Introduction

A growing body of evidence from observational studies suggests that low serum concentrations of vitamin D, measured as 25-hydroxyvitamin D [25(OH)D], are associated with type 2 diabetes and the metabolic syndrome, although the relationship has not been shown to be entirely consistent (see [1–6] for a review). As it is a major challenge to find strategies for the early detection of individuals at high risk of type 2 diabetes, and thereby possibly novel mechanisms for its prevention and early treatment, a few studies have addressed the issue of whether low serum 25(OH)D concentrations can predict the development of type 2 diabetes [7–12]. However, some of these reports are limited by, for example, the use of indirect measures or surrogate markers of vitamin D status [11] and type 2 diabetes [9], by a rather small study sample [9] or by an incomplete identification of incident type 2 diabetes [7, 8, 10]. Thus, further long-term prospective studies involving well-characterised cohorts seem warranted to investigate the predictive value of serum 25(OH)D for type 2 diabetes.

Assuming that an inverse association exists between 25(OH)D and risk of type 2 diabetes, the putative mechanism(s) behind such an association need to be elucidated. Studies in both animals and humans have suggested that vitamin D may exert positive effects on insulin secretion and insulin sensitivity, either directly, or indirectly through a modulation of calcium metabolism [13–16]. An alternative link between vitamin D and glucose homeostasis might be the IGF system [9]. We have shown that low serum levels of IGF-binding protein-1(IGFBP-1) in men and women with normal glucose tolerance (NGT) strongly predict risk of type 2 diabetes and prediabetes (impaired fasting glucose [IFG], impaired glucose tolerance [IGT] or the two combined) at a follow-up 8–10 years later [17, 18]. In addition, a combination of high concentrations of 25(OH)D and IGF-1 has been associated with a low risk of the metabolic syndrome [19].

The primary aim of this study was to investigate whether the development of prediabetes (IFG or IGT) and type 2 diabetes can be predicted by serum 25(OH)D concentration. For this purpose, we analysed 25(OH)D concentrations in a well-characterised cohort of 2,378 middle-aged women and men with either NGT or prediabetes at a baseline investigation in relation to their glucometabolic status (NGT, prediabetes or type 2 diabetes) at a follow-up investigation 8–10 years later. Moreover, associations between serum 25(OH)D concentrations and prediabetes were evaluated using data from the baseline investigation. Combinations of 25(OH)D and IGF-1 or IGFBP-1, respectively, in association with prediabetes and type 2 diabetes were studied both at baseline and in the prospective investigation.

Methods

Study population

The study population consisted of participants in an epidemiological survey, the Stockholm Diabetes Prevention Program, described elsewhere [20]. In brief, women and men aged 35–56 years without known diabetes participated in the baseline study over the period 1992–1998. The follow-up examination, 8–10 years later, included individuals who at baseline had either NGT or prediabetes (including IFG and IGT), but not those who had newly diagnosed type 2 diabetes at the baseline examination. A health examination, including measurements of weight, height and BP, and an OGTT were performed, and questionnaires covering lifestyle factors were answered at baseline and at follow-up.

Associations between 25(OH)D concentration and abnormal glucose regulation were evaluated in a nested case–control setting. Thus, serum concentrations of 25(OH)D were measured at baseline in all participants with abnormal glucose regulation, i.e. either prediabetes or type 2 diabetes at follow-up (490 women and 699 men), and in an equal number of participants, randomly selected within sex but pair-matched by exact age, from the group who had NGT at both baseline and follow-up.

Analyses were performed in three ways, two prospective and one cross-sectional. In prospective study 1, cases were individuals who had progressed from NGT at baseline to either prediabetes (304 women, 428 men) or type 2 diabetes (47 women, 87 men) at follow-up, or from prediabetes at baseline to type 2 diabetes (53 women, 92 men) at follow-up. The controls were 404 women and 607 men with NGT at both baseline and follow-up. In prospective study 2, which was a subset of prospective study 1, all cases had NGT at baseline and progressed to either prediabetes (304 women, 428 men) or type 2 diabetes (47 women, 87 men) at follow-up; the controls were 351 women and 515 men with NGT at both baseline and follow-up.

A cross-sectional study was also performed in which the cases were all individuals who had prediabetes at baseline, including not only those who progressed to type 2 diabetes, but also those who also had prediabetes at follow-up, a total of 139 women and 184 men. In this analysis, the controls were all participants with NGT at baseline, irrespective of glucose tolerance at follow-up, giving a total of 841 women and 1,214 men.

The investigation was approved by the ethics committee of Karolinska Institutet, and was carried out in accordance with the principles of the Declaration of Helsinki. All participants gave their informed consent.

Seasonal variations in serum 25(OH)D concentration

There was a seasonal variation in concentrations of 25(OH)D. To adjust for this, the months were categorised into four quarters: (1) November–January, (2) February–April, (3) May–July, and (4) August–October. According to sunlight exposure in Sweden, quarter 1 has the fewest sun hours. Linear regression analyses were performed with 25(OH)D as the dependent variable and the four quarters, coded as dummy variables using quarter 1 as the reference, as independent variables. Estimated β-coefficients were then used to adjust all individual 25(OH)D values for seasonal variation.

Classification of glucose tolerance

The 75 g OGTT was performed at baseline and at follow-up examination, and glucose regulation was defined according to American Diabetes Association guidelines [21], using a fasting serum glucose value of ≥5.6 mmol/l as the cut-off value for IFG.

Classification of established risk factors for diabetes and potential confounders

Physical activity during leisure hours was based on four response alternatives and categorised into three groups: low (sedentary), middle (moderate activity) and high (regular exercise and training). BMI was also divided into three groups (<25.0, 25.0–29.9 and ≥30.0 kg/m2). Waist circumference was categorised into three (<80, 80–87 >87 cm for women, and <94, 94–101 and >101 cm for men). Tobacco use (including smoking and snuff) was described as either never/former use or current tobacco use. Family history of diabetes (FHD) was designated positive when at least one first-degree (parents or siblings) or two second-degree relatives (grandparents, uncles or aunts) had been diagnosed with diabetes, or negative when no close relative had diabetes.

Socioeconomic position, based on self-reported occupation and classified according to the standard system from Statistics Sweden, was categorised into four groups (high, middle, low and self-employed). Education comprised three categories: low (elementary school), middle (senior high school, or technical and vocational school) and high (college or university). High BP was defined as a systolic BP ≥140 mmHg and/or a diastolic BP ≥90 mmHg and/or the use of antihypertensive treatment.

Assays Serum concentrations of 25(OH)D were determined by a chemiluminescence immunoassay (DiaSorin Inc., Stillwater, MN, USA) measuring both 25(OH)D3 and 25(OH)D2 with intra- and interassay CVs of 5% and 8%, respectively. Serum IGFBP-1 was measured using an in-house RIA with intra- and inter-assay (CV) values of 3% and 10%, respectively. IGF-1 was measured in serum by RIA after acid-ethanol extraction and cryoprecipitation, using des(1–3)IGF-1 as a tracer to minimise interference from IGFBPs. Intra- and interassay CVs were 4% and 11%, respectively. Venous serum glucose was assayed using the glucose oxidase method (Yellow Springs Glucose Analyzer, Yellow Springs, OH, USA).

Statistical analysis

Seasonally adjusted 25(OH)D values were used in all analyses, either categorised into quartiles or used as a continuous variable and reported per increment of 10 nmol/l. Categorisation into quartiles was performed according to the distribution between all participants, women and men combined (n = 2,378). Variables not normally distributed (IGFBP-1 and IGF-1) were log-transformed prior to analysis. Differences between groups were analysed by ANOVA for trend, and unpaired t test or χ2-square test for continuous and categorical data, respectively. ORs together with 95% CIs were calculated in multiple logistic regression analysis to explore the association between serum 25(OH)D concentrations and abnormal glucose regulation.

Potential confounders (physical activity during leisure time, BMI, waist, tobacco use, FHD, socioeconomic position, education and BP) 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 to the age-adjusted crude estimate in any of the outcome measures were included in the final multiadjusted model. Waist and BMI had similar influence on the crude estimates, so BMI was chosen for the final model. Thus, two logistic regression models are given: model 1 adjusted for age (and sex when combining male and female participants), and model 2 adjusted for age, BMI, FHD, physical activity and BP (and sex if applicable).

When categories of 25(OH)D were used in the logistic regression analyses, tests of linear trend were conducted by assigning median values of 25(OH)D in quartiles as a continuous variable. Since no statistical interaction was found between sex and 25(OH)D for any of the outcome measures, analyses are reported not only separately, but also for women and men combined. Analyses of combinations of 25(OH)D and IGF-I or IGFBP-1 were performed with the variables dichotomised at their medians, and statistical interaction was tested by including a product term into the regression models. Only participants for whom there were data on all variables were included in the study.

The analyses were performed using SAS statistical package version 9.2 for Windows (SAS Institute, Cary, NC, USA).

Results

Baseline characteristics according to quartiles of 25(OH)D

Women and men in the highest 25(OH)D quartile(s) were more physically active, had lower 2 h glucose levels during the OGTT and a lower prevalence of prediabetes, were less obese, with a lower BMI and smaller waist circumference, and had higher serum IGFBP-1 and IGF-1 levels compared with participants in the lower quartiles of the seasonally adjusted 25(OH)D concentrations (see electronic supplementary material [ESM] Table 1). In addition, women and men in the lowest quartile used more tobacco. There was no association between FHD and 25(OH)D concentrations. Furthermore, seasonally adjusted 25(OH)D concentrations were lower in women than in men (mean 57.0 nmol/l, 95% CI 55.9, 58.1 vs 61.2 nmol/l, 95% CI 60.2, 62.2; p < 0.0001).

Baseline characteristics and glucose tolerance at follow-up

Among all cases who had NGT or prediabetes at baseline and progressed to either prediabetes or type 2 diabetes at follow-up, both women and men had a higher prevalence of FHD, higher frequencies of hypertension, higher BMI and waist circumference, more prevalent current tobacco use, and lower concentrations of adjusted 25(OH)D and lower IGFBP-1 levels compared with controls with NGT at both baseline and follow-up (Table 1). In addition, women with prediabetes or type 2 diabetes at follow-up were less physically active and had higher levels of IGF-1, whereas the corresponding men had lower levels of IGF-1 than NGT participants.

Table 1 Baseline characteristics for the prospective studies
Full size table

Similar differences compared with controls were observed when analysing only data from participants who had NGT at baseline, although in men the levels of 25(OHD) did not differ (Table 1).

Prospective studies of the association of 25(OH)D concentration with development of prediabetes and type 2 diabetes

In prospective study 1 (Table 2), participants who had NGT or prediabetes at baseline and were in the highest seasonally adjusted quartile of 25(OH)D concentration had a reduced risk (OR 0.64, CI 0.50, 0.82) of progressing to prediabetes or type 2 diabetes compared with those in the lowest quartile. Yet after full adjustment for confounders (age, BMI, BP, FHD and physical activity), the reduction in risk did not remain. When analysing progression only to type 2 diabetes from NGT or prediabetes, a reduced risk (OR 0.58, CI 0.37, 0.90) was found after full adjustment. Moreover, for progression from prediabetes to type 2 diabetes, there was an even greater reduction in risk (OR 0.38, CI 0.21, 0.71), the reduction in risk being 23% for each 10 nmol/l increase in 25(OH)D.

Table 2 Prospective study 1
Full size table

In analyses stratified for sex, women in the highest quartile of 25(OH)D concentration had a decreased age-adjusted OR for progress to either prediabetes or type 2 diabetes compared with controls (OR 0.56, CI 0.38, 0.84; Table 2), but this association did not persist after full adjustment. The results were similar when analysing women progressing from NGT or prediabetes to type 2 diabetes. However, in women who progressed from prediabetes to type 2 diabetes, the OR was 0.73 (CI 0.58, 0.86) for a 10 nmol/l increase in 25(OH)D concentrations after adjustment (Table 2). For men in the highest quartile of the 25(OH)D distribution, there was also an age-adjusted reduced OR (OR 0.68, CI 0.49, 0.95), but, as in women, this association did not persist after full adjustment. However, in men belonging to the highest serum 25(OH)D quartile, there was a decrease in OR (OR 0.40, CI 0.19, 0.85) for progression from prediabetes to type 2 diabetes. Similar to the effect in women, a 10 nmol/l increase in 25(OH)D concentration reduced the risk of diabetes in men with prediabetes by 21% (OR 0.79, CI 0.68, 0.93).

In prospective study 2 (ESM Table 2), analyses were performed on participants progressing from NGT to either prediabetes, type 2 diabetes or prediabetes and type 2 diabetes combined. However, after full adjustment, no associations were found with 25(OH)D for any of the outcomes, either for the two sexes separately or when they were combined.

The association between combinations of baseline serum concentrations of 25(OH)D and IGFBP-1 or IGF-1, respectively, and type 2 diabetes at follow-up was analysed in participants with NGT or prediabetes at baseline who developed type 2 diabetes at follow-up (Fig. 1). In both women and men, high levels of IGFBP-1 (above the median) significantly predicted a reduced risk of type 2 diabetes, while at high concentrations of 25(OH)D (above the median), there was only a tendency towards a risk reduction. In combinations, no significant interaction term between categories of 25(OH)D and IGFBP-1 was found either for women (p = 0.168) or for men (p = 0.364). Although high levels of IGFBP-1 further reduced the risk of developing type 2 diabetes, this reduction was similar regardless of 25(OH)D concentration. Moreover, only in participants with high IGFBP-1 levels was the risk of type 2 diabetes significantly reduced compared with those having low levels of both variables. In women, there was no difference in risk of type 2 diabetes according to IGF-1 levels (above or below the median value), and the risk was not further significantly influenced in those with high 25(OH)D concentrations (p = 0.743 for the interaction term; Fig. 1). In partial contrast, men with a high (above the median) IGF-1 level showed a reduced risk of type 2 diabetes irrespective of 25(OH)D concentrations, and no interaction between IGF-1 and 25(OH)D was observed (p = 0.250).

Fig. 1
figure 1

Baseline serum levels of 25(OH)D and baseline serum levels of IGFBP-1 (a) or IGF-I (b), separately and in combination, associated with the development of type 2 diabetes at follow-up in women (grey lines) and men (black lines). Cases are individuals who had either NGT or prediabetes at baseline and developed type 2 diabetes at follow-up, whereas controls had NGT at both baseline and follow-up (prospective study 1). Low, < median; high, ≥ median. The seasonally adjusted median value for 25(OH)D was 58.1 nmol/l for both women and men; the median value of IGFBP-1 for women was 43 μg/l and for men was 25 μg/l; and the median value of IGF-I was 192 μg/l for both women and men. Logistic regression was performed. The data are expressed as OR and 95% CI, adjusted for age, BMI, FHD, physical activity and BP measured at baseline

Full size image

Cross-sectional study of the association between 25(OH)D concentrations and prediabetes

At baseline, participants with prediabetes had a higher prevalence of positive FHD and of hypertension, a higher BMI and waist circumference, and lower serum levels of IGFBP-1 and seasonally adjusted 25(OH)D compared with controls with NGT (ESM Table 3). Men but not women with prediabetes had lower serum IGF-1 levels. In addition, prediabetic women but not men were less physically active. In the logistic regression analyses of all participants, those in the two highest quartiles of 25(OH)D showed a decreased association of 25(OH)D with prediabetes after full adjustment (OR 0.49, CI 0.34, 0.72, and OR 0.58, CI 0.41, 0.83, respectively; Table 3). This was accounted for by significant associations in men (OR 0.41, CI 0.25, 0.68, and OR 0.57, CI 0.36, 0.91, respectively) but not in women. For each 10 nmol/l increase in serum 25(OH)D, the association with prediabetes was decreased by 18% in men (p < 0.001) and 11% in women (0.051).

Table 3 Cross-sectional baseline study
Full size table

In women, there were no significant associations between dichotomised serum concentrations of 25(OH)D, IGFBP-1 or IGF-1 and prediabetes at baseline (Fig. 2). However, in men with high levels of IGFBP-1 (above the median), the ORs associated with prediabetes were further reduced, but to a similar extent in participants with low and with high concentrations of 25(OH)D (p = 0.745 for the interaction). Combining categories of 25(OH)D and IGF-I in men revealed an interaction (p = 0.038), indicating that high levels of IGF-I decreased the OR associated with prediabetes in men with low levels of 25(OH)D, whereas the reduced OR in men with high levels of 25(OH)D was observed irrespective of IGF-I levels.

Fig. 2
figure 2

Baseline serum levels of 25(OH)D and of IGFBP-1 (a) or IGF-I (b), separately and in combination, associated with prediabetes at baseline in women (grey lines) and men (black lines). Cases are participants with prediabetes at baseline, and controls are participants with NGT at baseline (cross-sectional study). Low, < median; high, ≥ median. The seasonally adjusted median value for 25(OH)D was 58.1 nmol/l for both women and men; the median value of IGFBP-1 for women was 43 μg/l and for men was 25 μg/l; and the median value of IGF-I was 192 μg/l for both women and men. Logistic regression was performed. The data are expressed as OR and 95% CI, adjusted for age, BMI, FHD, physical activity and BP measured at baseline

Full size image

Discussion

We found that participants with high concentrations of serum 25(OH)D showed a reduced prevalence of prediabetes in the cross-sectional study, and that high concentrations of serum 25(OH)D in participants with prediabetes at baseline predicted a decreased risk of progressing to type 2 diabetes 8–10 years later. Conversely, low serum 25(OH)D was associated with increased risk of type 2 diabetes in prediabetic individuals. Notably, in persons with NGT at baseline, high 25(OH)D concentrations did not predict a significantly reduced risk of progressing to either prediabetes or type 2 diabetes. When analysing whether the associations with 25(OH)D were related to sex, it appeared in the cross-sectional study that they were significant only in men after controlling for confounders. It is noteworthy, however, that in both men and women with prediabetes at baseline, the risk of diabetes at follow-up was significantly decreased, by 21% and 27%, respectively, for every 10 nmol/l increase in 25(OH)D.

Our study has several strengths. First, in addition to the baseline investigation, the study has a prospective design in which cases with prediabetes and type 2 diabetes at follow-up were matched by age and sex to individuals who had NGT at both baseline and follow-up. Second, the participants were not aware of having diabetes or prediabetes when the exposure data were recorded, and thus the disease status was not likely to affect their behaviour or lifestyle. Third, the study sample was well-characterised, and adjustment for potential confounders was possible.

The positive association in the cross-sectional study between prediabetes and low serum 25(OH)D concentrations is in agreement with data from a recent US investigation [22]. Furthermore, similar to our study, the association was independent of confounders such as BMI and BP. In more general terms, a recent review of the role of vitamin D in prediabetes found that most cross-sectional and prospective studies in various populations demonstrate an inverse association between serum 25(OH)D concentrations and IGT, fasting plasma glucose, the metabolic syndrome and the incidence of prediabetes [23]. In addition, participants in the highest quartile of 25(OH)D were more physically active, had lower 2 h glucose levels, were less obese and had a smaller waist circumference compared with participants in the lower quartiles. These findings are consistent with previous cross-sectional studies in which vitamin D deficiency has been associated with other important risk factors for diabetes [24].

The effect of 25(OH)D concentrations on the risk of developing type 2 diabetes was detected in individuals with prediabetes and not with NGT at baseline, suggesting that this effect is not strong enough to affect the development of IGT or IFG. Instead, established risk factors such as overweight and physical inactivity may be critical in transforming the glucose tolerance status from normal to impaired. When prediabetes is already present, low serum 25(OH)D concentrations can promote a further deterioration in glucose tolerance.

There are several suggested mechanisms for how vitamin D might influence the progression from prediabetes to type 2 diabetes. Vitamin D receptors (VDRs) are expressed in more than 30 different tissues, including pancreatic islet and skeletal muscle cells [25], and serum vitamin D concentrations may exert modulating effects on both insulin secretion and insulin action [13–16]. The direct effect of vitamin D might be mediated by binding of the circulating active form, 1.25(OH)D3, to the pancreatic beta cell VDR, and activation of vitamin D within the beta cell by the 1-α-hydroxylase enzyme; alternatively, it may be brought about by an effect on insulin action through stimulating the expression of insulin receptors and enhancing insulin responsiveness for glucose transport [2]. The indirect effects of vitamin D can be mediated via regulating extracellular calcium and insulin secretion through calcium fluxes in the beta cells [26]. Vitamin D may also be involved in systemic inflammation processes, and improve insulin sensitivity and promote beta cell survival by directly modulating the effect of cytokines [2]. Furthermore, polymorphisms in the VDR gene may also be linked to insulin resistance, insulin secretion and fasting glucose levels. The latter seems to be a less likely explanation for our findings, however, since there was no association between FHD and 25(OH)D concentration.

Although we found that the predictive value of serum 25(OH)D concentration as a continuous trait was similar in women and men, the categorical data demonstrated a sex difference implying a more obvious role of vitamin D in men than in women. This difference was evident in relation to both the value of serum 25(OH)D as a predictor of the risk of developing type 2 diabetes from prediabetes, and the association between 25(OH)D and prediabetes in the cross-sectional study. Notably, these associations were also significant in women when adjusting only for age, but disappeared after full adjustment for possible confounders.

The major factor behind this adjustment effect appeared to be BMI, which seems to be a strong confounder in this context. Thus, serum 25(OH)D concentrations have been shown to be inversely correlated with various measures of obesity, including weight, BMI and waist circumference [27–29]. In a prospective Norwegian investigation, baseline serum 25(OH)D levels were inversely associated with type 2 diabetes after an 11-year follow-up, but this association disappeared in both sexes after adjustment for BMI [30]. Another factor that may influence the predictive value of vitamin D status in women is that there were rather few women with serum 25(OH)D concentrations in the higher quartiles and prediabetes who progressed to type 2 diabetes. Thus, it remains unclear whether our loss of significance in women after adjustment for BMI and other possible confounders is accounted for by a lack of power or any other factor.

In line with our previous reports, high IGFBP-1 concentrations predicted a reduced risk of diabetes in both men and women [17, 18], and these effects were evident regardless of 25(OH)D serum concentrations. Interestingly, in the presence of low serum IGFBP-1 concentrations, high 25(OH)D concentrations did not predict a reduced risk of progression to type 2 diabetes. Thus, these data do not support the findings of an earlier study by Forouhi et al. [9] demonstrating that high serum 25(OH)D concentrations were associated with improved glucose tolerance in participants with low, but not high, serum IGFBP-1 concentrations. High serum IGF-1 levels predicted a reduced risk of diabetes only in men, and there were no further significant effects due to 25(OH)D concentration. Thus, in agreement with Forouhi et al. [9], we were not able to present supporting evidence for a previous finding by Hypponen et al, in a British cohort [19], that high but not low serum concentrations of 25(OH)D combined with high serum IGF-1 levels are associated with a decreased risk of developing components of the metabolic syndrome. Interactions between the IGF system and vitamin D have been suggested [6, 31, 32]. Thus, serum 25(OH)D may regulate IGF-1 activity and, conversely, IGF-1 may increase vitamin D concentrations. Furthermore, vitamin D has been shown to stimulate IGFBP-1 production, probably via binding to VDRs [33].

In summary, high serum concentrations of 25(OH)D predict a reduced risk of progressing from prediabetes to type 2 diabetes. Furthermore, there was a cross-sectional association between prediabetes and serum 25(OH)D concentration, mainly in men. There were no significant interactions concerning diabetes risk between IGF-1 and IGFBP-1 on the one hand and 25(OH)D on the other. Our data suggest that vitamin D supplementation should be evaluated as for the prevention of type 2 diabetes in individuals with prediabetes.