Article Text
PDF
Abstract
Background Inflammation plays a central role in cardiometabolic disease and may represent a mechanism linking low socioeconomic status (SES) in early life and adverse cardiometabolic health outcomes in later life. Accumulating evidence suggests an association between childhood SES and adult inflammation, but findings have been inconsistent.
Methods We conducted a systematic review and meta-analysis of observational studies to quantify the association between childhood (age <18 years) SES and the inflammatory marker C reactive protein (CRP) in adulthood. Studies were identified in Medline and Embase databases, and by reviewing the bibliographies of articles published from 1946 to December 2015. Study-specific estimates were combined into meta-analyses using random-effects models.
Results 15 of 21 eligible studies (n=43 629) were ultimately included in two separate meta-analyses. Compared with those from the most advantaged families, participants from the least advantaged families had 25% higher CRP levels (ratio change in geometric mean CRP: 1.25; 95% CI 1.19 to 1.32) in minimally adjusted analyses. This finding was attenuated by the inclusion of adult body mass index (BMI) in adjusted models, suggesting BMI has a strong mediating role in CRP levels.
Conclusions We observed an inverse association between childhood SES and adulthood CRP, potentially mediated through BMI. Investigating how childhood SES is associated with childhood BMI and CRP would provide insight into the effective timing of social and clinical interventions to prevent cardiometabolic disease.
- Social and life-course epidemiology
- OBESITY
- Cardiovascular disease
- SOCIAL INEQUALITIES
Statistics from Altmetric.com
Introduction
Socioeconomic status (SES) is an important determinant of adult health outcomes.1 ,2 Early life SES in particular has been shown to predict risk factors and manifest disease in later life.3–6 Of the many health outcomes linked with early life SES, cardiometabolic conditions, including cardiovascular disease (CVD),7–9 diabetes10 ,11 and obesity,12 have shown the most consistent associations.
Chronic inflammation, indicative of prolonged low-grade activation of the immune system, is a suggested generic pathophysiological mechanism underlying the development of diverse non-communicable diseases,13 and especially of cardiometabolic health outcomes.14–16 Life course models17–21 posit that exposure to higher levels of inflammation, either during a critical period early in life or as an accumulation of a chain of risk factors over time, increase the likelihood of cardiometabolic disease later in life. In figure 1, we propose potential pathways through which low SES may act to increase inflammation, broadly corresponding to these alternate models. If SES in early life sets the stage for lifelong inflammation, then examining how intermediate variables alter associations between early life SES and adult inflammation may help delineate and quantify the direct and indirect pathways through which SES influences cardiometabolic health.
Directed acyclic graph outlining potential mechanisms through which childhood SES determines adult C reactive protein. Mechanisms broadly categorised into two life course models, the critical period model (A) and the chain of risk model (B).17 Model A1 (direct effect of childhood SES on adult CRP) posits a critical window early in life where later-life CRP is determined. Model B1 (indirect effect via adult SES), Model B2 (indirect effect via childhood BMI, and adult SES), Model B3 (indirect effect via childhood BMI to adult SES to adult BMI) and Model B4 (indirect effect via BMI in childhood and adulthood) posit that a chain of risk factors accumulate throughout life to increase CRP levels in adulthood. BMI, body mass index; CRP, C reactive protein; SES, socioeconomic status.
The cross-sectional association between SES and markers of inflammation in adulthood is well documented for a number of inflammatory biomarkers, including C reactive protein (CRP),19 ,22–24 fibrinogen23 ,25 ,26 and interleukin-6.19 ,24 ,27 CRP is the most commonly assessed marker as it is stable in stored samples and easy and inexpensive to measure; CRP has also been consistently associated with cardiometabolic outcomes in observational studies,14 but its role as a causal determinant of CVD is contested.28
Causal inferences have been challenging when examining the association between SES and CRP. Elevated levels of CRP may indicate an acute infection or injury in addition to chronic inflammation.29 Furthermore, measured and unmeasured confounding might bias cross-sectional analyses of SES and CRP. Finally, CRP may be a surrogate marker of poor health,13 which itself is responsible for lower adulthood SES, an example of reverse causation. Investigating temporally separate childhood SES and its relation to adult CRP provides evidence for the direction of a causal relationship between SES and CRP, if one exists.
Several studies have examined the impact of childhood SES on CRP and other inflammatory markers in adulthood.19 ,20 ,22 To the best of our knowledge, no previous systematic review has synthesised the available evidence examining childhood SES and later-life CRP. We therefore performed a systematic review and meta-analysis of the association between early life SES and subsequent adult markers of inflammation, with a focus on CRP, the most widely measured marker.
Methods
Protocol registration and information sources
The protocol was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO, CRD42016038683)30 on 4 May 2016. Searches of Ovid Medline and Embase databases were undertaken in December 2015 for studies reporting socioeconomic measures in childhood as exposures and blood biomarkers in adulthood as outcomes. The review was planned, conducted and reported in adherence to the standards of quality for reporting Meta-analyses of Observational Studies in Epidemiology (MOOSE).31
Literature search
The search strategy, developed together by a clinician and institution librarian, used subject headings and free text terms derived from previously identified relevant papers in authors' personal libraries.16 ,22 Keywords, MeSH terms and search limits for three topic areas—inflammation, SES and childhood—were combined in each database (see online supplementary table S1).
supplementary data
Study selection
Two investigators (KR, CG) independently screened abstracts and titles. Two investigators (RL, KR) further independently screened full texts according to inclusion criteria. Discrepancies were resolved through third-party adjudication (DB). Studies were included only if all of the following criteria were met: (1) they were observational studies of a general population, (2) childhood SES was reported as an exposure variable, and (3) adult inflammatory markers were reported as an outcome variable. We defined the exposure ‘childhood SES’ as a measure of parental or family occupational level, education, income or some other measure of SES, measured before 18 years of age. We defined ‘adult inflammatory marker’ as any blood-derived biomarker, measured after 18 years of age, associated with the inflammatory immune response. We specifically searched for studies containing the terms ‘C-reactive protein’, ‘interleukin’ and ‘fibrinogen’ in our search strategy.
All publication years from 1946 to 10 December 2015 were included. There was no minimum study size or follow-up duration, and no language restrictions. Bibliographies of eligible studies and previously published reviews were hand searched for additional relevant publications.
Data extraction
We narrowed our focus to CRP alone, as this marker was the most widely investigated. There were insufficient studies for other biomarkers to summarise the data meaningfully in meta-analyses. Two investigators (RL, BC) independently extracted two copies of the following data from the full texts: first author, year of publication, study population, sample size, age of participants at outcome, study duration, exposure and outcome methods, statistical analysis used, main estimates, standard errors (SEs) and confidence intervals (CI) estimates of the association between exposure and outcome, with covariates and stratification if applicable. We preferentially selected estimates that compared CRP between extreme categories of SES. In order to avoid overadjustment by potential intermediate variables between exposure and outcome, we preferentially extracted age-adjusted, sex-adjusted or minimally adjusted analyses. Where necessary, study authors were contacted for additional details.
Data synthesis and meta-analysis
To visualise overall trends, a summary table of results (positive association, no association, negative association) from age-adjusted and sex-adjusted analyses was compiled. Owing to heterogeneity in the reported analyses, only studies that reported β coefficients (or transformable variants) or odds/risk ratios were brought forward into meta-analyses (n=15). In studies that used CRP as a continuous variable, if possible, we converted analyses to a single type of statistic to ease comparison (n=9). Log-transformed β coefficients from linear regression became the baseline statistic into which all other analyses were transformed (see online supplementary tables S3–S6). In studies where CRP was treated as a binary variable (n=5), odds/risk ratios, comparing the odds for ‘high risk’ CRP in the low SES category with the high SES category, were extracted. Reciprocals of the ORs were calculated with the referent category as low SES. The random-effects model was used for all meta-analyses due to the high heterogeneity of the SES variables. When studies performed multiple relevant analyses, we included the most reliable measure in the following order: education preferably, and if not, occupation, and if not, income, and if not, miscellaneous exposures. This was to avoid overweighting studies that repeated measures on the same participants.
Risk of publication bias across studies was examined visually by funnel plots. A broadly symmetric plot indicated a lower risk of bias against the publication of negative results. We formally tested for asymmetry using Egger's test when >10 estimates were included in a single analysis.
Additional analyses
Post hoc stratified analysis of subgroups were undertaken to assess whether the association depended on (1) the study design and (2) the type of exposure. Further post hoc meta-analyses were undertaken excluding single studies with large weight. We additionally examined the influence of adult SES and adult body mass index (BMI) given their widely accepted relationships with adult inflammatory markers.
All statistical analyses were performed using Stata V.14.1 (StataCorp LP, Texas, USA) using meta-analyses commands such as metan, metafunnel and metabias.
Results
Study selection and characteristics
We screened the abstracts and titles of 1710 non-duplicate citations, excluding 1609 studies (figure 2). Of 101 studies that were potentially eligible on screening, 30 articles met all inclusion criteria. One additional study was included through a hand search of bibliographies of selected studies. We narrowed our focus to CRP alone, as other markers did not provide consistent evidence, and presenting such data may provide misleading conclusions. We excluded 10 studies and extracted data on childhood SES and adult CRP associations from 21 studies. Table 1 and online supplementary table S2 present the characteristics of each included study.
Summary of included paper characteristics related to C reactive protein
Flow chart detailing numbers of studies screened, assessed and included in the review, with reasons for exclusions.
Nine studies were performed in the USA, five in the UK, three in Nordic countries, one in New Zealand and three in low to middle income countries (Brazil and the Philippines). Five studies were cross-sectional32–36 and 16 were cohort studies with variable periods of follow-up ranging from one to five decades. Exposures included measures of parental occupational status (n=10), education (n=13), family income (n=3) and some measure of early life adversity (n=5). Seven studies excluded high CRP measurements (>10 mg/L), while Shanahan et al 29 defined >10 mg/L as the cut-off point for ‘very high CRP’.
Synthesis of results
Differences in analytical methods made direct comparison difficult, so we grouped the 21 studies examining CRP by the statistical methodology used in reporting associations (table 2).
Summary of study-specific unadjusted associations between childhood socioeconomic circumstance and adult C reactive protein levels, grouped by statistical method
The reported associations were classified according to the exposure used in analysis.
The majority of reported associations were inverse associations (participants with lower childhood SES had an increased CRP relative to those with a higher childhood SES). Three studies37–39 reported no significant associations between any measure of childhood SES and adult CRP, while further four studies23 ,34 ,40 ,41 reported at least one non-significant association. The remaining studies (n=14) reported significant associations between all measures of childhood SES and adult CRP.16 ,19 ,20 ,29 ,32 ,33 ,35 ,36 ,42–47
Meta-analysis
Fifteen studies had sufficient data for inclusion of study-specific estimates into two separate meta-analyses. Eight studies19 ,20 ,23 ,33 ,35 ,38 ,41 ,45 reported linear regression analyses of log-transformed CRP (figure 3). Two further studies32 ,37 reported geometric means. β coefficients estimating the difference between extreme groups in an unadjusted analysis were derived from the Carmelo study32 (see online supplementary table S5). The data presented in Gimeno et al 37 were discarded as the estimates were from the same study as Kivimaki et al.23 Thus, the first meta-analysis comprised nine studies of 24 934 participants (figure 3). Repeated analyses of the same population were excluded, in order of preference: income data excluded if occupation data were available, and occupation data excluded if education data were available. A sensitivity analysis excluding childhood adversity study-specific estimates was not substantially different (see online supplementary figure S1).
Unadjusted or minimally adjusted results from stratified meta-analysis of the association between childhood socioeconomic circumstance and adult geometric mean CRP levels (random-effects model), according to study design. Relative change in CRP comparing the lowest SES category with the highest SES category, according to either parental, maternal or paternal occupational rank, education level or childhood adversity. Solid diamonds represent study-specific point estimates; squares represent study-specific weights in stratified and final meta-analyses; horizontal bars represent 95% CIs, and hollow diamonds represent summary estimates, with the width of the diamond corresponding to the 95% CI of the summary. Income-based estimates were excluded if occupation-based estimates were available from the same study, and occupation estimates excluded if education estimates available, to avoid overweighting papers that conducted multiple analyses on the same population. Meta-analysis stratified according to study design and whether parental SES data were collected longitudinally or retrospectively. CRP, C reactive protein; SES, socioeconomic status.
Overall, there was a 25% increase (ratio change in CRP: 1.25, 95% CI 1.19 to 1.32) in adult geometric mean CRP in the lowest childhood SES group compared with the highest, measured by either parental education or occupational status, in a random-effects model. Sensitivity analyses restricted to either education or occupation subgroups did not yield substantially different summary estimates (see online supplementary figure S1).
Five further studies16 ,29 ,39 ,43 ,44 were grouped into a second meta-analysis of odds/risk ratios (n=18 695) (see online supplementary figure S2). The odds or risk for a high-risk CRP compared with a low-risk CRP was 23% higher in the low SES category compared with the high SES category (1.23; 95% CI 1.11 to 1.37). Post hoc sensitivity analysis excluding high-weighted estimates from the Shanahan study29 did not substantially alter the results (data not shown).
Publication bias was assessed via funnel plots separately for each meta-analysis. Visual inspection of the linear regression estimates funnel plot did not indicate the presence of small-study effects, and Egger's regression asymmetry test did not suggest significant small-study effects (p=0.3) (see online supplementary figure 3A). Visual inspection of the second meta-analysis funnel plot should be interpreted with caution given the limited number of studies included (n=5), but it did not demonstrate obvious bias (see online supplementary figure 3B).
Explanatory factors
Sex differences in C reactive protein
CRP, when examined by sex, was higher in women than in men. Two studies reported significant sex differences20 ,41 in initial analyses and stratified by sex in reported analyses. In particular, Nazmi et al 41 observed differing trends with SES and CRP for men and women, likely related to culture-specific and sex-specific SES–BMI relationships in Brazil.48
Body mass index
In four studies,19 ,23 ,38 ,41 the addition of adult BMI into the analysis as a mediator variable fully attenuated the association between childhood SES and adult CRP. Post hoc meta-analysis of BMI-adjusted associations between childhood SES and adult CRP yielded an estimate close to 1 (1.04, 95% CI 0.93 to 1.13) (see online supplementary figure S4). In separate logistic regression analysis, Gustafsson et al 44 (Umeå University, personal written communication 2016) demonstrated an attenuation of the OR towards the null hypothesis once adult BMI was included in the model. Path analysis in two further studies36 ,47 demonstrated strong associations between childhood SES and adult BMI, which in turn predicted adult CRP.
Adult socioeconomic status
We additionally examined whether childhood SES had an association with adult CRP independent of adult SES in a post hoc meta-analysis (see online supplementary figure S5). The log CRP ratio between highest and lowest SES groups was 1.13 (95% CI 1.04 to 1.23), suggesting that other residual confounding factors not accounted for by adult SES may also mediate this association.
Discussion
This meta-analysis of nine population-based cohort and cross-sectional studies showed that low childhood SES, predominantly quantified by parental education and occupation in minimally adjusted linear regression models, was associated with a moderately increased CRP level in adulthood. The geometric mean CRP level was 25% higher in those with the lowest SES in childhood, compared with the referent group of high SES. This finding was supported by a second meta-analysis of five additional studies which reported a moderate (23%) increase in the odds for ‘high risk’ CRP levels in those with low SES in childhood, compared with the referent highest SES category.
To the best of our knowledge, our systematic review is the first to report an association between childhood SES and adulthood CRP. One previous systematic review22 of observational studies, published through to 2006, reported cross-sectional associations between adult SES and CRP, but inconsistent evidence of an association between childhood SES and adult CRP. We report a more consistent relationship between childhood SES and adulthood CRP, likely reflecting inclusion of published studies in the last decade.
The clinical significance of our finding is imperfectly estimated by drawing parallels to the effect of statin therapy on adult CRP. An overall 25% decrease in CRP level between high and low SES groups found in our meta-analysis broadly corresponds to the expected anti-inflammatory effects from statin therapy. Mainly used to lower low-density lipoprotein (LDL) cholesterol, statin therapy also lowers CRP by around 17–43% in healthy adults.49–51 However, we should interpret this finding cautiously, as the causal role of CRP in CVD pathogenesis is contested.28 Consistent observational data demonstrates a strong association,52 but Mendelian randomisation studies suggest CRP is a consequence of confounding factors such as obesity, rather than a causal factor in CVD pathogenesis.28
Evidence for life course models
Our results support the chain of risk (indirect effects) model (pathway B3 or B4, figure 1), where childhood SES influences BMI to increase CRP in adulthood. The most compelling evidence for this is reported by Nazmi et al,41 a cohort study of Brazilian young adults. They demonstrate differing social gradients in CRP between men and women, dependent on the measure of SES. Higher family income was associated with higher CRP in men, but not in women. In contrast, higher maternal education was associated with lower CRP in women, but not men. They concluded that the social patterning of obesity influenced the association between childhood SES and adult CRP.41 Two further studies36 ,47 support this by using path analysis to demonstrate a significant role for BMI in the pathway between childhood SES and adult CRP. Comparison of the original meta-analysis of minimally adjusted associations between childhood SES and adult CRP (figure 3) with a post hoc meta-analysis of BMI-adjusted associations (see online supplementary figure S4) resulted in an attenuation of the summary estimate. This suggests that adult BMI is a plausible mediator of the association between childhood SES and adult CRP. Furthermore, analysis of the US National Health and Nutrition Examination Survey (NHANES) suggests that the inverse association between parental measures of SES and CRP in childhood is partly mediated by childhood BMI.53 This is in keeping with our proposed model where low SES affects childhood BMI and hence adult BMI, leading to increased adult CRP (pathway B4, figure 1). Few studies had additional childhood BMI data, precluding comment on the contribution of childhood or lifetime trajectories of BMI to adult BMI and inflammatory outcomes.
Alternative hypotheses may include a pathway through adult SES, in which one would expect attenuation of an association when analyses include adult SES as a covariate. We did not find evidence for this in the limited meta-analysis of adult SES-adjusted linear regression models (see online supplementary figure S5).
Plausible biological mechanisms
CRP is strongly associated with obesity and the relationship is complex. Mendelian randomisation studies have suggested that genetic variants associated with BMI are determinants of CRP levels,54 ,55 rather than the converse. This implies obesity leads to higher CRP, and cytokines from adipose tissue plausibly contribute to chronic inflammation.56 Conversely, ex vivo immunological data suggest lower childhood SES is associated with greater immune dysregulation,57 which may prime for heightened inflammatory responses to subsequent stimuli, such as infection.58 Epidemiological evidence suggests childhood infections are associated with increased BMI and other cardiometabolic outcomes, but only in those with low family incomes.59 ,60 A dysregulated host inflammatory response to infection may contribute to pathology in a number of key stages in atherosclerotic disease61 and lipid metabolism.62 The causal relationship between obesity and inflammation is unlikely to be unidirectional, and the mechanisms suggested are not mutually exclusive.
Strengths and limitations
Strengths of the current review are the consistency of results across two methodologies, over multiple exposure measures and in a number of different populations. Most studies (18/21) found an association between childhood SES and adult CRP. Only one study from those reporting any association found an inverse association between childhood SES and adult CRP.
Residual confounding is a concern in meta-analyses of observational studies. The strength of this review in examining childhood exposures and adult outcomes means most traditional ‘confounders’ will lie temporally between the exposure and outcome as intermediate variables. Adjustment for such variables in this situation is typically considered overadjustment and will increase bias towards the null hypothesis,63 but residual confounding through grandparental factors, for example, may still bias our result in an unpredictable manner.
Heterogeneity in study-specific analyses prevented inclusion of all studies into a meta-analysis, introducing a potential source of selection bias where only favourable studies with easily extractable analyses were included. We addressed this in two ways—first by contacting authors for further data where possible, and second by examining asymmetry in funnel plots and performing tests for publication bias where appropriate, although the small number of studies means that publication bias is difficult to assess. Replication of our findings in other cohorts is warranted.
Future directions
Understanding the determinants and effects of each component of childhood SES, including parental education, income and occupation, may provide novel targets for interventions. We report that parental occupation status and education were associated with adult CRP. Parental education, a widely used measure of SES, reflects a family's ability to improve health literacy64 and is reflected in rates of adverse health behaviours such as smoking, alcohol consumption and sedentary behaviour, all risk factors for cardiometabolic disease. Parental occupation directly impacts on material resources to the family, and also represents relative social standing, social networks and work-related stresses and autonomy.64 Differences in social health behaviours may provide a strong explanatory factor in socioeconomic health disparities, but should be interpreted within the appropriate cultural context—the social patterning of smoking, for example, is not as strong in France compared with the UK, and consequently contributes less to the social gradient in all-cause mortality.3
In conclusion, we report strong evidence of an association between low childhood SES and increased CRP in adulthood. There is support for a chain of risk (indirect effects) model where the association between childhood SES and adult CRP is mediated through adult BMI. Extending this model to other socially patterned health behaviours may increase understanding of the underlying mechanisms and focus research into novel interventions at an earlier age; early interventions may reduce cumulative exposure to low lifetime SES, or break the chains of risk originating from childhood.
What is already known on this subject
The mechanisms underlying the relationship between socioeconomic status and diverse health outcomes are unclear. Chronic inflammation is implicated in the pathogenesis of multiple non-communicable diseases, many of which are postulated to have their origins in early life. To better understand how socioeconomic status leads to inflammation-related chronic disease, we sought to quantify the association between childhood socioeconomic status and adult C reactive protein, a marker of chronic inflammation.
What this study adds
This meta-analysis shows that lower childhood socioeconomic status, when measured by parental occupation or education, is associated with higher adult C reactive protein. This association is attenuated by obesity, indicating a potential mediating mechanism by which early socioeconomic status may influence later-life health outcomes.
Acknowledgments
The study authors would like to acknowledge Ms Poh Chua for her assistance with Ovid Medline and Embase databases in helping to devise literature searches for this systematic review.
References
Footnotes
Contributors RSL and DPB contributed to study conception and design, directed the study implementation, acquisition, analysis and interpretation of data, and drafted the initial manuscript and approved the final manuscript submitted. AEA and FKM contributed to study design, analysis and interpretation of data, and critically revised the initial manuscript and further drafts for important intellectual content, and approved the final manuscript submitted. BC, KR and CEG contributed to acquisition and analysis of data, drafting and reviewing the manuscript and further drafts, and approved the final manuscript as submitted. MJ and MW contributed to the analysis and interpretation of the data, critically revising drafts for important intellectual content, and approved the final manuscript as submitted.
Funding This work was supported by the Australian National Health and Medical Research Council (Postgraduate Scholarship 1114567 to RSL, Senior Research Fellowships 1046518 to MW and 1064629 to DPB, Early Career Fellowship 1037449 and Career Development Fellowship 1111160 to FKM); by the Center for Integrative Approaches to Health Disparities (P60MD002249 to AEA); NIH National Institute of Diabetes, Digestive, and Kidney Diseases (R01DK087864 to AEA); the Carolina Population Center and its NIH Center grant (P2C HD050924 to AEA); the Australian Government (an Australian Postgraduate Award to CEG); and the Murdoch Childrens Research Institute (Murdoch Childrens Research Institute Top Up Scholarship to CEG). Research at the Murdoch Childrens Research Institute is supported by the Victorian Government's Operational Infrastructure Program.
Disclaimer The funding bodies did not play any role in the study.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.