Figures
Abstract
Background
Adipose tissue inflammation and dysregulated adipokine secretion are implicated in obesity-related insulin resistance and type 2 diabetes. We evaluated the use of serum adiponectin, an anti-inflammatory adipokine, and several proinflammatory adipokines, as biomarkers of diabetes risk and whether they add to traditional risk factors in diabetes prediction.
Methods
We studied 1300 non-diabetic subjects from the prospective Hong Kong Cardiovascular Risk Factor Prevalence Study (CRISPS). Serum adiponectin, tumor necrosis factor-alpha receptor 2 (TNF-α R2), interleukin-6 (IL-6), adipocyte–fatty acid binding protein (A-FABP) and high-sensitivity C-reactive protein (hsCRP) were measured in baseline samples.
Results
Seventy-six participants developed diabetes over 5.3 years (median). All five biomarkers significantly improved the log-likelihood of diabetes in a clinical diabetes prediction (CDP) model including age, sex, family history of diabetes, smoking, physical activity, hypertension, waist circumference, fasting glucose and dyslipidaemia. In ROC curve analysis, “adiponectin + TNF-α R2” improved the area under ROC curve (AUC) of the CDP model from 0.802 to 0.830 (P = 0.03), rendering its performance comparable to the “CDP + 2-hour post-OGTT glucose” model (AUC = 0.852, P = 0.30). A biomarker risk score, derived from the number of biomarkers predictive of diabetes (low adiponectin, high TNF-α R2), had similar performance when added to the CDP model (AUC = 0.829 [95% CI: 0.808–0.849]).
Citation: Woo Y-C, Tso AWK, Xu A, Law LSC, Fong CHY, Lam T-H, et al. (2012) Combined Use of Serum Adiponectin and Tumor Necrosis Factor-Alpha Receptor 2 Levels Was Comparable to 2-Hour Post-Load Glucose in Diabetes Prediction. PLoS ONE 7(5): e36868. https://doi.org/10.1371/journal.pone.0036868
Editor: Yu Wang, The University of Hong Kong, Hong Kong
Received: February 9, 2012; Accepted: April 8, 2012; Published: May 16, 2012
Copyright: © 2012 Woo et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This study was supported by the Health and Health Services Research Fund (#06070951) and the General Research Fund of the Hong Kong Research Grant Council (HKU 780210M). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Diabetes mellitus is an increasing global health problem. In a recent study [1], the prevalence of diabetes in the Chinese population in 2008 had increased to 9.7% among adults aged 20 years or older. Another 12.9% of the subjects in this study had impaired glucose tolerance (IGT), known to be associated with a highly elevated risk of diabetes development among various populations including Hong Kong Chinese [2]. Without appropriate public health measures, it is anticipated that the increase in diabetes prevalence in China will reach epidemic dimensions in the near future. On a brighter note, it has been shown in long-term prospective studies that lifestyle interventions can delay or even prevent the onset of type 2 diabetes in high risk individuals including subjects with IGT [3], [4], [5]. To identify the presence of IGT, however, it is necessary to perform an oral glucose tolerance test (OGTT) which is cumbersome and not consistently reproducible. From a public health point of view, it would be very helpful if alternative, more easily assessed biomarkers with high predictive value for diabetes development can be identified.
Adipose tissue inflammation and dysregulated adipokine secretion have been implicated in obesity-related insulin resistance and type 2 diabetes [6], [7]. High levels of the pro-inflammatory biomarkers, such as interleukin-6 (IL-6), tumor necrosis factor-alpha (TNF-α) or its soluble receptor tumor necrosis factor-alpha receptor 2 (TNF-α R2), and C-reactive protein (CRP), are found in obese individuals. In Chinese, high level of high sensitive C-reactive protein (hsCRP) has been shown to independently predict the deterioration of glycaemia [8]. In addition, IL-6 and TNF-α, the upstream cytokines of C-reactive protein in the inflammatory cascades, have also been linked with an increased risk of type 2 diabetes in a previous report [9], although other studies have shown inconsistent results [10], [11], [12]. Our group has previously demonstrated that a high baseline level of adipocyte fatty acid binding protein (A-FABP), a pro-inflammatory adipokine, is also predictive of type 2 diabetes in a Chinese cohort [13]. Among the various biomarkers, adiponectin, an anti-inflammatory, insulin-sensitizing adipokine with reduced expression in obesity, has been consistently linked with protection from type 2 diabetes [10], [11], [12], [14]. We have demonstrated, in a previous nested case-control study, that hypoadiponectinaemia could predict persistent hyperglycaemia in Chinese [14]. In this 5-year prospective study of a population-based Chinese cohort, we examined the association of the above obesity-related biomarkers with diabetes development in the Chinese population and investigated if any of these biomarkers have the potential to replace the OGTT for risk prediction.
Methods
Ethics
The study was approved by the ethics committee of the Faculty of Medicine, University of Hong Kong. All subjects gave written informed consent.
Participants
The Hong Kong Cardiovascular Risk Factor Prevalence Study (CRISPS) is a population-based, long-term follow-up study on the prevalence of cardiovascular risk factors in Hong Kong. [2], [14], [15], [16] In 1995–1996 (CRISPS1), 2,895 unrelated Chinese subjects were selected randomly by their telephone numbers to undergo a comprehensive assessment, including a 75 g OGTT in all subjects not on antidiabetic medications. Subjects were contacted for reassessment in 2000–2004 (CRISPS2) and in 2005–2008 (CRISPS3). The current study included only the non-diabetic subjects identified in CRISPS2 (baseline for this study), as defined by the World Health Organization 1998 criteria [17]. 1300 subjects who attended both CRISPS2 (baseline visit) and CRISPS3 (follow-up visit) and had complete baseline anthropometric and biochemical data were included for analysis. Subjects who had FG ≥7 mmol/L or 2-hG ≥11.1 mmol/L [17] at CRISPS3 or had been diagnosed to have diabetes between CRISPS2 and CRISPS3 were considered as incident cases of type 2 diabetes. At each attendance, medical histories were reviewed in detail. Anthropometric and biochemical parameters were measured as described previously [2], [14], [15], [16]. The presence of hypertension was defined as blood pressure ≥130/85 mmHg or receiving regular antihypertensive treatment. The presence of dyslipidaemia was defined as having high triglyceride (fasting triglycerides ≥1.69 mmol/L), low HDL-C (fasting HDL-C <1.29 mmol/L in women and <1.04 mmol/L in men), high LDL-C (fasting LDL-C ≥3.4 mmol/L) [18] or taking lipid lowering agents. Insulin resistance was estimated using the homeostasis model assessment index of insulin resistance (HOMA-IR), calculated by the formula (FG in mmol/L × fasting insulin in mIU/L /22.5) [19].
Biochemical Assessments
Serum levels of the biomarkers were measured from stored serum samples collected at CRISPS2 (baseline of this study). Serum A-FABP [13], IL-6 and TNF-α R2 were measured with commercial ELISA kits (BioVendor – Laboratory Medicine, Modrice, Czech Republic for A-FABP; Bender MedSystems GmbH, Vienna, Austria for IL-6; R&D Systems, Inc., Minneapolis, USA for TNF-α R2). hsCRP and total adiponectin levels were measured using in-house sandwich ELISA assays established in our laboratory as described previously [14].
Statistical Analyses
All statistical analyses were performed with SPSS Statistics 19 (SPSS, Chicago, IL). Results are presented as means ± SD or medians (interquartile range, IQR) as appropriate. Data that were not normally distributed, as determined using Kolmogorox-Smirnov test, were natural-logarithmically transformed to obtain near normality before analysis. A-FABP, IL-6, TNF-α R2 and adiponectin levels showed gender-specific dimorphisms and were sex-adjusted. Differences in baseline characteristics with subsequent glycaemia status were compared using χ2 tests for categorical variables and one-way ANOVA for continuous variables. The level of each biomarker was divided into high and low categories by an optimal cutoff derived from Youden Index (sensitivity + specificity −1) [20]. Multiple logistic regression was applied to estimate the odds ratio (OR) and 95% confidence interval (CI) of each biomarker by comparison of the two categories for incident diabetes. We used the low level as the reference for TNF-α R2, IL-6, hsCRP and A-FABP; and the high level as reference for adiponectin. In multivariate analyses, we adjusted for clinical parameters and conventional risk factors including age, sex, family history of diabetes in first degree relatives, smoking status, physical activity, hypertension, waist circumference (WC), fasting glucose (FG) and dyslipidaemia. Log-likelihood ratio test was used to compare the likelihood of incident diabetes before and after addition of one, followed by multiple, biomarker levels to the clinical diabetes prediction model (CDP), which included the aforementioned clinical parameters and diabetes risk factors. A biomarker risk score was counted by the number of biomarkers predictive of diabetes. Receiver Operating Characteristic (ROC) curves were used to assess the performance of different methods for diabetes risk prediction. The area under ROC curves (AUC) were compared with a nonparametric approach as described by DeLong et al [21]. Sensitivities, specificities, and positive and negative predictive values for the different prediction models were determined. Two-sided p-values less than 0.05 were considered statistically significant.
Results
Seventy-six of 1300 subjects (5.85%) had developed type 2 diabetes when reassessed after a median interval of 5.3 (inter-quartile range: 4.59–5.80) years. Table 1 shows that at baseline, compared to those without incident diabetes at follow-up, subjects with incident diabetes were significantly older, more obese, had greater BMI and WC, higher FG, 2 h-glucose, insulin, HOMA-IR, triglyceride and LDL-C, but lower level of HDL-C. They were more likely to have hypertension, dyslipidaemia, IGT or impaired fasting glucose (IFG) [17]. For the serum biomarkers, subjects with incident diabetes had lower levels of adiponectin (P = 0.001), but higher levels of hsCRP (P<0.001), IL-6 (P = 0.014), TNF-α R2 (P<0.001) and A-FABP (P<0.001).
The levels of biomarkers were divided into low and high categories by their respective optimal cutoffs, i.e. 5.94 ug/ml (M) and 6.03 ug/ml (F) for adiponectin; 2.13 ng/ml (M) and 2.05 ng/ml (F) for TNF-α R2; 21.72 ng/ml (M) and 26.33 ng/ml (F) for A-FABP; 0.71 pg/ml (M) and 0.34 pg/ml (F) for IL-6, and 0.62 mg/L for hsCRP. Each of the five serum biomarkers was tested individually by multivariate analysis to examine for its independent association with incident diabetes, with adjustments made for age, sex, family history of diabetes, smoking status, physical activity, hypertension, WC, FG and dyslipidaemia. All five biomarkers were, individually, significantly associated with incident diabetes. The ORs for hypoadiponectinaemia, TNF-α R2, A-FABP, IL-6 and hsCRP were 2.62 (P = 0.001), 2.19 (P = 0.004), 2.26 (P = 0.005), 2.25 (P = 0.009) and 2.26 (P = 0.010) respectively.
We constructed a clinical diabetes prediction model (CDP) using patients’ baseline demographic parameters and conventional risk factors (age, sex, WC, FG, smoking status, physical activities, family history of diabetes, presence of hypertension and dyslipidaemia). Effect of the addition of biomarkers to the CDP model was shown in Table 2. The largest increment in the log-likelihood of the model was observed after the addition of adiponectin. Addition of each of the other biomarkers to this “CDP + adiponectin” model resulted in further significant increase in log-likelihood (Table 2). Performance of the prediction models was examined by ROC curves analysis. Table 3 showed the comparison of the AUCs of different “CDP + biomarker(s)” models with the baseline CDP model and also the “CDP + 2 h-glucose” model. Adiponectin together with TNF-α R2, when added to the CDP, generated a model with an AUC of 0.830 (95% CI: 0.808–0.850), which was significantly better than the baseline CDP (AUC = 0.802, P = 0.0306) and was comparable to the “CDP + 2 h-glucose” model (AUC = 0.852 [95% CI: 0.831–0.870], P = 0.2954) (Fig. 1) (Table 3). This combination required the smallest number of biomarkers to achieve a performance comparable to 2 h-glucose. Adding fasting insulin level, a marker of insulin resistance, to the CDP did not show significant AUC improvement (AUC = 0.803 [95% CI: 0.780–0.824], P = 0.65). A biomarker risk score, counted by the number of biomarkers predictive of diabetes, i.e. low adiponectin and/or high TNF-α R2, was analyzed by multiple logistic regression as shown in Table 4. The ORs were 2.802 (P = 0.006) and 5.862 (P<0.001) for a score of 1 and 2 respectively. The AUC of the “CDP + biomarker risk score” model was 0.829 (95% CI: 0.808–0.849) which was similar to the “CDP + adiponectin + TNF-α R2” model. The sensitivity and specificity for diabetes prediction of the “CDP + biomarker risk score” model were 70.7% and 83.3% respectively. This model increased the positive predictive value from 21.5% to 31.3% without jeopardizing the negative predictive value (96.4% vs. 97.3%) when compared with the baseline CDP model (Table 5).
CDP, clinical diabetes prediction model.
Discussion
Various serum biomarkers representing the process of adipokine dysregulation and adipose tissue inflammation have been shown to be associated with the development of obesity-related type 2 diabetes. To look for the best predictive obesity-related biomarkers for type 2 diabetes in this 5-year prospective study among Southern Chinese, we focused on biomarkers which were previously shown to be associated with incident diabetes or worsening of glycaemic status in Chinese subjects. Adiponectin [14], A-FABP [13] and hsCRP [8] were therefore selected. IL-6 and TNF-α R2 were also included, being adipokines upstream of CRP in the inflammatory cascades. Individually, low adiponectin, high A-FABP, TNF-α R2 (a surrogate marker of TNF-α), IL-6 or hsCRP at baseline was independently predictive of 5-year diabetes risk in this cohort, after adjustment for nine conventional risk factors, including age, sex, family history of diabetes, smoking, physical inactivity, hypertension, waist circumference, fasting glucose and dyslipidaemia. Each of these biomarkers could increase the likelihood of diabetes development when added to a clinical prediction model comprising these conventional risk factors, with the increase being greatest following the addition of adiponectin. Based on ROC analysis, the combined use of serum adiponectin and TNF-α R2 resulted in a significant enhancement of diabetes prediction by this clinical prediction model, with the enhancement being comparable to that provided by 2-hour plasma glucose, as assessed with a 75 g OGTT.
The association of IL-6 with insulin resistance was suggested by the observation in animal studies that passive immunoneutralisation of IL-6 led to an improved insulin sensitivity in insulin resistant mice with transgenic NFκB activation [22]. Accordingly, elevated levels of IL-6 also predicted an increased risk of diabetes in postmenopausal women [9] and in the current study. On the contrary, however, blocking the action of IL-6 in rheumatoid arthritis patients led to enhanced plasma glucose levels [23], and IL-6 infusion acutely increased insulin-stimulated glucose disposal in humans, probably via AMP-activated protein kinase in skeletal muscles [24]. The role of IL-6 in insulin resistance in humans thus remains to be resolved. It has also been suggested that, as TNF-α can trigger IL-6 release, increased systemic IL-6 levels may reflect enhanced adipose tissue production of TNF-α, the actual driver behind obesity-related insulin resistance [23]. High hsCRP was also shown to be associated with an increased diabetes risk, when analyzed individually, in some previous studies. [8], [25], [26], [27] However, heterogeneity existed as was documented by a recent meta-analysis of 16 published studies [27], with the authors concluding that CRP may not be an independent risk factor for type 2 diabetes, as it is a downstream marker of the inflammatory process, stimulated by pro-inflammatory cytokines including IL-6 and TNF-α, the secretion of which are reciprocally regulated by adiponectin [28]. In the current study, hsCRP and IL-6 were not as useful as TNF-α R2, when used in addition to adiponectin for the prediction of diabetes risk. Similarly, serum A-FABP levels, shown to be associated with various cardiometabolic risk factors in Chinese [13] and Caucasians [29], and independently predictive of diabetes in a CRISPS subcohort [13], appeared to be less useful than TNF-α R2, when used in conjunction with adiponectin, in predicting diabetes development in this larger cohort. The mechanisms underlying the dysglycaemic effect of A-FABP are not fully understood although animal studies suggest several potential mechanisms [28]. In mice, genetic deficiency of A-FABP is associated with enhanced insulin signaling, in part via a reduction of TNF-α, IL-6 and other pro-inflammatory cytokines [28].
TNF-α R2 and adiponectin emerged as the better combination of biomarkers in our diabetes prediction models. TNF-α has been considered as a key mediator of obesity-related insulin resistance because of its increased expression in obesity and its inhibitory effect on insulin receptor signaling [30]. While earlier reports failed to demonstrate the association of serum TNF-α level with diabetes development [10], [12], a more recent study demonstrated that raised serum TNF-α R2 levels, measured as a surrogate marker for TNF-α because of its superior sensitivity and reliability when assayed in frozen plasma, were associated with a modest increase in diabetes risk [9]. Adiponectin, on the other hand, is an adipokine with insulin-sensitizing, anti-inflammatory and vasoprotective properties which has been extensively studied and reviewed in the past decade [28], [31]. Since the first report of its role in diabetes development in a case-control study of Pima Indians [32], an association between low adiponectin and the risk of diabetes has been consistently reported across diverse populations [11], [12], [26], [33], [34], [35]. Adiponectin has also been selected for construction of diabetes prediction models in previous studies [26], [36]. In a German study [26], adiponectin modestly increased the ROC AUC by 0.011 to 0.831 when added to a basic model which included already the inflammatory markers CRP, IL-6, soluble ICAM-1 and soluble E-selectin levels on top of demographic and lifestyle factors, family history of diabetes, blood pressure, lipid levels. In the Inter99 cohort, adiponectin, together with CRP, ferritin, IL-2 receptor A, fasting glucose and insulin, performed similarly as 2-hour plasma glucose (OGTT) or 2-hour insulin in ROC AUC analysis. In this study we demonstrated that only two obesity-related biomarkers, namely adiponectin and TNF-α R2, were required to achieve an effect comparable to 2-hour glucose, when added to a predictive model consisting of non-invasively assessable clinical parameters together with fasting glucose and lipid levels.
Different diabetes prediction models have been published previously. They could be derived from clinical parameters [37], serum biomarkers [36], or a combination of both [38]. The 2-hour post-OGTT plasma glucose has been shown to be the strongest single predictor for diabetes development in Southern Chinese [39]. Although the 2-hour glucose, when used alone, was shown to be inferior to a model derived from readily available clinical variables in the San Antonio Study (77.5% vs. 84.3%) [40], or equivalent to a diabetes prediction score involving six biomarkers [36] in the Inter99 cohort, it is noteworthy that the 2-hour glucose did further improve diabetes prediction based on clinical variables in the San Antonio Study [40], and even more so in the current study. Nonetheless, the OGTT is notorious for being inconvenient, cumbersome and associated with large intra-individual variation in glucose responses. In our pursuit of a potential alternative to 2-hour glucose we aimed at identifying biomarkers comparable to 2-hour glucose in improving the performance of our clinical diabetes prediction model. Fasting insulin, a marker of insulin resistance, was tested but did not result in significant improvement in diabetes risk prediction. In the process of model development, we categorized the biomarkers to low and high levels, with reference to mathematically derived optimal cut-offs, and formulated a simple biomarker risk score by counting the number of biomarkers predictive of diabetes development. As only two biomarkers were needed in our model, the biomarker risk score was simply 0, 1 or 2. This biomarker risk score was also shown to be an independent predictor of diabetes risk in our cohort and improved the performance of the clinical prediction model to a similar level as the addition of 2-hour glucose.
As our study is limited by the relatively small number of subjects with incident diabetes, we have not performed model validation in the current report. In addition, laboratory measurements for adiponectin and TNF-α R2 are still not readily available in clinical practice. However, our postulated diabetes prediction model involved only simple anthropometric measurements and a single fasting blood sample for identifying patients at increased risk of developing type 2 diabetes, which would be more convenient than the conventional oral glucose tolerance test. In summary, our data highlight the potential combined use of serum adiponectin and TNF-α R2 for diabetes prediction model construction in the Chinese population, as an attractive alternative to existing methods which warrant further validation in other populations. Identifying individuals at risk using simple and convenient prediction tools, followed by the commencement of strategic preventive measures, should be a useful approach to halt the epidemic of type 2 diabetes.
Acknowledgments
The authors wish to thank the participants of the Hong Kong Cardiovascular Risk Factors Prevalence Study.
Author Contributions
Conceived and designed the experiments: YCW AWKT BMYC KSLL. Analyzed the data: YCW CHYF LSCL KSLL. Contributed reagents/materials/analysis tools: KSLL. Wrote the paper: YCW. Investigators of the Hong Kong Cardiovascular Risk Factors Prevalence Study: NMSW THL BMYC KSLL. Data collection and advice on experiments and data analysis: AX SVL CHYF LSCL. Critical revision of the manuscript: AWKT AX THL SVL BMYC KSLL. Initiated and supervised the study: BMYC KSLL.
References
- 1. Yang W, Lu J, Weng J, Jia W, Ji L, et al. (2010) Prevalence of diabetes among men and women in China. N Engl J Med 362: 1090–1101.
- 2. Wat NM, Lam TH, Janus ED, Lam KS (2001) Central obesity predicts the worsening of glycemia in southern Chinese. Int J Obes Relat Metab Disord 25: 1789–1793.
- 3. Knowler WC, Barrett-Connor E, Fowler SE, Hamman RF, Lachin JM, et al. (2002) Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. N Engl J Med 346: 393–403.
- 4. Lindstrom J, Ilanne-Parikka P, Peltonen M, Aunola S, Eriksson JG, et al. (2006) Sustained reduction in the incidence of type 2 diabetes by lifestyle intervention: follow-up of the Finnish Diabetes Prevention Study. Lancet 368: 1673–1679.
- 5. Li G, Zhang P, Wang J, Gregg EW, Yang W, et al. (2008) The long-term effect of lifestyle interventions to prevent diabetes in the China Da Qing Diabetes Prevention Study: a 20-year follow-up study. Lancet 371: 1783–1789.
- 6. Shoelson SE, Lee J, Goldfine AB (2006) Inflammation and insulin resistance. J Clin Invest 116: 1793–1801.
- 7. Zhang X, Xu A, Chung SK, Cresser JH, Sweeney G, et al. (2011) Selective inactivation of c-Jun NH2-terminal kinase in adipose tissue protects against diet-induced obesity and improves insulin sensitivity in both liver and skeletal muscle in mice. Diabetes 60: 486–495.
- 8.
Ong KL, Tso AW, Xu A, Law LS, Li M, et al. (2011) Evaluation of the combined use of adiponectin and C-reactive protein levels as biomarkers for predicting the deterioration in glycaemia after a median of 5.4 years. Diabetologia.
- 9. Liu S, Tinker L, Song Y, Rifai N, Bonds DE, et al. (2007) A prospective study of inflammatory cytokines and diabetes mellitus in a multiethnic cohort of postmenopausal women. Arch Intern Med 167: 1676–1685.
- 10. Krakoff J, Funahashi T, Stehouwer CD, Schalkwijk CG, Tanaka S, et al. (2003) Inflammatory markers, adiponectin, and risk of type 2 diabetes in the Pima Indian. Diabetes Care 26: 1745–1751.
- 11. Ley SH, Harris SB, Connelly PW, Mamakeesick M, Gittelsohn J, et al. (2008) Adipokines and incident type 2 diabetes in an Aboriginal Canadian [corrected] population: the Sandy Lake Health and Diabetes Project. Diabetes Care 31: 1410–1415.
- 12. Daimon M, Oizumi T, Saitoh T, Kameda W, Hirata A, et al. (2003) Decreased serum levels of adiponectin are a risk factor for the progression to type 2 diabetes in the Japanese Population: the Funagata study. Diabetes Care 26: 2015–2020.
- 13. Tso AW, Xu A, Sham PC, Wat NM, Wang Y, et al. (2007) Serum adipocyte fatty acid binding protein as a new biomarker predicting the development of type 2 diabetes: a 10-year prospective study in a Chinese cohort. Diabetes Care 30: 2667–2672.
- 14. Tso AW, Sham PC, Wat NM, Xu A, Cheung BM, et al. (2006) Polymorphisms of the gene encoding adiponectin and glycaemic outcome of Chinese subjects with impaired glucose tolerance: a 5-year follow-up study. Diabetologia 49: 1806–1815.
- 15. Tan KC, Wat NM, Tam SC, Janus ED, Lam TH, et al. (2003) C-reactive protein predicts the deterioration of glycemia in chinese subjects with impaired glucose tolerance. Diabetes Care 26: 2323–2328.
- 16. Xu A, Tso AW, Cheung BM, Wang Y, Wat NM, et al. (2007) Circulating adipocyte-fatty acid binding protein levels predict the development of the metabolic syndrome: a 5-year prospective study. Circulation 115: 1537–1543.
- 17. Alberti KG, Zimmet PZ (1998) Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation. Diabet Med 15: 539–553.
- 18. Expert Panel on Detection, Evaluation , Treatment of High Blood Cholesterol in Adults (2001) Executive Summary of The Third Report of The National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, And Treatment of High Blood Cholesterol In Adults (Adult Treatment Panel III). JAMA 285: 2486–2497.
- 19. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, et al. (1985) Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28: 412–419.
- 20. Perkins NJ, Schisterman EF (2006) The inconsistency of “optimal” cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am J Epidemiol 163: 670–675.
- 21. DeLong ER, DeLong DM, Clarke-Pearson DL (1988) Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 44: 837–845.
- 22. Cai D, Yuan M, Frantz DF, Melendez PA, Hansen L, et al. (2005) Local and systemic insulin resistance resulting from hepatic activation of IKK-beta and NF-kappaB. Nat Med 11: 183–190.
- 23. Pedersen BK, Febbraio MA (2007) Point: Interleukin-6 does have a beneficial role in insulin sensitivity and glucose homeostasis. J Appl Physiol 102: 814–816.
- 24. Carey AL, Steinberg GR, Macaulay SL, Thomas WG, Holmes AG, et al. (2006) Interleukin-6 increases insulin-stimulated glucose disposal in humans and glucose uptake and fatty acid oxidation in vitro via AMP-activated protein kinase. Diabetes 55: 2688–2697.
- 25. McNeely MJ, Boyko EJ, Weigle DS, Shofer JB, Chessler SD, et al. (1999) Association between baseline plasma leptin levels and subsequent development of diabetes in Japanese Americans. Diabetes Care 22: 65–70.
- 26. Thorand B, Zierer A, Baumert J, Meisinger C, Herder C, et al. (2010) Associations between leptin and the leptin / adiponectin ratio and incident Type 2 diabetes in middle-aged men and women: results from the MONICA / KORA Augsburg study 1984–2002. Diabet Med 27: 1004–1011.
- 27. Lee CC, Adler AI, Sandhu MS, Sharp SJ, Forouhi NG, et al. (2009) Association of C-reactive protein with type 2 diabetes: prospective analysis and meta-analysis. Diabetologia 52: 1040–1047.
- 28. Tso AW, Xu A, Chow WS, Lam KS (2008) Adipose tissue and the metabolic syndrome: focusing on adiponectin and several novel adipokines. Biomark Med 2: 239–252.
- 29. Stejskal D, Karpisek M (2006) Adipocyte fatty acid binding protein in a Caucasian population: a new marker of metabolic syndrome? Eur J Clin Invest 36: 621–625.
- 30. Hotamisligil GS (1999) The role of TNFalpha and TNF receptors in obesity and insulin resistance. J Intern Med 245: 621–625.
- 31. Kadowaki T, Yamauchi T, Kubota N, Hara K, Ueki K, et al. (2006) Adiponectin and adiponectin receptors in insulin resistance, diabetes, and the metabolic syndrome. J Clin Invest 116: 1784–1792.
- 32. Lindsay RS, Funahashi T, Hanson RL, Matsuzawa Y, Tanaka S, et al. (2002) Adiponectin and development of type 2 diabetes in the Pima Indian population. Lancet 360: 57–58.
- 33. Snehalatha C, Mukesh B, Simon M, Viswanathan V, Haffner SM, et al. (2003) Plasma adiponectin is an independent predictor of type 2 diabetes in Asian indians. Diabetes Care 26: 3226–3229.
- 34. Choi KM, Lee J, Lee KW, Seo JA, Oh JH, et al. (2004) Serum adiponectin concentrations predict the developments of type 2 diabetes and the metabolic syndrome in elderly Koreans. Clin Endocrinol (Oxf) 61: 75–80.
- 35. Li S, Shin HJ, Ding EL, van Dam RM (2009) Adiponectin levels and risk of type 2 diabetes: a systematic review and meta-analysis. JAMA 302: 179–188.
- 36. Kolberg JA, Jorgensen T, Gerwien RW, Hamren S, McKenna MP, et al. (2009) Development of a type 2 diabetes risk model from a panel of serum biomarkers from the Inter99 cohort. Diabetes Care 32: 1207–1212.
- 37. Wilson PW, Meigs JB, Sullivan L, Fox CS, Nathan DM, et al. (2007) Prediction of incident diabetes mellitus in middle-aged adults: the Framingham Offspring Study. Arch Intern Med 167: 1068–1074.
- 38. Balkau B, Lange C, Fezeu L, Tichet J, de Lauzon-Guillain B, et al. (2008) Predicting diabetes: clinical, biological, and genetic approaches: data from the Epidemiological Study on the Insulin Resistance Syndrome (DESIR). Diabetes Care 31: 2056–2061.
- 39. Law LS, Tso AW, Tam S, Wat NM, Cheung BM, et al. (2011) Haemoglobin A1c is superior to fasting glucose in predicting the incidence of diabetes over 8 years among Chinese. Diabetes Res Clin Pract 91: e53–56.
- 40. Stern MP, Williams K, Haffner SM (2002) Identification of persons at high risk for type 2 diabetes mellitus: do we need the oral glucose tolerance test? Ann Intern Med 136: 575–581.