This is a nested case control study based on a larger prospective cohort. The study population participated in an epidemiological survey, the Stockholm Diabetes Prevention Program. The design of the study has been described previously (31). Briefly, individuals aged 35-56 years without known diabetes were recruited for the baseline study performed in 1992-94 for men and in 1996-98 for women. Due to the study design, a selection was made so that the study cohort was enriched with participants with FHD (see Supplemental Flow Diagram ). Thus, at baseline about 50% of the participants had a positive FHD, defined as known diabetes in at least one first-degree relative (father, mother, brother or sister) or at least two second-degree relatives (grand-parents, uncles or aunts). Eight to ten years later; in 2002-2004 for men and in 2004-2006 for women, a follow-up study was undertaken. All baseline participants were invited with the exception of those, who were diagnosed with T2D at baseline, had moved out of the Stockholm area or were deceased. In total 4,821 women and 3,128 men participated in the baseline study. Corresponding figures for the follow-up study were 3,329 (76% of invited) and 2,383 (87% of invited), respectively. Baseline and follow-up studies consisted of an extensive questionnaire covering lifestyle factors, a health examination including measurements of blood pressure, weight, height, hip and waist. Blood samples were taken after an overnight fast and 2-h after drinking 75 g glucose (oral glucose tolerance test (OGTT)). Individuals, who progressed from normal glucose tolerance (NGT) at baseline to AGT at follow-up, were selected as cases. Thus, cases consisted of persons who developed prediabetes (IGT and IFG+IGT) (133 women, 164 men) or overt T2D (55 women, 98 men). A similar number of controls were randomly selected among subjects having NGT and negative FHD at both baseline and follow-up (200 women, 277 men), matched to cases by sex and age in five-year interval. Only subjects having data on all variables were included in the analyses.

The study, approved by the Ethics Committee at Karolinska University Hospital, was carried out in accordance with the Declaration of Helsinki. All participants gave informed consent.

Individuals were categorized after an OGTT according to 1999 WHO criteria (WHO 1999) as having NGT, impaired fasting glucose (IFG, between 6.1 and 7.0 mmol/L), impaired glucose tolerance (IGT, 2-h value between 7.8 and 11.1 mmol/L) or T2D.

Subjects having IGT or IFG plus IGT were defined as having prediabetes. Moreover, subjects classified as having T2D at follow-up are those diagnosed at the follow-up examination as well as those, who were diagnosed by a physician during the period between baseline and follow-up examinations. The physicians in primary care diagnosed T2D when the fasting glucose levels were higher than 7 mmol/l at two or more occasions or symptoms accompanied by glucose levels higher than 11.1 mmol/l or HbA1c above 47 mmol/mol. All subjects categorized as having prediabetes in the present study were newly diagnosed by the OGTT at the follow up occasion.

Body mass index (BMI, kg/m²) was divided into three groups (<25.0, 25.0-29.9 and ≥30.0). Similarly, waist was categorized in three groups: <80, 80-87, and >87 cm for women, and <94, 94-101, and >101 cm for men. Physical activity during leisure time and smoking, based on the response alternatives were categorized into three groups ( Table 1 ). Socioeconomic position, based on self-reported occupation, classified according to the standard system from Statistics Sweden, was categorized into four groups (unskilled/skilled manual workers, low-level non-manual employees, medium- and high-level non-manual employees and self-employed/farmers). Education comprised three categories; low (elementary school), middle (senior high school, technical and vocational school), and high (college, university). Blood pressure was dichotomized into normal (diastolic and systolic blood pressure below 90 and 140 mmHg, respectively, without hypertension treatment) and high (systolic blood pressure ≥140 and/or diastolic blood pressure ≥ 90 mmHg and/or on anti-hypertensive treatment).

All samples were assayed in duplicate. Serum IGFBP-1 was measured by an in-house radioimmunoassay (RIA) using a polyclonal antibody and human IGFBP-1 as standard (32). Intra- and inter-assay CV were 3% and 10%, respectively. Serum IGFBP-2 was determined by an in-house time-resolved immunofluorometric assay (TR-IFMA) based on commercial reagents as previously described, with intra- and inter-assay CVs averaging 5 and 12%, respectively (33). Serum adiponectin was determined by an in-house TR-IFMA based on commercial antibodies as previously described, with intra- and inter-assay CVs averaging <5 and <13%, respectively (34). This assay was later demonstrated to detect all major forms of adiponectin in human serum (high-molecular, medium molecular and low molecular weight forms) (34). Serum IGF-II was determined following acid ethanol extraction of serum, using a highly sensitive TR-IFMA as previously described, intra- and inter-assay CVs averaging <5 and <15%, respectively (35). IGF-I was measured in serum by RIA after acid-ethanol extraction and cryo-precipitation, using des-(1-3)-IGF-I as tracer to minimize interference by IGFBPs (36). Intra- and inter-assay CV were 4% and 11%, respectively.

Venous serum glucose was assayed using the glucose oxidase method (Yellow Springs Glucose Analyzer, Yellow Springs, OH, USA).

Immunoreactive insulin was assayed by in-house RIA, using a polyclonal antibody, human insulin as standard and charcoal addition to separate antibody-bound and free insulin. The intra-assay CV was 5.8-8.4% and the inter-assay CV was 11.5-16.9%. In this assay proinsulin has a 100% cross reactivity (37).

Results are presented separately for women and men. Clinical characteristics are presented as means and 95% confidence intervals (CI). Variables not normally distributed (i.e. adiponectin, IGFBP-1, IGFBP-2, IGF-I, IGF-II and insulin) were log-transformed prior to analyses. Differences between more than two groups were for continuous variables analyzed by one-way ANOVA and if significant, followed by Scheffé post hoc test and for categorical variables by chi-square test. Paired t-test was performed when comparing two occasions within group and unpaired t-test when comparing two groups. Odds ratios (ORs) together with 95% confidence intervals (CIs) were calculated by multiple logistic regression analyses to explore associations between serum variables and an abnormal glucose regulation.

Potential confounders (BMI, waist, physical activity during leisure time, smoking, socioeconomic position, education and blood pressure) were tested separately in logistic regression models including the exposure variable and age. The change-in-estimate method was used, meaning that variables that contributed to at least a 10% change of the age-adjusted crude estimate in any of the outcome measures were included in the final multi-adjusted model. Waist and BMI, just like socioeconomic position and education, had similar influence on the crude estimates and accordingly BMI and socioeconomic position were chosen for the final model. Thus, two logistic regression models are given; model 1 adjusted for age and model 2 adjusted for age, BMI, physical activity, smoking, socioeconomic position and blood pressure.

In the regression model serum variables are categorized in quartiles according to the distribution within sex or used as continuous ²log-transformed variables and thus reported per halving (adiponectin, IGFBP-1 and IGFBP-2), or per doubling (IGF-I and IGF-II). Co-linearity between biomarkers was evaluated by estimating variance inflation factor (VIF) which was found to not exceed 2.0.

Receiver operating characteristics (ROC) curve analyses were used to evaluate and compare the predictive power of the biomarkers. In PROC LOGISTIC (SAS statistical package) the ROCCONTRAST statement, implementing a nonparametric approach, was used to analyze differences between the respective areas under the curve (AUC). The analyses were performed using SAS statistical package version 9.2 for Windows (SAS Institute Inc., Cary, NC, USA) and Statistica StatSoft version 10 (Tulsa, OK, USA). P-values <0.05 were considered statistically significant.

At baseline, all subjects had normal fasting glucose, OGTT and blood pressure. Significant differences were observed in all continuous parameters except for age and serum IGF-II levels in those who later developed prediabetes or T2D compared to the controls, who remained NGT ( Table 1 ). FHD was reported at baseline in 71% of women, who later developed prediabetes, and in 80% of women, who developed T2D. BMI and waist circumference were at baseline progressively higher in those, who later developed prediabetes or T2D. Conversely, baseline levels of adiponectin, IGFBP-1 and IGFBP-2 were significantly lower in individuals, who later developed prediabetes or T2D compared to the controls.

Those who later developed T2D were at baseline more sedentary with less regular physical activity and they had lower socioeconomic status compared to the controls ( Table 1 ). Hypertension was more common in those who later developed prediabetes or T2D.

Eight years later at follow up the number with known FHD had increased to 75% of those with prediabetes and to 84% of those with T2D.

After adjustment for known risk factors, significant ORs for future T2D were obtained for the lowest quartiles of adiponectin (OR:29.42), IGFBP-2 (OR:9.52) and IGFBP-1 (OR:5.41), and for the highest quartile of IGF-I (OR:3.79). IGF-II did not predict T2D. For the prediction of prediabetes, significant ORs were obtained for all five proteins, the strongest for IGFBP-2 (OR:7,48) and adiponectin (OR:6,06) ( Table 2 ).

The OR for T2D associated with low adiponectin was not affected by the adjustment for confounding risk factors in contrast to that of IGFBP-1 and -2, which decreased after extended adjustment ( Table 2 and Supplemental Table S1 ).

The control men, who stayed NGT, demonstrated compared to the control women higher concentrations of insulin and IGF-II (p<0.001) and lower levels of fasting adiponectin, IGFBP-1 and IGFBP-2 (p<0.001) ( Table 1 ). The glucose levels were not different.

At baseline all subjects had normal fasting glucose, OGTT and blood pressure. Significant differences were observed in all continuous parameters except for age and serum IGF-I levels in those NGT men, who later developed prediabetes or T2D compared to controls, who stayed NGT ( Table 1 ). At baseline FHD was reported in 63% of men, who later developed prediabetes, and in 77% of the men, who developed T2D. BMI and waist measurements were higher in those who later developed prediabetes compared to controls, but in contrast to women, with no further increments in those who later developed T2D. Fasting and 2h OGTT plasma glucose levels differed at baseline between men, who later developed prediabetes or T2D ( Table 1 ). As for women, levels of adiponectin, IGFBP-1 and IGFBP-2 were significantly lower at baseline in those who later developed prediabetes or T2D as compared to controls.

Among men, who later developed T2D, there were more sedentary activities and more current smokers ( Table 1 ). Hypertension was more common in those, who later developed prediabetes or T2D.

At follow up, 10 years later, the number with known FHD had increased to 66% of those with prediabetes and to 82% of those with T2D.

After adjustment for known risk factors, significant ORs for future T2D were obtained for the lowest quartiles of IGFBP-1 (OR:14.89), IGFBP-2 (OR:7,63) and adiponectin (OR: 2.35). Neither IGF-I nor IGF-II predicted AGT. For the prediction of prediabetes, only IGFBP-1 (OR:13.44) and IGFBP-2 (OR:4.03) yielded significant ORs ( Table 2 ). Extended adjustment for confounding risk factors did not significantly alter the ORs for T2D or prediabetes ( Table 2 and Supplemental Table S1 ).

In both women and men, low serum levels of adiponectin, IGFBP-1 and IGFBP-2 showed a concentration dependent risk to develop T2D ( Table 2 ). When included in the same regression model and following multivariate analysis and adjustment of established risk factors, only adiponectin and IGFBP-2 remained significantly and independently associated with future prediabetes and T2D in women, while only IGFBP-1 remained associated with future prediabetes and T2D in men ( Table 3 ).

ROC curves for the three variables adiponectin, IGFBP-1 and IGFBP-2 were evaluated in comparison with the ROC curve for BMI. In women, the AUC-ROCs were; 0.81, 0.79, and 0.79 for adiponectin, IGFBP-1, and IGFBP-2, respectively, compared to 0.80 for BMI ( Figure 1A ). These AUCs did not differ significantly. In men, the corresponding AUC-ROCs were 0.64, 0.78, 0.71, respectively, and for BMI 0.69 with the value for IGFBP-1 being significantly higher compared to the other proteins (p<0.001 to p=0.003). When adiponectin, IGFBP-1, and IGFBP-2 were combined the ROC-AUC reached 0.87 for women and 0.79 for men, which was significantly higher than for BMI alone, p=0.049 for women and p=0.002 for men. Including BMI in the model did not markedly change the AUC ROCs ( Figure 1B ).

Thus, in women, the sensitivity was similar for BMI and the three biomarkers. However, when adiponectin, IGFBP-1 and IGFBP-2 were combined, they were together more sensitive than BMI. In contrast to women, in men IGFBP-1 showed a significantly higher sensitivity than adiponectin, IGFBP-2 or BMI. When combined the three biomarkers were as in women resulting in a higher sensitivity than BMI.