Pradhan AD, Manson JE, Rifai N, Buring JE, Ridker PM. C-Reactive Protein, Interleukin 6, and Risk of Developing Type 2 Diabetes Mellitus. JAMA. 2001;286(3):327-334. doi:10.1001/jama.286.3.327
Author Affiliations: Center for Cardiovascular Disease Prevention (Drs Pradhan, Manson, Ridker, Rifai, Buring), the Leducq Center for Molecular and Genetic Epidemiology of Cardiovascular Disorders (Dr Ridker), the Divisions of Cardiology (Dr Ridker) and Preventive Medicine (Drs Pradhan, Manson, Ridker, and Buring), Brigham and Women's Hospital and Harvard Medical School; the Department of Pathology (Dr Rifai), Children's Hospital Medical Center and Harvard Medical School; and the Department of Ambulatory Care and Prevention, Harvard Medical School (Dr Buring), Boston, Mass.
Context Inflammation is hypothesized to play a role in development of type 2
diabetes mellitus (DM); however, clinical data addressing this issue are limited.
Objective To determine whether elevated levels of the inflammatory markers interleukin
6 (IL-6) and C-reactive protein (CRP) are associated with development of type
2 DM in healthy middle-aged women.
Design Prospective, nested case-control study.
Setting The Women's Health Study, an ongoing US primary prevention, randomized
clinical trial initiated in 1992.
Participants From a nationwide cohort of 27 628 women free of diagnosed DM,
cardiovascular disease, and cancer at baseline, 188 women who developed diagnosed
DM over a 4-year follow-up period were defined as cases and matched by age
and fasting status with 362 disease-free controls.
Main Outcome Measures Incidence of confirmed clinically diagnosed type 2 DM by baseline levels
of IL-6 and CRP.
Results Baseline levels of IL-6 (P<.001) and CRP
(P<.001) were significantly higher among cases
than among controls. The relative risks of future DM for women in the highest
vs lowest quartile of these inflammatory markers were 7.5 for IL-6 (95% confidence
interval [CI], 3.7-15.4) and 15.7 for CRP (95% CI, 6.5-37.9). Positive associations
persisted after adjustment for body mass index, family history of diabetes,
smoking, exercise, use of alcohol, and hormone replacement therapy; multivariate
relative risks for the highest vs lowest quartiles were 2.3 for IL-6 (95%
CI, 0.9-5.6; P for trend = .07) and 4.2 for CRP (95%
CI, 1.5-12.0; P for trend = .001). Similar results
were observed in analyses limited to women with a baseline hemoglobin A1c of 6.0% or less and after adjustment for fasting insulin level.
Conclusions Elevated levels of CRP and IL-6 predict the development of type 2 DM.
These data support a possible role for inflammation in diabetogenesis.
Type 2 diabetes mellitus (DM) is estimated to affect 15 million Americans,
is dramatically increasing in incidence, and is associated with an augmented
risk for cardiovascular disease, especially among women.1- 3
Because of the resultant macrovascular and microvascular injury typical of
this disease, the economic and functional burdens are greatest during mid-to-late
adulthood. Compounding these issues, as many as one third of individuals with
type 2 DM are undiagnosed, and approximately 20% have diabetic retinopathy
or evidence of systemic vasculopathy at clinical presentation.4
Although the main physiological abnormalities are insulin resistance
and impaired insulin secretion,5- 7
the specific underlying determinants of these metabolic defects remain uncertain.
An accumulating body of evidence suggests that inflammation may play a crucial
intermediary role in pathogenesis, thereby linking diabetes with a number
of commonly coexisting conditions thought to originate through inflammatory
mechanisms. In this regard, substantial experimental evidence and more recent
cross-sectional data suggest that interleukin 6 (IL-6) and C-reactive protein
(CRP), 2 sensitive physiological markers of subclinical systemic inflammation,
are associated with hyperglycemia, insulin resistance, and overt type 2 DM.8- 15
Indeed, it recently has been postulated that type 2 DM may represent a disease
of the innate immune system,16 a hypothesis
of particular interest because both of these inflammatory biomarkers also
are known to predict the development of cardiovascular disease in otherwise
healthy populations.17- 20
Interleukin 6, a major proinflammatory cytokine, is produced in a variety
of tissues, including activated leukocytes, adipocytes, and endothelial cells.
C-Reactive protein is the principal downstream mediator of the acute phase
response and is primarily derived via IL-6–dependent hepatic biosynthesis.
In rodent models of glucose metabolism, the in vivo infusion of human recombinant
IL-6 has been shown to induce gluconeogenesis, subsequent hyperglycemia, and
compensatory hyperinsulinemia.21 Similar metabolic
responses have been observed in humans after administration of subcutaneous
recombinant IL-6.10 Cross-sectional investigations
further support a role for inflammation in the etiology of diabetes; several
studies have demonstrated elevated levels of IL-6 and CRP among individuals
both with features of the insulin resistance syndrome and clinically overt
type 2 DM.11- 15
Despite these observations, to our knowledge, there are no published
prospective data evaluating the relationship between IL-6, CRP, and the development
of type 2 DM. Therefore, we evaluated whether baseline plasma levels of these
inflammatory markers might independently predict future risk for this disease
among apparently healthy individuals.
We designed a prospective, nested case-control study involving participants
in the Women's Health Study (WHS), an ongoing randomized clinical trial initiated
in 1992 evaluating the use of low-dose aspirin and vitamin E in the primary
prevention of cardiovascular disease and cancer among female health professionals
aged 45 years and older.22 Among participants
in this primary prevention trial, 27 628 individuals (69% of the WHS
cohort) also were free of baseline type 2 DM and provided whole blood samples
at enrollment. These samples were centrifuged and then stored in liquid nitrogen
until laboratory analysis. Plasma samples collected in EDTA were used for
IL-6, CRP, and insulin level determination. Packed red blood cell samples
were used for measurement of hemoglobin (Hb) A1c.
Case subjects were WHS participants who provided blood specimens, were
free of reported diabetes at enrollment, and subsequently developed newly
diagnosed diabetes during a 4-year observation period. Candidate cases were
initially identified by self-report on yearly follow-up questionnaires and
were subsequently verified through telephone interview conducted by a physician
(A.D.P.). Based on revised American Diabetes Association diagnostic criteria,23 cases were confirmed if 1 or more of the following
conditions were met: (1) presence of more than 1 classic symptom of hyperglycemia
(ie, polyuria, polydipsia, weight loss with or without polyphagia, and blurred
vision) plus either a fasting plasma glucose of 126 mg/dL (7.0 mmol/L) or
higher or random plasma glucose 200 mg/dL (11.1 mmol/L) or higher; (2) in
the absence of symptoms, 2 or more elevated plasma glucose concentrations
(fasting plasma glucose of ≥126 mg/dL [7.0 mmol/L], random plasma glucose ≥
200 mg/dL [11.1 mmol/L], or 2-hour plasma glucose ≥200 mg/dL [11.1 mmol/L]
during oral glucose tolerance testing); or (3) use of insulin or an oral hypoglycemic
agent. The primary care physician's office was contacted for supporting documentation
as necessary. Candidate cases who did not meet the diagnostic criteria, who
were found to have prevalent diabetes at enrollment, who died, or who were
otherwise lost to follow-up were excluded. In addition, to reduce misclassification
bias that might be due to undiagnosed diabetes at study entry, individuals
diagnosed within the first year of follow-up (n = 69) also were excluded.
For each woman who developed confirmed type 2 DM, 2 control subjects
were chosen at random among individuals free of self-reported DM at the time
the case subject reported her event. Control subjects were matched by age
(within 1 year) and by fasting status of submitted blood specimens. Fasting was defined as 10 hours or longer since last meal
prior to sample collection. The final study group undergoing laboratory investigation
included 288 confirmed cases and 576 matched controls.
Because of the high prevalence of undiagnosed diabetes among middle-aged
US population and because this study was designed to evaluate the role of
inflammation as a determinant of future disease, we limited our sample to
individuals with a baseline HbA1c of 6.5% or less. Participants
with missing values for baseline clinical covariates of interest also were
eliminated from the analysis (body mass index [BMI; calculated as weight in
kilograms divided by the square of height in meters], 3% of cases and 1.5%
of controls; history of hypertension, 0.5% of cases and 0.7% of controls;
history of hyperlipidemia, 0.5% of controls; and use of hormone replacement
therapy [HRT], 0.3% of controls). The primary sample thus comprised 188 cases
and 362 age-matched controls with HbA1c of 6.5% or less on entry
into the cohort. Based on available published data regarding the exposure
rate for elevated CRP levels (≥0.22 mg/dL) among US women,24
we estimated our power to be approximately 87% for detecting a relative risk
(RR) of 1.8. Exposure rates for elevated IL-6 levels were not available at
the time of study design. The distribution of cases according to year of follow-up
at time of diagnosis is as follows: 39 cases in year 2; 76 cases in year 3;
and 73 cases in year 4. Among approximately two thirds of women who were fasting
at the time of sample collection, we also measured specific insulin levels
as an indicator of underlying insulin resistance.
Baseline plasma samples were thawed and assayed for IL-6, CRP, and specific
insulin levels. Hemoglobin A1c was measured by immunoassay (Hitachi
911 Analyzer; Roche Diagnostics, Indianapolis, Ind). The plasma concentration
of IL-6 was measured by a commercially available enzyme-linked immunosorbent
assay (R&D Systems, Minneapolis, Minn). C-Reactive was measured via a
high-sensitivity latex-enhanced immunonephelometric assay on a BN II analyzer
(Dade Behring, Newark, Del).25 Double antibody
systems (Linco Research, St Louis, Mo), with less than 0.2% cross-reactivity
between insulin and its precursors, were used to measure specific concentrations
of plasma insulin. In addition, as insulin levels may be falsely lowered in
the presence of hemolysis,26 samples with free
hemoglobin values higher than 50 mg/dL (spectrophotometric method, Hitachi
911 Analyzer) were excluded from fasting subgroup investigations. Blood samples
were analyzed in randomly ordered case-control triplets so as to reduce systematic
bias and interassay variation. Blinded quality control specimens were analyzed
simultaneously with the study sample. The coefficients of variation for IL-6,
CRP, and insulin were 12.7%, 12.0%, and 14.7%, respectively.
We used repeated measures analysis (SAS PROC mixed; SAS Institute, Cary,
NC) to evaluate differences in means and a matched χ2 statistic
to assess for differences in proportions, both approaches allowing for correlation
within matched case-control sets. Because the distributions of IL-6, CRP,
and insulin are skewed, differences in medians were tested with the Wilcoxon
rank-sum test. Analysis of linear trends was used to assess associations between
increasing level of each biomarker and risk of future diabetes after the sample
was divided into quartiles based on the distribution of controls. Quartile-specific
risk estimates were obtained through conditional logistic regression adjusting
for BMI, family history of diabetes in a first-degree relative, smoking, physical
activity, alcohol consumption, and use of HRT. Continuous and categorical
variables were specified according to best fit through comparison of competing
regression models. In particular, BMI was controlled for on a continuous linear
scale and insulin was expressed using both insulin and insulin-squared terms.
Sensitivity analyses, using a HbA1c cutpoint of 6.0% for
exclusion of prevalent diabetes at baseline, were performed to check the robustness
of our findings. In addition, although baseline abnormalities in fasting insulin
levels may be considered an intermediary factor in causal pathways, we adjusted
for this metabolic parameter in fasting subgroup analyses to assess the residual
predictive role of each inflammatory marker under study. Spearman partial
correlation coefficients were calculated for each inflammatory marker against
fasting insulin level and against other continuous metabolic variables while
controlling for age and BMI.
Conditional logistic regression was used to examine the joint role of
IL-6 and CRP in predicting diabetes after dividing the primary sample into
4 groups based on the 75th percentile cutpoints for each biomarker. Finally,
to assess consistency of risk relationships among obese and nonobese individuals,
the study sample was divided into 6 groups based on a BMI cutpoint value of
29 kg/m2 (the upper tertile of BMI for our study population) and
low, medium, and high tertiles of the inflammatory markers.
Baseline characteristics of women who were subsequently diagnosed with
diabetes (case subjects) and those remaining free of diabetes (control subjects)
are shown in Table 1. As would
be predicted, women who subsequently developed diabetes had a higher mean
BMI, were more likely to have a family history of diabetes in a first-degree
relative, and were more likely to have a history of hypertension or hyperlipidemia.
In addition, case subjects exercised less frequently and consumed less alcohol.
There were no statistically significant differences in ethnicity, smoking,
or HRT use.
Median baseline levels of IL-6 and CRP were significantly higher among
cases than among controls (P<.001) (Table 1). Moreover, increasing levels of both inflammatory markers
were associated with a higher risk of developing future diabetes; in age-matched
analyses, the RRs of incident type 2 DM for increasing quartiles of IL-6 were
1.0, 2.5, 4.1, and 7.5, respectively (P<.001 for
trend) and the RRs for increasing quartiles of CRP were 1.0, 2.2, 8.7, and
15.7, respectively (P<.001 for trend) (Table 2). Adjustment for BMI markedly attenuated
these relationships, although persistent positive effects of IL-6 (P = .008 for trend) and CRP (P<.001 for
trend) were observed. Indeed, CRP remained a significant predictor in fully-adjusted
models that included BMI, family history of diabetes, smoking, physical activity,
alcohol consumption, and HRT use. Overall, the RR for future diabetes increased
28% (95% confidence interval [CI], −1% to 65%; P
= .07) per quartile increase in baseline IL-6 levels and 64% (95% CI, 22%-218%; P = .001) per quartile increase in baseline CRP levels.
Similar findings were observed in analyses limited to those with an HbA1c of 6.0% or less at baseline. For example, in this subset, fully adjusted
RRs of incident diabetes across quartiles of CRP were 1.0, 1.8, 3.8, and 4.9,
respectively (P = .02 for trend).
In the subgroup of participants who provided fasting blood specimens,
the median insulin level also was significantly higher in case subjects than
control subjects (77.5 vs 39.3 pmol/L; P<.001).
Therefore, we evaluated whether relationships between IL-6, CRP, and the future
risk of diabetes were independent of fasting insulin level. As shown in Table 3, adjustment for baseline fasting
insulin level further attenuated the effects of IL-6 levels. However, the
risk relationship for CRP was not materially altered after adjustment for
this factor. In addition, in this subgroup, Spearman partial correlation coefficients
between inflammatory markers and both fasting insulin and BMI were statistically
significant (P<.001) (Table 4). C-Reactive protein was more strongly correlated than IL-6
with each parameter tested. Hemoglobin A1c of 6.5% or less was
not strongly associated with either biomarker.
To assess potential joint effects, we computed the RR of DM after dividing
the original sample into 4 groups based on the 75th percentile of control
distributions for IL-6 and CRP (Figure 1).
As shown, the RR of type 2 DM was highest among women with both high IL-6
and CRP levels, suggesting a multiplicative effect above that seen for either
high IL-6 or high CRP levels alone.
To investigate effect modification by BMI, we determined the RR of incident
diabetes among women with a BMI less than 29 kg/m2or 29 kg/m2 and greater (Figure 2).
After controlling for BMI within strata, higher baseline plasma levels of
IL-6 and CRP were associated with increased risk of incident disease. Notably,
even among nonobese women, increasing IL-6 and CRP levels conferred an augmented
stepwise elevation in risk.
Among individuals with a baseline HbA1c of 6.5% or less,
we performed secondary analyses to address the additional issues of potential
surveillance bias, possible confounding by degree of baseline chronic hyperglycemia,
and residual confounding by obesity. Case subjects were noted to have a higher
baseline prevalence of diagnosed hypertension and hyperlipidemia and on this
basis might have been screened more frequently for other major cardiovascular
risk factors, including diabetes. We attempted to control for this potential
surveillance bias by adjusting for history of hypertension and hyperlipidemia
and found equivalent results; the BMI-adjusted RR in the highest vs the lowest
quartile for IL-6 was 3.0 (95% CI, 1.3-7.0; P = .007
for trend) and for CRP was 4.0 (95% CI, 1.5-10.6; P<.001
for trend). To evaluate for confounding by chronic hyperglycemia, we controlled
for baseline HbA1c; in analyses simultaneously controlling for
BMI, the adjusted RR for the highest compared with the lowest quartile of
IL-6 was 2.0 (95% CI, 0.6-6.2; P = .77 for trend)
and for CRP was 3.4 (95% CI, 0.9-12.1; P = .03 for
trend). To assess residual confounding by obesity, we used waist-hip ratio
(WHR) and waist circumference as alternate indices of adiposity among approximately
two thirds of the study cohort for whom these parameters were available. The
distribution of baseline characteristics in this subgroup was similar to those
observed for the primary study sample. In statistical models controlling for
WHR in addition to BMI, the RR in these analyses comparing the highest with
the lowest quartile for IL-6 was 4.1 (95% CI, 1.4-11.8; P = .004 for trend) and for CRP was 3.1 (95% CI, 1.1-9.4; P = .001 for trend). Similar results were obtained after adjustment
for waist circumference. In addition, after confining the study cohort to
normal weight subjects (BMI ≤ 25 kg/m2), the median CRP level
remained significantly elevated among case subjects vs control subjects (0.46
vs 0.15 mg/dL; P = .01). Finally, we found that randomized
treatment assignment to vitamin E or low-dose aspirin did not affect the observed
risk relationships during the 4-year observation period.
In this prospective study of apparently healthy middle-aged women, 2
markers of systemic inflammation, CRP and IL-6, were found to be determinants
of risk for type 2 DM. In particular, CRP was a powerful independent predictor
after adjustment for obesity, clinical risk factors, and fasting insulin levels.
Parallel associations were found for IL-6, although lower in magnitude and
of borderline statistical significance after multivariate adjustments. These
findings were robust in sensitivity analyses limited to subjects with a baseline
HbA1c of 6.0% or less and were consistently noted in both obese
and nonobese individuals.
To our knowledge, no prior epidemiological evidence has been available
linking baseline CRP and IL-6 levels to incident DM. However, cross-sectional
studies have reported increased concentrations of these inflammatory markers
in both the insulin resistance syndrome and overt type 2 DM.11- 15
Our data also extend prior work in which other inflammatory markers, such
as white blood cell count, fibrinogen, and low serum albumin,27
and inflammation-associated hemostasis variables, such as factor VIII and
von Willebrand factor,28 were associated with
risk of diabetes; although in these latter investigations, risk relationships
largely disappeared after adjustment for obesity.
The current prospective data support a possible role for inflammation
in diabetogenesis and are in accord with previous hypotheses originated by
Pickup and Crook16 that type 2 DM may be a
manifestation of an ongoing cytokine-mediated acute phase response initiated
by the body's innate immune system. Of particular relevance to the current
findings, CRP is thought to exhibit several characteristics that imply a fundamental
role in natural host defense. Specifically, CRP is a member of the pentraxin
family of oligomeric proteins involved with pattern recognition in innate
Reported immunoregulatory functions of CRP include enhancement of leukocyte
reactivity, complement fixation, modulation of platelet activation, and clearance
of cellular debris from sites of active inflammation.29,32,33
Although our data support etiological associations, at this time explicit
mechanisms remain speculative and require further study.
Several alternative, perhaps coordinate, explanations for our results
warrant further discussion. First, it is possible that the associations observed
in this study of diabetes reflect underlying atherosclerosis among case subjects.17- 20 However,
it is worth noting that the 4-year cardiovascular event rate among our study
cohort was low (1 case subject and 1 control subject with incident stroke,
myocardial infarction, coronary angioplasty, or coronary artery bypass graft
surgery), even among those individuals with greatest baseline elevations of
either IL-6 or CRP levels.
Obesity-mediated cytokine production is another important and perhaps
central mechanism for systemic elevations of both of these biomarkers. The
primary cytokine involved in hepatic CRP synthesis is IL-6, also an important
adipocyte signaling molecule released from visceral and subcutaneous fat stores.
Indeed, approximately 25% of in vivo systemic IL-6 originates from subcutaneous
adipose tissue34 and is thought to modify adipocyte
glucose and lipid metabolism and body weight.35- 37
Furthermore, omental fat cells have been shown to secrete as much as 2 to
3 times more IL-6 in vitro than cells derived from subcutaneous stores,38 an intriguing finding as venous drainage from omental
fat provides direct access to the portal system, and abdominal adiposity is
strongly linked to insulin resistance.39- 42
Although we found a stronger relationship between baseline elevations of CRP
and incident diabetes than that seen for IL-6, this may partially reflect
the considerably longer plasma half-life of CRP that thereby may provide a
more stable indication of subclinical inflammation. Furthermore, because of
diurnal variation in IL-6 release and because study specimens were obtained
at different times of the day, random misclassification of both cases and
controls may have biased our results for IL-6 toward the null. In this regard,
plasma CRP levels may represent a more appropriate integrated measure of basal
In the present analysis, BMI was used as the principal measure of obesity,
and as expected, significantly attenuated RR estimates for both IL-6 and CRP.
Adjustment for WHR or waist circumference, indices of abdominal adiposity,
similarly weakened the observed risk relationships. However, significant associations
for both IL-6 and CRP were nonetheless observed in multivariate models adjusting
for these factors. In addition, a stepwise RR gradient was evident even among
nonobese individuals (Figure 2).
Regardless of the apparent diminution of effects in statistical models controlling
for obesity, the graded crude association between increasing concentrations
of these inflammatory markers and risk for diabetes suggests a possible role
in disease expression whether this process is obesity-driven or autonomously
Another potential mechanism that may explain our results is the relationship
between inflammation and endothelial dysfunction. Altered endothelial permeability
and diminished peripheral blood flow may limit insulin delivery and promote
insulin resistance in metabolically active tissues.43
Indeed, in the dynamic state, interstitial insulin concentration appears to
be the rate-limiting step for determining insulin effectiveness.44
In this regard, the local administration of proinflammatory cytokines has
been shown to impair endothelium-dependent vascular relaxation in human veins
in vivo,45 and plasma IL-6 response to mild
inflammatory stimuli is correlated with temporary but profoundly diminished
response to both physical and pharmacological vasodilatory agents.46 Elevated CRP levels have been similarly associated
with blunted endothelial vasoreactivity and, perhaps more importantly, CRP
normalization is associated with subsequent improvement of regional blood
flow.47 In addition, cross-sectional analyses
show that CRP is strongly correlated with markers of endothelial activation
and dysfunction.48 The increasingly recognized
associations between endothelial dysfunction, inflammation, and insulin resistance
provide an additional plausible physiologic basis for our findings.
Finally, it is possible that elevated IL-6 and CRP levels may largely
reflect adipocyte activation. For instance, IL-6 and downstream CRP production
may be associated with the corelease of other pathogenic substances arising
from otherwise stimulated adipocytes. Other potential mediators of insulin
resistance deriving from adipose stores include tumor necrosis factor α,49,50 leptin,51,52
free fatty acids,53 and resistin.54
Nonetheless, under the assumption that elevated levels of IL-6 and CRP purely
reflect altered adipocyte function, the ready availability of reliable and
sensitive markers of this process may represent a novel approach for early
identification of both obese and nonobese individuals at increased risk for
the clinical development of this disease.
Several limitations of our study warrant further discussion. First,
because our cohort was composed of primarily healthy middle-aged women, our
results may not be generalizable to other age groups or to men who may be
at risk for type 2 DM. Second, it is possible that undetected diabetes at
study entry might have biased our results. However, to minimize the impact
of this factor we excluded individuals with a baseline HbA1c of
greater than 6.5% from our primary sample and also conducted sensitivity analyses
using a lower threshold for case and control inclusion. Furthermore, all women
who were diagnosed with diabetes within the first year of follow-up were eliminated.
A third limitation is potential residual confounding by obesity in our multivariate
analyses. Although BMI, WHR, and waist circumference are the more common clinical
measures of adiposity, multivariate adjustment on this basis may not fully
account for the metabolic consequences of obesity. In this regard, it is important
in analyses to note adjusting for WHR and waist circumference in addition
to BMI, we found consistent results statistically significant and undiminished
in magnitude. Finally, we used a single baseline plasma measurement of each
biomarker. We therefore could not evaluate the effects of changes in plasma
levels of inflammatory markers over time. However, several longitudinal analyses
have found that levels of CRP are stable during long-term follow-up, as long
as measurements are not made within 2 weeks of an acute infection.55,56
In conclusion, in this prospective evaluation of 2 markers of systemic
inflammation in the prediction of incident diabetes, elevated CRP was found
to be a powerful independent risk determinant. Interleukin 6 levels also were
elevated among individuals at risk, although these associations were markedly
attenuated after multivariate adjustment. Our epidemiological observations,
coupled with emerging experimental evidence, support a possible role for inflammation
in the pathogenesis of type 2 DM. Our data also raise the possibility that
inflammatory markers, like CRP, might provide an adjunctive method for early
detection of risk for this disease.