Interleukin-6, Diabetes, and Metabolic Syndrome in a Biracial Cohort: The Reasons for Geographic and Racial Differences in Stroke Cohort

The burden of type 2 diabetes in the U.S. disproportionately affects Black individuals, who are 60% more likely than non-Hispanic White adults to develop diabetes and twice as likely to die as a result of diabetes (1). This disparity contributes to disproportionately increased risks of heart disease (2), stroke (3), and end-stage kidney disease (4). Diabetes risk also differs by sex; Black women are more than twice as likely as White women and Black men are 52% more likely than White men to develop diabetes (5). Metabolic syndrome is characterized by a cluster of at least three conditions from among dysglycemia, abdominal obesity, dyslipidemia, and increased blood pressure. Metabolic syndrome increases the risk of diabetes (6), and as with diabetes, there are race–sex differences; non-Hispanic Black women and non-Hispanic White men are more likely than other race–sex groups to have metabolic syndrome (7).

Interleukin-6 (IL-6) is implicated in the pathogenesis of diabetes (8,9). A higher IL-6 level was consistently associated with greater risk of incident diabetes, with a 2020 meta-analysis of 15 published studies finding a 24% greater risk per 1-unit increment of log-transformed IL-6 (10). Adipose tissue is a major source of IL-6, produced in part by activated macrophages, which accumulate in response to the release of hormones, cytokines, and free fatty acids by adipocytes (11). In turn, IL-6 promotes further macrophage infiltration in adipose tissue, mediating the transition from acute to chronic inflammation (12), and induces insulin resistance (8,9). IL-6 increases with obesity (13); thus, this feedback loop is chronically activated in obesity (8) and may indicate a role of IL-6 in diabetes pathogenesis. It is not known whether a higher IL-6 level is a risk factor for future metabolic syndrome, although cross-sectional studies have reported higher IL-6 levels (14) with metabolic syndrome.

Compared with White individuals, Black individuals have higher circulating concentrations of some proinflammatory biomarkers, including IL-6, despite having lower visceral adipose tissue (15). Studies examining the relationship of IL-6 with diabetes incidence by race are limited; we are aware of only one prospective cohort study, which observed participants over 4.7 years and showed no difference in this association by race (16).

In a subcohort of the Reasons for Geographic and Racial Differences in Stroke (REGARDS) (17) study, we studied associations of baseline IL-6 with incident diabetes and metabolic syndrome. We sought to 1) quantify associations of baseline IL-6 with diabetes and metabolic syndrome and 2) identify if associations differed by race, sex, or central adiposity. We hypothesized that a higher IL-6 level would be positively associated with risks of diabetes and metabolic syndrome and that this relationship would differ by race, sex, and central adiposity.

REGARDS is a national prospective cohort study of 30,239 participants from the contiguous U.S. who self-identified as non-Hispanic Black or non-Hispanic White (17). Black adults and those residing in the southeastern stroke belt (18) of the U.S. (North Carolina, South Carolina, Georgia, Tennessee, Mississippi, Alabama, Louisiana, and Arkansas) (19) were oversampled. Participants age ≥45 years were recruited by telephone between 2003 and 2007 from commercially available contact information. They completed a computer-assisted telephone interview and a standardized in-home visit to collect demographic information, clinical measurements, and blood and urine samples (20). REGARDS contacted participants (or proxies) every 6 months to collect health information. Approximately 10 years after enrollment in 2013–2016, 16,150 participants had a second extended telephone interview and in-home visit in which similar data were collected. Institutional review boards of all participating institutions approved the study, and informed consent was obtained from all participants.

The REGARDS Biomarkers as Mediators of Racial Disparities in Risk Factors (BioMedioR) study was designed to study relationships of circulating biomarkers with racial disparities in hypertension and diabetes and is described in detail elsewhere (21). The BioMedioR prospective subcohort was selected from 13,912 REGARDS participants who underwent both in-home assessments and had data on diabetes and hypertension status at both visits. Differences in characteristics of those lost to follow-up in REGARDS (22.0% of White participants and 28.4% of Black participants) were previously found not to have contributed to attrition bias in estimating racial differences (22). BioMedioR included 4,400 participants with 1,100 each of randomly sampled Black men, Black women, White men, and White women. This design allowed excellent power, as evaluated through extensive simulations across multiple sampling scenarios, to detect race and sex interactions with biomarkers (21) and allowed laboratory assays to be completed with cost efficiency (i.e., measurements were not required in 13,912 to answer the research questions). As previously reported, associations of risk factors with incident hypertension and diabetes were similar in the BioMedioR sample and the larger REGARDS sample (21). Two separate analytic samples of BioMedioR participants with baseline IL-6 data were created to study diabetes and metabolic syndrome risk separately; we excluded participants with prevalent disease or who were missing criteria to determine disease status from respective samples.

Fasting blood was drawn from each participant at both visits, locally centrifuged for 10 min, and then sent overnight on ice to the central laboratory at the University of Vermont, where it was centrifuged at 30,000g, aliquoted, and stored at 193 K (20). At both visits, calorimetric reflectance spectrophotometry was used to measure glucose, HDL, and triglycerides with the Ortho Vitros Clinical Chemistry System 950 IRC (Johnson & Johnson Clinical Diagnostics). The HOMA for insulin resistance (HOMA-IR) was calculated as the product of glucose (mmol/L) and insulin (μU/mL) divided by 22.5, with insulin measured by two-step chemiluminescence immunoassay using the Roche Elecsys 2010 system (Roche Diagnostics) (23).

Baseline blood samples from the first in-home visit for BioMedioR participants were retrieved from the biorepository, and IL-6 was measured using ultrasensitive ELISA (V-PLEX Human Proinflammatory Panel I; Meso Scale Diagnostics, LLC, Rockville, MD). The laboratory analytic interassay coefficient of variation ranged from 6.2% to 9.3% at varying concentrations of control samples.

Diabetes was defined as fasting glucose ≥6.99 mmol/L, random glucose ≥11.10 mmol/L, or self-reported use of diabetes medications (24). Incident diabetes was identified in participants who did not have diabetes at the first in-home visit (2003–2007) but were classified as having diabetes at the second (2013–2016).

Metabolic syndrome was defined using the harmonized definition (25) requiring presence of three of the following five risk factors: dysglycemia (fasting glucose ≥5.55 mmol/L or self-reported diabetes medication use), elevated blood pressure (systolic ≥130 and/or diastolic ≥85 mmHg or self-reported antihypertensive use), elevated triglycerides (≥1.69 mmol/L or self-reported fibrate, niacin, or omega-3 fatty acid use), low HDL (<1.04 [men] or <1.29 mmol/L [women] or self-reported fibrate or niacin use), or central adiposity (waist circumference ≥102 [men] or ≥88 cm [women]). Incident metabolic syndrome was identified in participants without metabolic syndrome at the first in-home visit but classified as having metabolic syndrome at the second.

All covariates were from the baseline examination unless otherwise noted. Data were collected using validated tools and methods (17). Demographic variables included self-reported age (years), sex (female or male), and race (Black or White). Geographic region was based on self-reported home location and categorized as the stroke buckle (an area of the U.S. with high stroke mortality in the coastal plains of North Carolina, South Carolina, and Georgia), the stroke belt (an area of the U.S. with high stroke mortality not reaching the same level as that in the stroke buckle; includes nonbuckle parts of North and South Carolinas and Georgia plus Tennessee, Mississippi, Alabama, Louisiana, and Arkansas), or the rest of the country (18,19).

Socioeconomic variables were self-reported and included education (less than high school, high school, some college, or college graduate) and annual household income (<$20,000, $20,000–$35,000, $35,000–$75,000, or ≥$75,000).

Anthropometric variables included waist circumference (measured midway between the lowest rib and iliac crest), BMI (kg/m²), and systolic and diastolic blood pressures (mmHg), calculated as the mean of two resting measurements taken 5 min apart.

Lifestyle variables were self-reported and included tobacco use (current, past, or never) and duration, weekly exercise frequency (none, one to three times, or more than four times), and current alcohol use (none, moderate [one to seven (women) or one to 14 weekly drinks (men)], or heavy [more than seven (women) or >14 weekly drinks (men)]). Regular dietary intake over the previous year was estimated with Block 98 Food Frequency Questionnaires (26) and used to calculate total energy intake (kcal). Dietary pattern scores were defined by factor analysis of intake of 56 food groups (27); a pattern heavy in fried food, processed meats, added fats, egg products, and sweetened beverages similar to dietary patterns observed in the southern U.S. was deemed the southern dietary pattern. Southern diet was considered continuously in analyses, with a higher score indicating greater concordance with this pattern.

Survey weighting of all analyses was applied to account for sample selection, using a Taylor series as a finite population correction (21). In each analytic sample (for incident diabetes and incident metabolic syndrome), participant characteristics were tabulated by IL-6 quartile (28). Categorical variables were reported as percentage and continuous variables as mean (SD) or median (interquartile range [IQR]) based on skewness. IL-6 was log (base e) transformed when used as a continuous variable because of its skewed distribution.

The age-adjusted relative risk (RR) of incident diabetes and metabolic syndrome for Black compared with White participants was assessed for all participants in each analytic sample using Poisson regression with robust variance estimation (29).

Five Poisson regression models were used to calculate RRs of diabetes and metabolic syndrome for each quartile of IL-6, with the first quartile serving as reference, and per a 1-SD increment of log IL-6 (0.77 and 0.76 for the diabetes and metabolic syndrome analyses, respectively). Confounding variables were specified a priori for the association between IL-6 and diabetes and/or metabolic syndrome. Models 1–4 assessed groupings of related variables to provide detailed assessment of which confounding factors were most relevant. Model 1 included age, sex, race, and region. Model 2 included model 1 variables plus triglycerides and HDL. Model 3 included model 2 variables plus baseline tobacco use, tobacco pack-years, weekly exercise frequency, alcohol use, waist circumference, and BMI. Model 4 was our primary model of interest, which included model 3 variables plus education and annual household income. Model 5 included model 4 variables plus HOMA-IR. Across these models, participants missing covariate data were excluded. Visualizations were performed with restricted cubic splines using the Harrell method and model 4 covariates, with the RR expressed relative to the 12.5th percentile IL-6 (28).

Because diet data were frequently missing, a sensitivity analysis was designed to assess confounding by dietary factors. Here, only participants with complete dietary data were included in an analysis with model 4 variables and another with model 4 variables plus southern dietary pattern and total energy intake.

To determine whether there was effect modification by race, sex, or central adiposity (waist circumference ≥88 [women] or ≥102 cm [men]) on the association of log IL-6 with incident diabetes or metabolic syndrome, using the primary adjusted model of interest (model 4), interaction terms for each of these variables with IL-6 (log IL-6 ∗ sex, log IL-6 ∗ race, and log IL-6 ∗ central adiposity) were included in separate models. When testing modification by central adiposity, BMI was removed as a covariate. Interactions were considered significant if the P value for the cross-product term was <0.10 (more liberal than typical to allow interactions of interest with moderate effect to be detected). Poisson regression models were subsequently stratified by those variables showing significant interaction.

Analyses were conducted in R 4.2.0 (R Foundation for Statistical Computing, Vienna, Austria). The REGARDS protocol was approved by institutional review boards at all participating institutions. According to REGARDS policy, the aims and analysis plan were prespecified, reviewed, and approved by the REGARDS publications committee, which also reviewed the final manuscript and assured that the a priori plans were followed.

Supplementary Fig. 1 shows the inclusion flow diagram. Ultimately, 3,399 participants were included in the diabetes analysis; 460 of these (13.5%) developed incident diabetes. For the metabolic syndrome analysis, 1,871 participants were included; 439 (23.5%) had incident metabolic syndrome. Median (IQR) follow-up time was 9.5 (8.6, 9.9) years for both analytic samples.

Baseline characteristics of participants in the diabetes analytic sample by IL-6 quartile are listed in Table 1. Black participants comprised 32% of this sample and represented 55% of those in the highest quartile of IL-6. The mean (SD) BMI was 28.6 (5.5) kg/m², and waist circumference was 93.6 (14.9) cm; both increased with increasing IL-6 quartile, as did other covariates.

Baseline characteristics by IL-6 quartile in the metabolic syndrome analytic sample are listed in Table 2. Black participants comprised 32% of this sample and 57% of those in the highest quartile of IL-6. Mean (SD) BMI was 27.1 (4.7) kg/m², and mean waist circumference was 89.7 (14.5) cm; as in the diabetes sample, risk factor prevalence increased with increasing IL-6 quartile.

Supplementary Fig. 2 presents baseline log IL-6 distribution for both analytic samples by incident disease status. In the diabetes sample, median (IQR) baseline IL-6 was 0.79 (0.54, 1.24) pg/mL. In the metabolic syndrome sample, it was 0.73 (0.51, 1.12) pg/mL. For both analytic samples, median log IL-6 at baseline was higher in the group that went on to develop disease.

Supplementary Table 1 summarizes risk of incident diabetes or metabolic syndrome by race. In the diabetes analytic sample, 14% of participants developed incident diabetes, with a greater percentage seen in Black than in White participants (18% vs. 10%). The age-adjusted RR of diabetes for Black compared with White participants was 1.90 (95% CI 1.59–2.27). The rate of incident diabetes increased across successive IL-6 quartiles overall and in both race groups.

In the metabolic syndrome analytic sample, 23% of participants developed incident metabolic syndrome at follow-up. The age-adjusted RR of metabolic syndrome for Black compared with White participants was 1.06 (95% CI 0.90–1.25). As with diabetes, the rate increased across IL-6 quartiles.

Strong interactions were observed for log IL-6 ∗ race and log IL-6 ∗ central adiposity on incident diabetes (P = 6 × 10⁻⁵ and 0.001, respectively). A three-way interaction for log IL-6 ∗ race ∗ central adiposity was therefore tested (including all two-way interactions), and this was highly statistically significant (P = 8 × 10⁻⁵). There was no sex interaction (P = 0.80). Because of these interactions, stratified analyses were pursued, with model 4 (primary model) results including spline plots shown in Fig. 1, which shows a log-linear relationship for White participants without central adiposity. Similarly, in Supplementary Table 2, both in the quartile-based analysis and per SD increment of log IL-6, the association of IL-6 with incident diabetes was greatest in White participants without central adiposity. Specifically, in model 4, they had an RR of 2.34 (95% CI 1.17–4.66) for the fourth compared with first IL-6 quartile; per SD increment of log IL-6, this RR was 1.44 (1.26–1.64). Black participants with central adiposity had the next strongest association in the quartile-based analysis (RR 2.02 [1.00–4.07]) (Supplementary Table 2); however, this and associations in other stratified groups did not maintain statistical significance. For each SD increment log IL-6, White participants with central adiposity had the second greatest association (RR 1.34 [1.11–1.63]) (Supplementary Table 2).

Supplementary Table 2 summarizes similar patterns across all five models; in model 1, the RR gradually rose across quartiles in all four race/central adiposity groups, with the steepest rise and highest RRs across quartiles in White participants without central adiposity (RR 1.27 [95% CI 0.65–2.48], 2.04 [1.08–3.86], and 3.04 [1.65–5.60] in model 1 quartiles 2, 3, and 4, respectively). Adding covariates in successive models modestly attenuated the RRs across quartiles. The P values for trend across quartiles were statistically significant from models 1 to 4 among White participants, whereas statistical significance was lost for Black participants without central adiposity. Trends were similar per SD of baseline log IL-6.

For model 5 adding HOMA-IR, the RRs among White participants regardless of central adiposity were attenuated; there was a modest impact among Black participants with central adiposity, but among Black participants without central adiposity, the RR was accentuated (Supplementary Table 2).

In a sensitivity analysis to understand confounding by diet, which was missing in 28.1% of participants, we reanalyzed model 4 in each subgroup, including only participants with complete diet data for comparison with the same model with diet added (Supplementary Table 3). Here, there was no confounding observed by diet.

For metabolic syndrome, the interaction P values for log IL-6 ∗ race and log IL-6 ∗ central adiposity were 0.07 and 9 × 10⁻⁴, respectively; there was no three-way interaction (P = 0.73) or sex interaction (P = 0.23). Figure 2 shows the RR of metabolic syndrome stratified by race and central adiposity separately in model 4. For those with central adiposity, a higher IL-6 level was associated with greater metabolic syndrome risk in the spline regressions. Examined by quartile and per SD increment, the relationship was less clear in those with central adiposity, whereas for those without central adiposity, all three analyses were concordant, with the RR for participants in the fourth IL-6 quartile relative to the first being 1.69 (95% CI 1.21–2.37) (Supplementary Table 4). With race stratification, the per SD increment and spline regression demonstrated a greater association in White than in Black participants; this was less apparent considering IL-6 quartile.

Supplementary Table 4 shows similar patterns across all five models; for participants without central adiposity, IL-6 in the third and fourth quartiles compared with quartile 1 showed statistically significant elevated risk in all models, with modest attenuation in models 1–4. The only other subgroup with significant elevated risk was White participants, only in models 1 and 2. The P values for trend across quartiles were statistically significant in all models among participants without central adiposity only. Trends were similar per SD of baseline log IL-6, although associations became statistically significant among participants with central adiposity.

For model 5 adding HOMA-IR, the RRs attenuated among those with central adiposity and remained essentially the same among those without central adiposity. Model 5 RRs were accentuated among Black participants and attenuated among White participants (Supplementary Table 4).

Diet data were missing in 20.8%, so we analyzed those with complete diet data in a sensitivity analysis (Supplementary Table 5); no confounding by diet was observed in this analysis.

In this biracial U.S. cohort, higher IL-6 level was a risk factor for both diabetes and metabolic syndrome; however, associations were generally greater in lower-risk subgroups (i.e., in White compared with Black participants) and in those without central adiposity.

Our findings are consistent with the understanding that IL-6–mediated inflammation plays a causal role in the development of insulin resistance (10). Chronic inflammation is implicated in the insulin resistance seen in diabetes and metabolic syndrome (30). This is likely a result of IL-6 production by activated macrophages in adipose tissue, which accumulates in proportion to adipocyte size (8,11).

Our findings do not support the literature evidence that IL-6 might be a suitable marker for risk assessment for future diabetes in clinical settings. A meta-analysis of 15 prospective studies including 5,421 incident cases and 31,562 noncases showed a 24% greater hazard of developing diabetes per 1-unit increment log-transformed higher IL-6 (hazard ratio 1.24; 95% CI 1.17–1.32) but presented no information on racial differences (10). The current findings address this gap and detected differences in the association by race and central adiposity. Specifically, although our findings among White participants were similar to the literature (10), there was no robust association in Black participants, and there was a race difference that was highly statistically significant, so general clinical applicability of IL-6 measurement is unlikely. This finding contrasts with results of the Multi-Ethnic Study of Atherosclerosis, which reported an association of higher IL-6 level with risk of incident diabetes over 4.7 years but found no significant difference in the association by race (16). This study had shorter follow-up and four ethnic groups which were all free of clinical cardiovascular disease at baseline. These factors and low power may have affected the ability to detect race differences.

In cross-sectional and case–control studies, IL-6 level was higher among patients with versus without metabolic syndrome (14). We are aware of no prospective studies on IL-6 and risk of incident metabolic syndrome and no previous data comparing this association in Black and White participants.

The generally greater association of higher IL-6 level with risk of incident diabetes or metabolic syndrome in participants without versus with central adiposity at baseline may suggest that IL-6 from sources other than central adipose tissue may more effectively induce insulin resistance than that from central adipose tissue. However, in light of existing models describing insulin resistance resulting from inflammation of central adipose tissue, it is possible that those with central adiposity and a relatively high IL-6 level at baseline may have been excluded from our analysis because they developed diabetes at a younger age, leading to the observed findings. Alternatively, the observed difference may be due to other types of bias or confounding not accounted for in the analysis or to chance.

Interventions that might alter IL-6–related diabetes and metabolic syndrome include lifestyle changes (e.g., improved physical activity (31) and healthier diet (32)) or pharmacologic agents (e.g., biguanides (33), sodium–glucose cotransporter 2 inhibitors (34), and monoclonal antibodies (35)). However, findings here suggest that targeting IL-6 in middle and older age may not help to reduce the racial disparity in diabetes affecting Black individuals, because the association of IL-6 with diabetes risk was less apparent in Black compared with White individuals. Future studies should focus on extending our findings to younger individuals, because it is possible inflammation-related diabetes emerges at a younger age in Black individuals than the age group considered here. Existing research shows that the experience of persistent discrimination contributes to elevated inflammation in Black individuals (36,37), and higher IL-6 levels have been observed in racially segregated populations (38), so further study on the relationship between inflammation and glucose disorders in Black individuals is warranted. This should include evaluation of other inflammation-sensitive biomarkers that might reveal possible interventions to narrow the racial disparity.

The present analysis has limitations. As with all studies, the roles of bias, incomplete control for confounding, and chance must be considered in the interpretation of results. Diabetes status was determined on a single fasting or random glucose measurement or by self-reported use of hypoglycemic medication. This may have misclassified some participants, especially because hemoglobin A_1c was not available. Diabetes and metabolic syndrome status were assessed using threshold criteria that incompletely account for the continuous nature of the underlying disease. IL-6 was assessed once at baseline and might have changed over time before the development of disease. It is also possible that acute illness misclassified some participants as having a higher IL-6 level; this was likely to occur with random distribution among participants and thus is unlikely to have affected results. Bias related to participants dying or being lost to follow-up between the two in-home visits could have affected results in unpredictable ways, but previous work has demonstrated minimal evidence of selection bias resulting from attrition of racial disparities in REGARDS (22). There were subtle differences considering results from spline plots and the RR per SD or quartile of IL-6, likely reflecting smaller sample sizes in stratified analyses. Strengths of the present analysis include the prospective study design minimizing reverse causality in associations, the geographic diversity of the cohort supporting generalizability to Black and White U.S. adults, and the oversampling of Black participants. Additionally, this large cohort was observed for >9 years, with satisfactory retention and rigorous data collection.

In conclusion, we confirm that IL-6 is a marker for diabetes risk and describe a similar association between IL-6 and metabolic syndrome. Important race and central adiposity subgroup findings suggest directions for future study. These findings may inform further research to reduce the burden of these conditions and reduce the racial disparity in diabetes.

Acknowledgments. The authors thank the other investigators, staff, and participants of the REGARDS study for their valuable contributions. A full list of participating REGARDS investigators and institutions can be found at https://www.uab.edu/soph/regardsstudy. The authors also appreciate the support and guidance of investigators from the Study Design and Molecular Epidemiology Core of the Vermont Center for Cardiovascular and Brain Health.

Funding. REGARDS is supported by cooperative agreement U01 NS041588 cofunded by the National Institute of Neurological Disorders and Stroke and the National Institute on Aging, National Institutes of Health (NIH), U.S. Department of Health and Human Services. Additional support was provided by the Larner College of Medicine Dean’s Office (B.J.P.). Funding for support from the Study Design and Molecular Epidemiology Core, Vermont Center for Cardiovascular and Brain Health, was provided by grant P20 GM135007 from the National Institute of General Medical Sciences, NIH.

This content is the sole responsibility of the authors and does not necessarily represent the official views of the National Institute of Neurological Disorders and Stroke, the National Institute of Health, the U.S. Department of Veterans Affairs, or the U.S. Government. Federal employee disclaimer (C.N.): The identification of specific products or scientific instrumentation is considered an integral part of the scientific endeavor and does not constitute endorsement or implied endorsement on the part of the author, U.S. Department of Defense (DoD), or any component agency. The views expressed in this manuscript are those of the author and do not reflect the official policy of the U.S. Department of Army/Navy/Air Force, DoD, or U.S. Government.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. B.J.P. and K.S.W. conducted data analyses and generated tables and figures. B.J.P. and M.C. contributed to discussion and wrote the first draft of the manuscript. T.B.P., C.D.N., S.E.J., D.K.M., D.L.L., and N.C.O. contributed to discussion and reviewed and edited the manuscript. All authors approved the final version of the manuscript. B.J.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This study was presented orally at the American Heart Association EPI|Lifestyle Scientific Sessions, Boston, MA, 28 February–3 March 2023.