Additional Contribution of Emerging Risk Factors to the Prediction of the Risk of Type 2 Diabetes: Evidence From the Western New York Study

Authors

  • Saverio Stranges,

    1. Department of Social and Preventive Medicine, The State University of New York at Buffalo, Buffalo, New York, USA
    2. Clinical Sciences Research Institute, University of Warwick, Coventry, United Kingdom
    Search for more papers by this author
  • Lisa B. Rafalson,

    1. Department of Social and Preventive Medicine, The State University of New York at Buffalo, Buffalo, New York, USA
    Search for more papers by this author
  • Jacek Dmochowski,

    1. Department of Social and Preventive Medicine, The State University of New York at Buffalo, Buffalo, New York, USA
    2. Department of Mathematics and Statistics, UNC Charlotte, Charlotte, North Carolina, USA
    Search for more papers by this author
  • Karol Rejman,

    1. School of Nursing, The State University of New York at Buffalo, Buffalo, New York, USA
    Search for more papers by this author
  • Russell P. Tracy,

    1. Department of Pathology, University of Vermont, Burlington, Vermont, USA
    2. Department of Biochemistry, University of Vermont, Burlington, Vermont, USA
    Search for more papers by this author
  • Maurizio Trevisan,

    1. Department of Social and Preventive Medicine, The State University of New York at Buffalo, Buffalo, New York, USA
    Search for more papers by this author
  • Richard P. Donahue

    Corresponding author
    1. Department of Social and Preventive Medicine, The State University of New York at Buffalo, Buffalo, New York, USA
      (rpd1@buffalo.edu)
    Search for more papers by this author

(rpd1@buffalo.edu)

Abstract

Objective: To examine whether several biomarkers of endothelial function and inflammation improve prediction of type 2 diabetes over 5.9 years of follow-up, independent of traditional risk factors.

Methods and Procedures: A total of 1,455 participants from the Western New York Study, free of type 2 diabetes at baseline, were selected. Incident type 2 diabetes was defined as fasting glucose exceeding 125 mg/dl or on antidiabetic medication at the follow-up visit. Sixty-one people who met the case definition (8/1,000 person years) were identified and individually matched with up to three controls on gender, race, year of study enrollment, and baseline fasting glucose (<110 or 110–125 mg/dl). Biomarkers were measured from frozen baseline samples.

Results: In conditional logistic regression analyses accounting for traditional risk factors (age, family history of diabetes, smoking, drinking status, and BMI), E-selectin was positively related (3rd vs. 1st tertile: odds ratio 2.77, 95% confidence interval (CI) 1.13–6.79, P for linear trend = 0.023) and serum albumin was inversely related (3rd vs. 1st tertile: odds ratio 0.36, 95% CI 0.14–0.93, P for linear trend = 0.032) to type 2 diabetes incidence. The addition of E-selectin, serum albumin, and leukocyte count to a basic risk factor model including only traditional risk factors significantly increased the area under the receiver operating characteristic curve (AUC) (from 0.646 to 0.726, P value = 0.04).

Discussion: These results support the role of endothelial dysfunction and subclinical inflammation as important mechanisms in the etiopathogenesis of type 2 diabetes; moreover, they indicate that novel biomarkers may improve the prediction of type 2 diabetes beyond the use of traditional risk factors alone.

Introduction

While aging, family history, nonwhite race/ethnicity, obesity, and physical inactivity are well-established risk factors for the development of type 2 diabetes (1,2,3,4); growing evidence suggests that other emerging risk factors may play an independent role in the etiology and pathogenesis of this disease. In particular, recent epidemiologic studies indicate that subclinical inflammation and endothelial dysfunction may represent potentially important mechanisms leading to type 2 diabetes (5,6,7,8). Little attention, however, has been paid to whether these novel markers add to the identification of future diabetic patients beyond that of easily determined risk factors (9). Therefore, we sought to examine whether multiple biomarkers representing emerging risk factors for type 2 diabetes (i.e., markers of inflammation/fibrinolysis, markers of endothelial dysfunction, and indices of insulin sensitivity/resistance) would contribute to the prediction of type 2 diabetes beyond the use of traditional risk factors alone (i.e., obesity, family history, smoking, alcohol consumption). The data were derived from a 6-year longitudinal investigation of the Western New York Study, a population-based study of diabetes and cardiovascular risk factors among residents of Erie and Niagara Counties, New York.

Methods and Procedures

Study population

Participants in this report were originally enrolled as healthy control participants in the Western New York Health Study, an epidemiologic case-control investigation of patterns of alcohol intake and coronary heart disease in Erie and Niagara Counties, New York, conducted from 1996 to 2001 (59.5% initial response rate). The details of the methodology have been described previously (10,11,12). In brief, the initial cohort was selected from drivers' license lists and Health Care Finance Administration lists. Eligible participants for this study were men and women aged 35–79 years selected from the baseline examination without known clinical cardiovascular disease (self-report) or type 2 diabetes (fasting plasma glucose >125 mg/dl or self-report) and who were capable of completing this study protocol (n = 2,652). Exclusion criteria included self-report of any medical condition that would prohibit participation (e.g., all cancers except skin cancer, type 1 diabetes, physical or mental impairment, permanent change in residence to out of state, deceased, or inability to contact and determine eligibility). This left 2,139 persons eligible, of whom 1,455 completed the full clinical examination (68.0%) at the follow-up visit in 2003–2004 (5.9 ± 0.8 years). This protocol was approved by the University at Buffalo Health Science Institutional Review Board and all participants provided written informed consent prior to participation.

During the follow-up, 72 incident cases of type 2 diabetes were identified, for an incidence rate of 8.02 per 1,000 person years (95% confidence interval (CI) 6.2–10.3). This rate is comparable to other studies of largely non-Hispanic white populations that used a similar definition of diabetes classification (2,3). For these analyses, an incident case of type 2 diabetes was further defined as a person who had been diagnosed by their physician since their baseline examination and was taking antidiabetic medications, or a person whose fasting glucose at the follow-up visit exceeded 125 mg/dl (n = 61) (13). Self-reported cases of type 2 diabetes but without pharmacologic treatment were not included in this analysis (n = 11). Approximately 47.5% had been diagnosed and treated since the baseline examination and the remainder was discovered by a fasting glucose level exceeding 125 mg/dl at the follow-up visit. Incident cases of type 2 diabetes were individually matched with up to three controls on gender, race/ethnicity, year of study enrollment, and according to whether their baseline fasting glucose was <110 mg/dl or 110–125 mg/dl in order to control to the extent possible for differences associated with glycemia. Not all cases could be matched to three controls, thus this analysis focuses on 61 cases and 158 controls.

Study protocol

Both at the baseline and at the follow-up visit, participants underwent a clinical examination that included resting blood pressure and anthropometric measurements. Several questionnaires that were first administered at the baseline examination were re-administered at the follow-up visit. These assessed lifestyle and health habit information including: cigarette use, physical activity, alcohol use, general health and well-being, personal and family health history, medication use, and socioeconomic status. A positive family history of diabetes was defined as a positive report in a first-degree relative. Hypertension was defined as blood pressure ≥140/90 mm Hg or use of antihypertensive medications. Anthropometric measurements were performed by trained and certified interviewers on participants who wore light clothing and no shoes. Body mass index was calculated as weight (kg) divided by height in meters2.

Laboratory measurements

Fasting glucose concentrations were determined using the glucose oxidase method (Beckman instruments Fullerton, CA) and the inter-assay coefficient of variation (CV) was <5%. After identification of those who progressed (or not) to type 2 diabetes, the baseline aliquots of serum or plasma were retrieved and sent through overnight courier to the Laboratory for Clinical Biochemistry Research, University of Vermont or the Department of Endocrinology at the University of Pittsburgh for assay (14). High-sensitivity C-reactive protein (hsCRP) was measured using the Behring Nephelometer II from Dade Behring using a particle-enhanced immunonephelometric assay. Intra-assay CVs ranged 2.3–4.4% and inter-assay CVs ranged 2.1–5.7%. Interleukin-6 (IL-6) was measured by ultra-sensitive enzyme-linked immunosorbent assay (R&D Systems, Minneapolis, MN). Using this method, we determined a routine CV of 6.3%. Plasminogen activator inhibitor-1 assay was originally developed by Collen and colleagues (15). It is done as a two-site enzyme-linked immunosorbent assay. The analytical CV for this assay is 3.5%. Soluble E-selectin was measured using a high-sensitivity quantitative sandwich enzyme (Parameter Human sE-selectin Immunoassay; R&D Systems, Minneapolis, MN). Intra-assay and inter-assay CVs ranged 4.7–5.0% and 5.7–8.8%, respectively. Human Soluble Intercellular Adhesion Molecule-1 (sICAM-1) was measured by an enzyme-linked immunosorbent assay (Parameter Human sICAM-1 Immunoassay; R&D Systems, Minneapolis, MN). The laboratory CV was 5.0%. Adiponectin was assayed from a kit provided by Linco Research (St Charles, MO). This kit is an radioimmunoassay using a double antibody-PEG separation. The assay has a detection limit of 1 ng/ml. The inter-assay CV ranged 12.2–14.5% and the intra-assay CV ranged 3.7–6.1%. Fasting insulin was assayed from a kit provided by Linco Research that has minimal cross-reactivity with human proinsulin. The assay has a lower detection limit of 2 μU/ml with inter-assay CV of 3.6–8.4% and intra-assay CV of 2.2–4.4%. Serum albumin was assayed by a nephelometric immunoassay using a monospecific antiserum to human albumin. White blood cell (WBC) count was assessed by an automated cell counter. Insulin resistance was estimated using the homeostatic model assessment (16).

Statistical analysis

All analyses were conducted using the Statistical Package for Social Sciences (SPSS-13.0; SPSS, Chicago, IL) and SAS 9.1 (Cary, NC). Differences in baseline characteristics and baseline biomarkers between incident cases of type 2 diabetes and their matched controls were evaluated using independent sample t-tests for continuous variables and χ2-test for categorical variables. Two principal sets of analyses were conducted. In the first set, conditional logistic regression analyses were performed, conditioned on the matching factors (i.e., gender, race/ethnicity, year of baseline visit, and baseline fasting glucose) to evaluate the association of traditional and emerging risk factors with the risk of type 2 diabetes. All biomarkers were a priori divided into tertiles based on the distribution among controls. Traditional risk factors considered in this report included: the baseline values of age, reported family history of diabetes (yes vs. no), smoking status (current smokers vs. current nonsmokers), drinking status (current drinkers vs. current nondrinkers), and in a separate model body mass index. Analyses that substituted waist circumference or abdominal height for BMI yielded similar results and thus are not presented.

In the second set of analyses, we wished to evaluate whether the emerging risk factors would improve the diagnostic accuracy of incident type 2 diabetes. We compared the area under the curve using receiver operating curves (ROCs) with the nonparametric approach of DeLong and co-workers (17). All calculations were based on predictive probabilities obtained from three different models. Variables considered for inclusion were derived from the results of the analyses shown in Table 3. Because cases and controls were matched on several variables, these were not considered in the development of the ROCs. Model 1 used stepwise unconditional logistic regression among the entire sample of 1,455 participants. The dependent variable was type 2 diabetes case/control status while the independent variables considered for inclusion were: gender, race/ethnicity, year of study enrollment, age, family history of diabetes, smoking, drinking status, and BMI. Only gender, family history of diabetes, and BMI were significant. Unconditional logistic regression was employed wherein the significant variables from Model 1 (gender, BMI, and family history of type 2 diabetes mellitus) were then forced into Model 2 using the 61 cases and 158 controls to compare how well the case-control study results resembled the results from the entire cohort. Thus, these models contain identical predictor variables. Model 3 (the extended model) was developed by adding the emerging risk factors (i.e., WBC, serum albumin, and E-selectin), also selected by stepwise regression, to Model 2.

Table 3.  Multivariate odds ratios of type 2 diabetes for traditional and emerging risk factors. The Western New York Study, 1996–2004
inline image

Results

The results for the first aim of the study are presented in Tables 13. Table 1 shows the baseline characteristics of the study participants according to case/control status. Mean values of anthropometric measures (e.g., BMI, waist circumference, and abdominal height) were significantly higher among participants who became diabetic than among matched controls. Family history of diabetes and hypertension at baseline were also significantly more prevalent among diabetic subjects than among controls. There was a statistically significant difference in mean baseline concentrations of hsCRP, WBC, serum albumin (lower among the cases), and adiponectin (lower among the cases).

Table 1.  Baseline characteristics of participants by diabetic case-control status: The Western New York Study, 1996–2004
inline image

Table 2 displays the results of age-adjusted analyses using conditional logistic regression models to evaluate the association between emerging risk factors and incident type 2 diabetes. Higher levels of hsCRP, IL-6, WBC, and E-selectin were associated with an increased risk of type 2 diabetes, although for IL-6 the association was of borderline significance (P for trend = 0.077). In contrast, higher levels of serum albumin and adiponectin were associated with a significantly reduced risk of type 2 diabetes. No significant associations were found for fibrinogen, plasminogen activator inhibitor-1, and sICAM-1.

Table 2.  Odds ratios of type 2 diabetes for tertiles of emerging risk factors: The Western New York Study, 1996–2004
inline image

Table 3 shows the results of conditional logistic regression models to examine the predictive ability of emerging risk factors for type 2 diabetes after taking the effects of traditional risk factors into account without (reduced model) and with BMI (fully adjusted model). Overall, findings suggest an attenuation of the observed associations after taking the effects of BMI into account. Specifically, for markers of inflammation, there was a tendency for higher levels of hsCRP and WBC to be associated with an increased risk of type 2 diabetes; however, the association of hsCRP with the risk of type 2 diabetes was greatly attenuated after adjustment for BMI. In contrast, higher levels of serum albumin were associated with a significantly reduced risk of type 2 diabetes; after adjustment for BMI, this relative protection was still statistically significant (3rd vs. 1st tertile: OR 0.36; 95% CI 0.14–0.93, P for trend = 0.032). For markers of endothelial function, E-selectin was a strong, significant predictor of type 2 diabetes in both models. In the fully adjusted model, the odds ratio for diabetes comparing the highest to the lowest tertile of E-selectin was 2.77 (95% CI 1.13–6.79), with a significant linear trend (P = 0.023). For adiponectin, higher levels were associated with a significantly reduced risk of type 2 diabetes (3rd vs. 1st tertile: OR 0.34; 95% CI 0.12–1.00); however, after adjustment for BMI, this decreased risk, while still sizable, became not statistically significant.

Figure 1 addresses the second major aim of the study and displays the results of ROC analyses derived from unconditional logistic regression for three different models described above. The results, expressed as area under ROC (AUC) and 95% CI were as follows: Model 1—Basic Model (gender, BMI, and family history of type 2 diabetes) in entire population: 0.647 (0.562, 0.731); Model 2—Basic Model in Case/Control Sample: 0.646 (0.562, 0.730); Model 3—Extended Model in Case/Control Sample: 0.726 (0.653, 0.798). In both Models 1 and 2, the predictor variables derived from the stepwise procedure described above included gender, BMI, and family history of diabetes. In Model 3, indicator variables for E-selectin, WBC, and serum albumin were added.

Figure 1.

Receiver operating curves (ROC) from three different models*. The Western New York Study, 1996–2004.

A formal comparison between AUCs for Model 1 vs. Model 2 showed no significant difference (difference = 0.0008, s.e. = 0.01, 95% CI (−0.0187, 0.0204), P value = 0.93) indicating that the predictive probabilities from the full cohort and the case-control sample were consistent. A comparison between Model 2 and Model 3, i.e., without and with the emerging risk factors in the case/control sample, indicated that the addition of the three emerging risk factors (WBC, serum albumin, and E-selectin) significantly increased the AUC (difference = −0.0797, s.e. = 0.0382, 95% CI (−0.1545, −0.0048), P value = 0.04).

Discussion

In this community-wide population sample, we examined the association of several emerging risk factors with the risk of developing type 2 diabetes during an average follow-up of 5.9 years. We also examined whether novel biomarkers would give additional contribution to the risk prediction of type 2 diabetes obtained using the traditional risk factors alone. Findings indicate that biomarkers of endothelial dysfunction (i.e., E-selectin) and subclinical inflammation (i.e., serum albumin and WBC) are significant predictors of type 2 diabetes, independent of traditional risk factors; moreover, the addition of these emerging risk factors to a basic risk factor model significantly improved the prediction of type 2 diabetes (i.e., from 64.6 to 72.6%) in this cohort, though the use of ROCs and c-statistics has been recently questioned (18).

An increasing number of biomarkers have been proposed in the pathogenesis of type 2 diabetes. Our findings provide additional support to the growing evidence on the potential role of inflammation and endothelial dysfunction as important mechanisms in the etiopathogenesis of type 2 diabetes (5,6,7,8,19,20,21,22,23,24,25,26,27,28). Specifically, markers of chronic subclinical inflammation such as low levels of serum albumin and high levels of WBC were associated with an increased risk of type 2 diabetes. In contrast, for CRP and IL-6, the association with the risk of type 2 diabetes was not significant and greatly attenuated after adjustment for BMI and other traditional risk factors. Similar to our findings, several reports suggest that the associations between markers of inflammation and risk of type 2 diabetes may be significantly reduced after accounting for measures of obesity, and may be stronger among leaner individuals (5,6,23,25,28). It is plausible that in a generally overweight/obese population such as ours, CRP and IL-6 are not strongly associated with type 2 diabetes because of their relatively high correlation with body weight or body fat distribution (29,30), unlike endothelial factors. In fact, chronic, low-grade inflammation is one of the mediating mechanisms in the biologic pathway of obesity leading to the development of type 2 diabetes (31). It should be noted, however, that the inverse association between serum albumin, a marker of systemic inflammation, and risk of type 2 diabetes in this study was still significant after adjustment for BMI. Similar to our finding, others have found low levels of serum albumin to be associated with risk of type 2 diabetes, coronary heart disease, cancer, and overall mortality (5,32,33,34,35). Whether these observations reflect true associations or may be confounded by other conditions (e.g., renal or liver disease, malnutrition, anemia) that can cause hypoalbuminemia and increase the risk of type 2 diabetes and other chronic disease is unclear. Moreover, in this study, high levels of WBC were associated with an increased risk of type 2 diabetes, although not in a linear fashion. Adjustment for BMI had little effect on these results, consistent with findings from previous prospective analyses (5,36). WBC, however, is strongly influenced by smoking and may represent an indicator of smoking intensity (37). Unfortunately, we did not have adequate power to address this issue and perform stratified analyses by smoking status.

In this study, elevated levels of E-selectin, a biomarker of endothelial dysfunction, were associated with a strong, significant increased risk of developing type 2 diabetes. This association was independent of BMI and other traditional risk factors. Endothelial dysfunction may represent a further important mechanism in the etiopathogenesis of type 2 diabetes. These findings are consistent with recent studies reporting significant associations of biomarkers and other measures of endothelial dysfunction with the risk of type 2 diabetes (7,8,38,39,40). Finally in this study, levels of adiponectin were inversely related to the risk of type 2 diabetes. Although this association did not achieve statistical significance after accounting for BMI, the odds ratio suggests a strong effect. Several studies have reported an association between high levels of adiponectin and reduced risk of developing type 2 diabetes across different ethnic groups in both men and women (41,42,43,44,45,46).

Other investigators have developed prediction equations for type 2 diabetes (9,47,48,49). These authors have used a different set of predictor variables than ours, but are in general agreement that conventional risk factors, based on anamnestic data, may help to predict a large number of future diabetic patients. In this study, we found a significant improvement in the prediction of type 2 diabetes with the addition of three novel biomarkers (i.e., E-selectin, serum albumin, and leukocyte count) to a basic risk factor model including only traditional risk factors. It is possible that in a larger study in which other biomarkers could be included in the predictive model, the AUC would likely increase further; however, the degree of improvement will depend at least in part on the correlation between the basic model and the extended model (50).

Several limitations of this study deserve mention. First, although the incidence rate of type 2 diabetes is consistent with other studies of largely non-Hispanic white populations (2,3), there were a limited number of participants who became diabetic, thus limiting the power of these analyses. The interpretation of these analyses should thus be guided by both the effect size estimates as well as the results of formal statistical testing (50). Moreover, like most large-scale epidemiologic studies, we measured the risk factors at one occasion. A single measurement of the exposures of interest leaves open the possibility of misclassification here as well. Finally, data on the emerging biomarkers were not available for full cohort, so there is still possibility that the model obtained with these biomarkers for the full cohort may be less consistent than the basic model.

In conclusion, our results support the notion that endothelial dysfunction and subclinical inflammation may represent further important mechanisms in the etiopathogenesis of type 2 diabetes. Moreover, we found that novel biomarkers may improve the prediction of type 2 diabetes beyond the use of traditional risk factors alone.

Acknowledgment

This study was supported by grant NIH R01 DK60587 to R.P.D. We acknowledge the assistance of Mya Swanson in data management, file preparation, and analyses.

Disclosure

The authors declared no conflict of interest.

Ancillary