLBI Quality Assessment Tool for Before-After (Pre-Post) studies with no control group” (NHLBI, 2020; Fig. S8). This tool assesses internal validity, reliability, and systematic error in study design. We modified the measure to drop items assessing the quality of intervention (see Fig. S8). The quality of each study was categorized by two reviewers (inter-rater reliability = 0.75) as “good” (low risk of bias, considered valid), “fair” (some bias, largely valid), or “poor” (substantial bias, low validity) in accordance with the guidance documents. When necessary, a third reviewer examined the article to reach a coding agreement.

Publication bias was of limited concern, since our question about stability of immune measures over time was not a primary measure or focus of the majority of studies included in the meta-analysis.

2.6. Analytic strategy

2.6.1. Calculation of effect sizes

For each inflammatory marker and time interval, we conducted a separate random effects meta-analysis on the correlations of biomarker measures between time points. When several correlations were repeated in a single cohort within a study (e.g., cytokines measured once a day for one week) the observed value on day 2 would be used twice: once to calculate the correlation between day 1 and day 2; and again for the correlation between day 2 and day 3. Thus, each pairwise comparison of time points could result in multiple instances of correlation for a given time interval from the same group of participants. In this case, multiple correlations would have some homogeneity since they are from the same group of participants within the same study. Therefore, we aggregated correlations that were from the same group of participants within the same time interval before performing the meta-analysis. Specifically, correlations were averaged and variances/standard deviations were pooled within each cluster to take dependency into account (Marín-Martinez and Sánchez-Meca, 1999). A cluster was defined as any single study or single group of participants that generated multiple correlations for a given/defined time interval. To determine dependency in the process of aggregating values, we calculated the intra-cluster correlation (ICC) to determine the portion of the variability attributable to homogeneity of the values within clusters.

Aggregated correlations were then transformed into their corresponding Fisher’s Z score to formulate confidence intervals and carry out quantitative meta-analysis. Given the heterogeneity in the populations from which our data was drawn, we used a random effects model to pool the Fisher’s Z transformed correlations (Lipsey and Wilson, 2001). Specifically, the “meta” package in R was used to determine the point estimate and 95 % confidence interval for the transformed correlations for the various cytokine/time interval combinations. The study from which the correlations were derived was included in the model as a random effect to account for potential dependencies among correlations coming from the same study. The corresponding point estimates and confidence intervals were then transformed back into the correlation scale. For instances in which there was only one aggregated correlation for a particular cytokine and time interval combination, a confidence interval was constructed using the sample size and standard deviation from the study, but no quantitative analyses were performed. For the purposes of this review, we define stability coefficients of <0.50 as low, 0.50-0.60 as modest, 0.60-0.75 as moderate, and >0.75 as strong.

2.6.2. Moderator analyses

Variables considered for moderator analyses included assay method of measurement, sample matrix, time of day for blood draw, fasting status, age, and sex (for a summary by study, see Table S1, Table S2). To estimate effect sizes and confidence intervals by moderator type, the same meta-analytic analyses were performed on subsamples of the data separated by moderator code.

2.6.3. Assessment of heterogeneity

For each point estimate, heterogeneity among effect sizes was assessed using the Cochrane’s Q test, with a p-value < 0.05 suggesting significant heterogeneity. An I² statistic was also examined to estimate how much variation in the effect size estimate could be attributed to “true” between-studies variability. As recommended by Higgins (Higgins et al., 2003), I² < 25 % was interpreted as low, 50 % as moderate, and 75 % as high heterogeneity. Low values for I² indicate very little between-study variation among the studies included in the effect size estimate.

2.6.4. Assessment of study quality

Sensitivity analyses were performed excluding “poor” quality studies to determine their impact on the results. We also conducted a sensitivity analysis including only studies rated “fair” or “good” that were specifically designed to answer our question of interest (notated in Table S1).

3. Results

The 50 studies included in the meta-analyses are summarized in Table S1 and References are listed in Figure S10. Fourteen articles were rated as “good,” 31 articles were rated as “fair,” and 5 articles were rated as “poor” (Table S1). Studies were rated as “poor” primarily due to in-consistencies in assay or sample processing methods across timepoints; all also had relatively small sample sizes (N’s = 19–106). Eleven of the included studies were specifically designed to answer our question of interest (i.e., the main focus of the article was to examine the stability of a number of physiometric variables over repeated time points among healthy adults; Table S1), with 22 additional studies contributing data from epidemiologic samples, and 15 studies contributing data from healthy control groups. The studies represented a total of 48,674 individuals, who were on average 56.7 % male and 53.6 years of age [range: 13–79 years]. Studies were performed in a number of regions across the world, including North America, Europe, Asia, Australia, and Iceland, and individuals were representative of these regions in racial & ethnic makeup. Thus, our sample of studies was broadly representative of a Northern American and Eurasian population.

The studies included in the meta-analysis reported on a range of immune markers, including CRP, fibrinogen, IL-6, TNF-α, IL-8, IL-1β, IFN-γ, IL-10, IL-1Rα, IL-2, IL-4, IL-5, sTNF-RII, IL-12, IL-13, and IL-1α. Overall, there were 44 combinations of cytokines and time intervals with correlational data available for meta-analysis (Table 1, Table S3). Among samples eligible for quantitative meta-analysis, there were 22 instances in which multiple correlations for a given time interval were derived from a single study or single group of participants. In these cases, calculated ICCs were generally found to be between 0.70 and 0.90. These ICC values further justified aggregating multiple correlations from the same study within the same time interval. After aggregation, there were 35 instances in which there was more than one study per interval, and 9 instances in which there was only one aggregated correlation for a particular cytokine and time interval combination (Table 1, Table S3). There were not enough studies for reliable estimation of an effect size (k > 5) for most markers of inflammation across time intervals (Lipsey and Wilson, 2001). For this reason, the primary results focus on CRP, IL-6, TNF-α, and fibrinogen. Results for the other immune markers are included in supplemental materials (Table S3, Fig. S5).

3.1. Individual immune markers

Stability estimates (correlations), 95 % confidence intervals, and sample sizes (k = number of studies contributing to each estimate) for each interval are provided for CRP, IL-6, TNF-α, and fibrinogen in Table 1. A visual plot of the stability estimates over increasing time intervals is provided in Fig. 1. Forest plots for CRP, IL-6, TNF-α, and fibrinogen by time interval are presented in Figs. 2-5. Stability estimates and visual plots for all other inflammatory markers and sensitivity analyses are presented in the Supplement (Table S3, Fig. S5, Table S4, Table S5).

Fig. 1.

Stability Estimates and Confidence Intervals for Measurement of Immune Markers Across Time (in years).

Fig. 2.

Forest Plots for CRP by Time Interval.

Fig. 5.

Forest Plots for fibrinogen by Time Interval.

3.2. CRP

Overall, CRP demonstrated decreasing stability estimates with increasing time interval (r = 0.79 [24 h – 3 months] → r = 0.35 [7 + years]; Table 1, Fig. 1). Strong stability was observed for intervals <3 months, while there was modest stability for intervals 6 months – 3 years, and low stability for intervals 5 years or greater (Table 1, Fig. 1). Fig. 2 shows the forest plots of weighted effect sizes and confidence intervals for each study contributing to the stability estimate for each interval. The stability estimates were all relatively precise, as indicated by relatively small 95 %CI’s (Table 1, Fig. 2). Significant heterogeneity in the overall effect size estimate was observed for all intervals (Table 2, Fig. 2), indicating that observed variation was primarily due to between-study differences. Results were similar when studies that were rated as “poor” quality were removed from the analyses (Table S4).

Table 2.

Tests for Heterogeneity and I² Values for Stability Estimates of Immune Markers Across Time Intervals.

Biomarker	Interval	Cochran’s Q	p-value	I2
CRP	1 day-3mo	102.2	<0.001	93.2 %
CRP	6mo-1 yr	528.8	<0.001	98.5 %
CRP	3 years	694.2	<0.001	98.1 %
CRP	5 years	412.8	<0.001	96.1 %
CRP	7 + years	599.1	<0.001	98.3 %
IL-6	1 day-3mo	83.8	<0.001	91.7 %
IL-6	6mo-1 yr	231.8	<0.001	96.5 %
IL-6	3 years	104.37	<0.001	92.3 %
IL-6	5 – 7 + yrs	202.7	<0.001	95.1 %
TNF-α	1 day-3mo	174.4	<0.001	97.7 %
TNF-α	6mo-1 yr	81.4	<0.001	95.1 %
TNF-α	3 years	235.33	<0.001	98.7 %
TNF-α	5 – 7 + yrs	17.1	<0.001	94.2 %
Fibrinogen	1 day-3mo	19.5	<0.001	94.9 %
Fibrinogen	6mo-1 yr	58.1	<0.001	94.8 %
Fibrinogen	3 years	24.82	<0.001	96.0 %
Fibrinogen	5 – 7 + yrs	14.5	0.0129	65.4 %

Notes: CRP = C-reactive protein; IL = interleukin; TNF = tumor necrosis factor; I² = heterogeneity statistic; [ ] = 95% confidence interval; Q = Cochrane’s Q test.

3.3. IL-6

The pattern of results for IL-6 were broadly similar to those of CRP, with decreasing stability estimates with increasing time interval (r = 0.61 [24 h – 3 months] → r = 0.39 [5 + years]; Table 1, Fig. 1). These estimates reveal moderate stability for intervals <3 months, modest stability for intervals 6 months – 3 years, and low stability for intervals ≥ 5 years (Table 1, Fig. 1). There was significant heterogeneity in the stability estimate for all intervals (Table 2), suggesting that variation around these estimates was primarily due to between-study differences.

Fig. 3 shows the forest plots of weighted effect sizes and confidence intervals. In general, there were larger 95 %CI’s for shorter time intervals (24 h – 3 months, 6 months – 1 year), suggesting less precision in the stability estimates across shorter time periods (Table 1, Fig. 3). Removing studies rated as “poor” quality resulted in greater precision in the estimate for these time intervals, and the overall pattern of results was similar (Table S4).

Fig. 3.

Forest Plots for IL-6 by Time Interval.

3.4. TNF-α

Overall, TNF-α demonstrated strong stability over intervals <3 months and intervals of 1 year to 3 years, moderate stability over intervals 6 months – 1 year, and low stability over intervals >5 years (r = 0.80 [24 h – 3 months] → r = 0.30 [5+ years]; Table 1, Fig. 1). Fig. 4 shows the forest plots of weighted effect sizes and confidence intervals for each study contributing to the stability estimates for TNF-α. There was significant heterogeneity in the stability estimate for all intervals (Table 2), suggesting that variation around these estimates was primarily due to between-study differences. However, only the intervals <1 year had ≥ 5 studies contributing to the effect size estimates, suggesting that inferences over time intervals longer than 1 year may not be reliable (Lipsey and Wilson, 2001). Removing studies rated as “poor” quality revealed a similar pattern of results (Table S4); however, after removal of “poor” quality studies there were no intervals with ≥ 5 studies contributing to the stability estimate.

Fig. 4.

Forest Plots for fibrinogen by Time Interval.

3.5. Fibrinogen

For fibrinogen, strong stability was observed across intervals <3 months, while there was moderate stability for intervals 6 months – 1 year, and modest stability for intervals 3 years or greater (r = 0.88 [24 h – 3 months] → r = 0.53 [5+ years]; Table 1, Fig. 1). However, only the 5+ year time interval had ≥ 5 studies contributing to the effect size estimate; thus, inferences for time intervals other than 5+ years may not be reliable (Lipsey and Wilson, 2001). Removing studies rated as “poor” quality resulted in a similar pattern of results, although these estimates were based on even smaller sample sizes (Table S4).

Fig. 5 shows the forest plots of weighted effect sizes and confidence intervals for each study contributing to the stability estimates for fibrinogen. Of note, the 5+ year interval had reduced heterogeneity (Table 2), suggesting that the estimates for this time period were more consistent across studies.

3.6. Other inflammatory markers

Other markers of inflammation that were reported across studies included IL-8, IL-1β, IFN-γ, IL-10, IL-IRα, IL-2, IL-4, IL-5, sTNF-RII, IL-12, IL-13, and IL-1α (Table S3, Table S4, Fig. S5). Of these, IL-1β and IL-8 had the broadest coverage with the greatest number of studies contributing to each interval. However, the only interval with ≥ 5 studies contributing to the estimate was the 1 day – 3 month interval for IL-1β (r = 0.62; Table S3). Although we calculated an effect size for all possible intervals (Table S3, Fig. S5), these results should be interpreted with caution given that they are based on <5 independent studies (Lipsey and Wilson, 2001).

3.7. Assessment of moderators

There was insufficient data to conduct statistical tests of moderation for any of the proposed moderators (Table S2, Table S6, Table S7). However, we were able to qualitatively examine the frequencies of reported moderators for CRP and IL-6 (Table S6, Table S7), and we generated preliminary estimates of stability for these markers by assay type, sample matrix, and fasting status (Tables 3a-d). Time of day of blood draw was confounded with fasting status, with studies coded as fasting also coded as having blood drawn in the morning. Thus, we did not perform moderator analyses by time of day of blood draw. Only estimates containing ≥ 5 studies per interval are interpreted.

Table 3a.

Stability Estimates by CRP Assay Type.

CRP Assay Type	Interval
	1 day – 3 months	6 months – 1 year	3 years	5 years	>7 years
ELISA (single analyte)	r = 0.658*	r = 0.654*	r = 0.392		r = 0.510*
	[0.237, 0.871]	[0.147, 0.889]	[0.259, 0.511]		[0.430, 0.582]
	k = 2	k = 2	k = 6		k = 1
ELISA (Multiplex)			r = 0.766*		r = 0.383*
			[0.475, 0.906]		[−0.086, 0.713]
			k = 2		k = 1
Immunoturbidometric assay	r = 0.940*	r = 0.290*	r = 0.210*	r = 0.230*	r = 0.354
	[0.902, 0.964]	[0.232, 0.346]	[−0.092, 0.498]	[0.162, 0.286]	[0.181, 0.506]
	k = 1	k = 1	k = 1	k = 1	k = 6
Latex aggregation/Latex-enhanced immunoturbidometric assay	r = 0.904*	r = 0.211*	r = 0.472*	r = 0.416*	r = 0.187*
	[0.521, 0.984]	[0.182, 0.240]	[−0.139, 0.823]	[−0.137, 0.772]	[0.161, 0.213]
	k = 2	k = 1	k = 2	k = 2	k = 3
Immunonephelometric assay		r = 0.620*	r = 0.646*	r = 0.457	r = 0.374*
		[0.599, 0.641]	[0.549, 0.726]	[0.320, 0.575]	[0.181, 0.539]
		k = 1	k = 2	k = 6	k = 3
Radial immunodiffusion	r = 0.750*
	[0.682, 0.805]
	k = 1
Bead-based immunoturbidometric assay (but not latex beads)				r = 0.320*
				[0.289, 0.350]
				k = 1

Notes: CRP = C-reactive protein; ELISA = enzyme-linked immunosorbent assay; r = correlation between measurements or stability in these measurements across this interval; [ ] = 95 % confidence interval; k = # of independent study samples; n/a = no data available.

^*Estimate is based on <5 studies and thus warrants less confidence (Lipsey and Wilson, 2001).

Table 3d.

Moderator Analyses for CRP, IL-6 by Fasting Status.

Biomarker	Fasting	Interval
		1 day – 3 months	6 months – 1 year	3 years	5 years	>7 years
CRP	No		r = 0.435*	r = 0.632*	r = 0.359*	r = 0.199*
			[0.200, 0.623]	[0.316, 0.822]	[0.046, 0.608]	[0.175, 0.222]
			k = 1	k = 2	k = 2	k = 2
	Yes	r = 0.725*	r = 0.511	r = 0.415	r = 0.343	r = 0.432
		[0.629, 0.800]	[0.268, 0.693]	[0.223, 0.575]	[0.222, 0.454]	[0.272, 0.569]
		k = 4	k = 5	k = 7	k = 9	k = 5
IL-6	No		r = 0.787*	r = 0.727*	r = 0.551*
			[0.399, 0.936]	[0.407, 0.888]	[0.229, 0.764]
			k = 3	k = 3	k = 1
	Yes	r = 0.539*	r = 0.499*	r = 0.444*	r = 0.371
		[0.472, 0.600]	[0.267, 0.676]	[0.386, 0.498]	[0.263, 0.470]
		k = 3	k = 3	k = 2	k = 7

Notes: CRP = C-reactive protein; IL = interleukin; r = correlation between measurements or stability in these measurements across specified interval; [ ] = 95 % confidence interval; k = # of independent study samples.

^*Estimate is based on <5 studies and thus warrants less confidence (Lipsey and Wilson, 2001).

For CRP, a majority of studies reported that they used immunonephelometric assays (k = 7/37) and immunoturbidometric assays (k = 7/37) to assess levels in serum (k = 21/37), in the morning hours (k = 15/37), and among patients who were fasting (k = 16/37; Table S2). Other commonly used assays included latex-aggregation or latex-enhanced immunoturbidometric assays (k = 6/37) and single analyte ELISAs (k = 5/37), with only one study assessing levels of CRP using radial immunodiffusion (k = 1/37) and one study using a bead-based assay other than latex (k = 1/37; Table S2). Stability estimates by assay type for CRP are presented in Table 3a. For intervals with ≥ 5 studies, when compared with the overall findings, studies using a single analyte ELISA across 3 years yielded a lower stability estimate (r = 0.392 vs r = 0.530 95 %CI[0.399, 0.639]), studies using immunonephelometric assay across 5 years yielded a higher stability estimate (r = 0.457 vs r = 0.355 95 %CI [0.285, 0.420]), and studies using immunoturbidometric assay across 7+ years were similar (r = 0.354 vs r = 0.347 95 %CI[0.234, 0.450]; Table 3a). Stability of CRP did not differ significantly as a function of sample matrix (i.e., plasma vs serum; Table 3c, Fig. S6) or whether individuals fasted or not prior to the blood draw (Table 3d, Fig. S6).

Table 3c.

Moderator Analyses for CRP, IL-6 by Sample Matrix.

Biomarker	Sample Matrix	Interval
		1 day – 3 months	6 months – 1 year	3 years	5 years	>7 years
CRP	Plasma	r = 0.827*	r = 0.494	r = 0.416	r = 0.443	r = 0.351
		[0.584, 0.934]	[0.312, 0.641]	[0.314, 0.508]	[0.302, 0.566]	[0.227, 0.463]
		k = 4	k = 6	k = 9	k = 7	k = 8
	Serum	r = 0.746*	r = 0.516*	r = 0.735*	r = 0.343	r = 0.324*
		[0.472, 0.889]	[0.233, 0.718]	[0.395, 0.898]	[0.231, 0.446]	[0.046, 0.556]
		k = 4	k = 3	k = 2	k = 7	k = 3
IL-6	Plasma	r = 0.553*	r = 0.670	r = 0.529	r = 0.351
		[0.449, 0.642]	[0.479, 0.800]	[0.457, 0.595]	[0.209, 0.480]
		k = 3	k = 7	k = 8	k = 8
	Serum	r = 0.548*	r = 0.278*	r = 0.660*
		[0.152, 0.793]	[−0.196, 0.646]	[0.505, 0.774]
		k = 4	k = 2	k = 1

Notes: CRP = C-reactive protein; IL = interleukin; r = correlation between measurements or stability in these measurements across specified interval; [ ] = 95 % confidence interval; k = # of independent study samples.

^*Estimate is based on<5 studies and thus warrants less confidence (Lipsey and Wilson, 2001).

For IL-6, a majority of studies reported that they assessed levels using a single analyte ELISA (k = 14/22), using serum samples (k = 16/22), collected in the morning hours (k = 8/22), from individuals who were fasting (k = 8/22; Table S2). IL-6 measurements using a multiplex assay were generally more stable across time than measurements using a single analyte ELISA (Table 3b). There were no clear differences in stability of IL-6 by sample matrix (i.e., plasma vs serum; Table 3c, Fig. S6). However, samples measured from individuals who had fasted were generally less stable across time than those drawn from individuals who did not fast (Table 3d, Fig. S6).

Table 3b.

Moderator Analyses for IL-6 by Assay Type.

Biomarker	Assay Type	Interval
		1 day – 3 months	6 months – 1 year	3 years	5 years
IL-6	ELISA (single analyte)	r = 0.660*	r = 0.553	r = 0.427*	r = 0.318
		[0.456, 0.799]	[0.382, 0.687]	[0.380, 0.471]	[0.210, 0.419]
		k = 4	k = 5	k = 4	k = 9
	ELISA (Multiplex)	r = 0.562*	r = 0.659*	r = 0.707	r = 0.643*
		[0.107, 0.822]	[0.170, 0.888]	[0.514, 0.832]	[0.316, 0.833]
		k = 4	k = 4	k = 5	k = 2

Notes: IL = interleukin; r = correlation between measurements or stability in these measurements across specified interval; [ ] = 95 % confidence interval; k = # of independent study samples.

^*Estimate is based on<5 studies and thus warrants less confidence (Lipsey and Wilson, 2001).

4. Discussion

Single measurements of inflammatory biomarkers such as CRP, IL-6, TNF-α, and fibrinogen are often used in PNI research as markers of risk for diseases with inflammatory pathophysiology among otherwise healthy individuals (e.g., CHD and cancer; Dibaba et al., 2019; Guo et al., 2013; Kannel et al., 1987; Kaptoge et al., 2014; Zhang et al., 2018). However, to date, the temporal stability of these peripheral markers is poorly understood and has not been widely investigated or accounted for in the prediction of health risk. Thus, the current review sought to quantitatively investigate the stability of peripheral markers of inflammation among healthy individuals when assessed on multiple occasions separated by at least 24 h. Our hypothesis that most inflammatory markers would demonstrate modest to moderate stability over short periods of time was broadly supported, with temporal stability in the strong to moderate range for time periods <6 months. We also posited that temporal stability would decrease over time, with higher concordance when measured over weeks/months compared to years. Our results supported this prediction; most inflammatory markers measured over periods 1 day to 3 months had strong stability estimates (r >0.75), estimates measured over time periods 6 months – 1 year and 3 years were moderate (r = 0.60-0.75) to modest (r = 0.50-0.60), and estimates of stability for 5 years or greater were generally low (r <0.50). This suggests that inflammatory markers measured on only one occasion may represent an adequate index of individual risk over time periods up to 6 months, while prediction of risk across intervals >6 months to 3 years may require repeated assessment. Our results also suggest that in order to accurately characterize individual risk over intervals >3 years, repeated measurements are absolutely necessary.

Markers that were commonly assessed across studies that also had broad coverage across most time intervals included CRP, IL-6, TNF-α, and fibrinogen. In general, CRP and IL-6 demonstrated decreasing stability with increasing time interval, with moderate to strong stability estimates over intervals ≤ 3 months (CRP: r = 0.79; IL-6: r = 0.61), modest to moderate stability estimates over time periods 6 months to 3 years (CRP: r’s = 0.53-0.51; IL-6: r’s = 0.60-0.57), and low stability estimates thereafter (≥5 years: CRP: r’s = 0.36-0.35; IL-6: r = 0.39). TMF-α similarly demonstrated strong stability over intervals ≤ 3 months (r = 0.80), modest stability over time periods 6 months to 1 year (r = 0.60), and low stability estimates over time periods >5 years (r = 0.30). Inconsistent with other findings, the stability of TNF-α over 3 years was strong (r = 0.82), although this estimate was based on <5 studies and should be interpreted with caution. Fibrinogen also followed the same pattern of results, although the estimate for 5+ years was in the modest range (r = 0.53); notably, the estimate for this interval was the only one for fibrinogen based on at least 5 independent studies. Other commonly assessed inflammatory markers included IL-1β. IL-1β showed modest to moderate temporal stability across all intervals ≤ 3 years; however, the only stability estimate based on >5 studies was the 1 day – 3 month time period. Thus, these results are less reliable and warrant replication. In general, however, the results for IL-1β followed the same pattern for the commonly measured markers (i.e., CRP, IL-6, TNF-α, and fibrinogen), with single assessments strongly predicting measures taken 1 day to 3 months later. In addition, our findings suggest that TNF-α may be equivalent in terms of stability over time to other commonly measured markers of inflammation in PNI (e.g., IL-6, CRP). Given growing evidence for TNF-α as an independent predictor of cardiovascular disease (Subirana et al., 2018; Yuan et al., 2020), researchers may want to consider including TNF-α as a marker of peripheral inflammation in future studies. Indeed, whereas IL-6 has both pro- and anti-inflammatory signaling properties and is modulated by many factors (e.g., acute physical activity), TNF-α has a primary pro-inflammatory function and responds more selectively to acute threat (Del Giudice and Gangestad, 2018).

Our findings suggest that the most commonly measured inflammatory biomarkers used in the prediction of incident disease risk, CRP and IL-6, demonstrate strong stability over intervals <6 months, modest to moderate stability over periods of >6 months to 3 years, and lower stability over periods >3 years. These estimates are consistent with findings from large epidemiologic samples (e.g., Al-Delaimy et al., 2006; Clendenen et al., 2010; Danesh et al., 2004; Engelberger et al., 2015; Gu et al., 2009; De Bacquer et al., 1997; Rosenson et al., 1994; Sakkinen et al., 1999; Clendenen et al., 2010; Epstein et al., 2013; Engelberger et al., 2015; Koelman et al., 2019; Hofmann et al., 2011). Importantly, our findings extend evidence of biomarker stability taken from epidemiologic samples to include data from additional studies that include a diverse range of demographic factors, including age, sex, and geographic location, as well as a number of different sampling methods. In addition, these findings are similar to a seminal reference paper for the stability of CRP used by the AHA and CDC to make clinical recommendations, which measured stability of CRP up to 6 months and found these estimates to be strong (Macy et al., 1997; see in Pearson et al., 2003; Myers et al., 2004). Our estimates of the temporal stability of CRP and IL-6 are also similar to those reported for common cardiovascular risk factors such as blood pressure and lipid levels. For example, stability estimates of resting blood pressure over 2 to 12 years range from 0.61 to 0.35 for systolic blood pressure (Hottenga et al., 2006; Jackson et al., 2015; Katzmarzyk et al., 2000), with slightly lower estimates of stability for diastolic blood pressure (Hottenga et al., 2006; Katzmarzyk et al., 2000). Similarly, stability estimates for measures of blood lipids over 4 to 12 years range from 0.65 to 0.37 for both low-density lipoproteins (LDL) and triglycerides (Jackson et al., 2015; Katzmarzyk et al., 2000). Thus, our findings are not only consistent with stability estimates of inflammatory biomarkers from large epidemiologic samples and clinical findings for CRP, they also approximate stability estimates for blood pressure and lipid levels over the same time period(s).

Our findings of modest to moderate stability for most markers of inflammation across intervals >6 months and <3 years suggest that repeated measurement of inflammatory biomarkers over these intervals may be necessary to accurately characterize health risk, while our findings of low stability over intervals >3 years suggest that repeated measurements are absolutely necessary. While experts currently do not agree on the frequency with which CRP measurements should be repeated clinically, current recommendations indicate that two measurements should be taken 2 weeks apart to determine 10-year CHD risk (Pearson et al., 2003). Our results suggest that healthy individuals may require repeated assessments of inflammatory biomarkers anywhere from 6 months to 3 years apart to accurately estimate future health risk. This recommendation is similar to clinical guidelines for repeated assessment of other markers of cardiovascular disease risk. For example, clinical guidelines for blood pressure suggest reassessment every 1 – 2 years for healthy adults (Chobanian et al., 2003; Flack and Adekola, 2020; Takahashi et al., 2012). Furthermore, repeated measurement of fasting lipid levels is recommended every 4–6 years among healthy adults, starting at age 20 (NHLBI, 2002). Thus, although future research is warranted to determine the optimal interval for repeated measurement of inflammatory biomarkers, our findings suggest that measures should be repeated at least every 3 years to accurately predict disease risk.

Although we collected data on a number of potential moderators of the stability of inflammatory biomarkers across time, these factors were not reported consistently enough across studies to quantify their contribution, even for the most commonly occurring biomarkers and time interval combinations (i.e., CRP, IL-6; Table S6, Table S7). Thus, we were not able to estimate how much variation in our effect sizes was due to between-subjects factors (e.g., age, sex, or fasting state of the participant) compared to analytic factors (e.g., differences in the matrix from which the analytes are measured; Engelberger et al., 2015). However, our statistical tests for heterogeneity suggest that for a majority of estimates, heterogeneity was primarily due to between-subjects factors (Table 2). Of note, sensitivity analyses that excluded studies rated as “poor” quality (primarily due to analytic factors) did not impact our primary findings (Table S4). This suggests that analytic variance contributed little to observed heterogeneity of stability estimates across studies. A majority of studies measured CRP using immunonephelometric and immunoturbidometric assays and IL-6 using single analyte ELISAs (Table S2). In addition, most samples for IL-6 and CRP were measured in in serum, in the morning, and under fasting conditions (Table S2). Although we were not able to perform statistical tests of moderation, we estimated preliminary effect sizes for the stability of CRP and IL-6 separately by assay type, fasting status, and sample matrix (plasma vs serum; Tables 3a-d). The data coverage was too sparse to compare CRP assay types to each other, although estimates with ≥ 5 studies per interval suggested that compared to our main findings, measurements using single analyte ELISA may underestimate stability measurements across 3 years, while measurements using immunonephelometric assays may overestimate stability across 5 years. For IL-6, results suggested that repeated measurements using multiplex assay were more stable across time than measurements using a single analyte ELISA, although this interpretation should be taken with caution given that a majority of these estimates were based on <5 studies per interval. For both CRP and IL-6, there was no clear difference in the temporal stability of measures between plasma and serum. Similarly, stability of CRP did not vary as a function of fasting status, which is consistent with prior literature (Myers et al., 2004). For IL-6, fasting measures appeared to be less stable over time than non-fasting measures. However, this observation should be interpreted with caution given the low number of studies included in this assessment and evidence that IL-6 varies as a function of diet (Zhou et al., 2010; Pearson et al., 2003). Indeed, future work is necessary to better understand within- and between-individual factors that may contribute to observed variance in estimates of stability.

4.1. Limitations

While there are a number of strengths to our current study and findings, they are not without limitations. Specifically, there were a small number of studies contributing to some of the inflammatory biomarkers and time interval combinations, leading to limited confidence in these combined effect sizes. In addition, a few of the effect size estimates were contained in wide confidence intervals suggesting low precision of the combined effect size estimate. We also observed significant heterogeneity across studies even when a large number of independent study results were included in the combined effect size. Although we collected information on a number of potential moderators, including between-subjects factors (measured or controlled at the study level) and analytic factors (Table S2, Table S6, Table S7), we were not powered to test whether these factors contribute to observed heterogeneity. Although we calculated preliminary stability estimates for CRP and IL-6 separately by assay type, fasting status, and sample matrix, most estimates were based on fewer than five studies and thus findings warrant replication before conclusions about best practices are drawn. In addition, future research should more thoroughly investigate factors that contribute to variation in temporal stability of inflammatory biomarkers.

4.2. Implications/Future directions

Our findings suggest that single measures of most inflammatory biomarkers provide a reliable index of stable individual differences in the short term (<6 months), with the stability of CRP and IL-6 over time periods ≥ 6 months and <3 years approximating stability of other widely used measures of cardiovascular risk (e.g., blood pressure, blood lipids) assessed across similar time periods. We also found similar estimates of stability for TNF-α over intervals of 1 day through 3 years. Given growing evidence that TNF-α may be an independent predictor of cardiovascular disease, this marker may be of interest to include in future studies in PNI.

Given evidence that momentary exposures (e.g., acute stress, pollution) impact circulating levels of inflammatory mediators, repeated assessments may provide a more reliable measure of stable individual differences. Thus, repeated measurement in the short term (days to weeks) may provide a more accurate measure of individual difference that shows greater stability over time and more accurately identifies individuals at future health risk (Danesh et al., 2008; Koenig et al., 2003). Further research is warranted to consider this possibility. The current findings suggest that, at a minimum, measurements of inflammatory biomarkers should be repeated every 3 years in the prediction of risk. These findings align with clinical recommendations for repeated measurement of other common cardiovascular risk markers (e.g., blood pressure, blood lipids) and may be more stringent than current clinical recommendations for repeated measurement of CRP; however, more work is necessary to characterize the optimal intervals for repeated measurement of each inflammatory marker. Finally, although we found significant heterogeneity in the effect size estimate for most inflammatory biomarkers and time intervals, we were underpowered to explore the contribution of the many contemporaneous factors (e.g., stress, mood, physical activity, pollution) that impact circulating levels of inflammatory mediators. Gaining an understanding of factors that contribute to this variance will likely improve our ability to accurately use circulating markers of inflammation in the prediction of disease risk among otherwise healthy individuals. Future work is also necessary to examine the stability of these markers among individuals with inflammatory-related diseases.

Data availability

Data will be made available on request.