- Top of page
- Subjects and methods
- Competing interests
The prevalence of Type 2 diabetes mellitus (DM) is increasing worldwide, implying huge challenges for the health systems in the future [1,2]. Effective means for prevention and treatment of this condition are therefore needed. The basis of the development of the disease is insulin resistance combined with a relative deficit of insulin secretion from pancreatic B cells [1,3].
Serum 25-hydroxyvitamin D (25(OH)D) is the commonly used biomarker of a person's vitamin D status, and exists in two forms; 25-hydroxyvitamin D3 (25(OH)D3 and 25-hydroxyvitamin D2 (25(OH)D2). The presence of vitamin D receptors (VDRs) and the enzyme 25-hydroxyvitamin D-1α-hydroxylase, which converts 25(OH)D into its active form, 1,25-dihydroxyvitamin D (1,25(OH)2D), in B cells have been well described [4,5]. Studies suggest that high concentrations of 1,25(OH)2D lead to an increase in insulin secretion [6,7], possibly mediated through increased calcium influx into the B cells . Systemic inflammation contributes to the development of insulin resistance and eventually Type 2 DM , and an anti-inflammatory effect of vitamin D may also possibly affect the development of Type 2 DM, although this has been little studied . In line with this, cross-sectional data from a number of epidemiological studies show an inverse association between serum 25(OH)D concentrations and glucose concentrations [9–12], insulin resistance [10,12–16] and prevalence of Type 2 DM . There are, however, few prospective studies evaluating this relationship [10,17,18], and a recent review pointed out the need for more prospective studies to clarify and quantify the association between serum 25(OH)D concentrations and the risk of Type 2 DM . We wanted to test the hypothesis that low serum 25(OH)D concentrations are associated with increased risk of subsequent Type 2 DM in a population-based cohort with 11 years of follow-up.
- Top of page
- Subjects and methods
- Competing interests
There were 4829 non-smoking and 2336 smoking participants with valid serum 25(OH)D measurements in the baseline examination of the Tromsø Study 1994–95. After exclusion of persons registered with Type 1 DM (n = 15), participants reporting diabetes at baseline and/or having journal-based previous diagnosis of Type 2 DM (n = 159), participants with baseline HbA1c > 6.5% (48 mmol/mol) (n = 42) or missing (n = 375) at baseline, participants registered as moved out of Tromsø before baseline (n = 14) or with missing information on any of the variables used in the models (n = 67), 4157 non-smoking participants were available for the final analyses. Among smokers, 1962 participants were included after exclusion of participants with Type 1 DM (n = 4), self-reported and/or registered Type 2 DM at baseline (n = 48), baseline HbA1c > 6.5% (48 mmol/mol) (n = 11) or missing (n = 239), participants registered as moved out of Tromsø before baseline (n = 2) or with missing information on any of the variables used in the models (n = 70). This group also included exclusively cigar (n = 15) and pipe smokers (n = 19).
Table 1 shows baseline characteristics for the participants according to serum 25(OH)D quartile in non-smokers and smokers. In non-smokers, age, BMI and serum PTH tended to decrease with increasing quartiles, while the proportion of former smokers and physical activity score increased slightly with increasing quartiles of serum 25(OH)D. In smokers, BMI and serum PTH decreased across serum 25(OH)D quartiles, while number of cigarettes increased.
Table 1. Baseline characteristics of participants with serum 25(OH)D measurements in the Tromsø Study 1994–95 followed for development of Type 2 diabetes mellitus in a period of 11 years after baseline
| ||All||Serum 25(OH)D|
|Quartile* 1||Quartile 2||Quartile 3||Quartile 4||Test for trend (P-value)|
| n||4157||1030||1042||1040||1045|| —|
| Serum 25(OH)D (nmol/l, range)||5.0–192.2||5.0–53.1||34.8–62.3||43.4–73.5||52.3–192.2|| —|
| Serum 25(OH)D (nmol/l)||52.8 (16.8)||34.5 (8.0)||47.2 (6.4)||56.6 (6.8)||72.8 (13.9)|| NA|
| Sex (% females/males)||62/38||62/38||62/38||62/38||62/38||1.00|
| Age (years)||59.6 (10.1)||60.8 (10.6)||60.0 (9.7)||59.1 (10.6)||58.6 (9.2)||< 0.01|
| BMI (kg/m2)||26.3 (3.8)||27.0 (4.4)||26.5 (3.9)||26.0 (3.6)||25.7 (3.3)||< 0.01|
| Former smokers (%)||50||46||49||49||55||< 0.01|
| Physical activity score ||3.4 (2.4)||3.2 (2.4)||3.3 (2.3)||3.5 (2.3)||3.7 (2.4)||< 0.01|
| Serum PTH measured (n)||1938||563||518||446||411|| NA|
| Serum PTH [pmol/l, median (interquartile range)]||2.7 (1.5)||3.0 (1.9)||2.7 (1.4)||2.5 (1.3)||2.3 (1.4)||< 0.01|
| n||1962||473||488||507||494|| —|
| Serum 25(OH)D (nmol/l, range)||5.0–179.5||5.0–67.2||52.9–79.3||64.4–91.3||76.0–179.5|| —|
| Serum 25(OH)D (nmol/l)||73.0 (20.3)||49.9 (9.3)||65.4 (5.9)||77.1 (6.5)||98.6 (15.2)|| NA|
| Sex (% females/males)||60/40||60/40||55/45||59/41||64/36||0.13|
| Age (years)||57.3 (10.3)||57.7 (11.6)||57.2 (10.6)||57.3 (9.6)||56.8 (9.4)||0.22|
| BMI (kg/m2)||24.7 (3.7)||25.1 (4.3)||24.9 (3.7)||24.7 (3.4)||24.1 (3.2)||< 0.01|
| Physical activity score ||3.1 (2.4)||3.0 (2.4)||3.0 (2.3)||3.2 (2.4)||3.2 (2.4)||0.12|
| Number of years smoked||34.2 (11.5)||33.3 (12.6)||34.1 (11.7)||34.8 (11.1)||34.4 (10.8)||0.08|
| Number of cigarettes/day||11.3 (6.1)||10.3 (5.6)||11.4 (6.5)||11.8 (6.0)||11.9 (6.2)||< 0.01|
| Serum PTH measured (n)||890||252||212||225||201|| NA|
| Serum PTH [pmol/l, median (interquartile range)]||2.3 (1.4)||2.6 (1.9)||2.3 (1.6)||2.1 (1.2)||2.0 (1.3)||< 0.01|
Type 2 DM was registered in 183 participants (4.4%) of the non-smokers and in 64 (3.3%) of the smokers during the follow-up period. Year of diagnosis was known for 144 of the non-smokers and 47 of the smokers. Age- and sex-adjusted survival analyses revealed an increased hazard ratio for being diagnosed with Type 2 DM throughout the follow-up period for both non-smokers and smokers in the first quartile compared with the fourth, and for smokers also in the second quartile (Tables 2 and 3). However, after additional adjustment for BMI, the hazard ratios were attenuated and no longer significant. Further adjustment for physical activity score and, for non-smokers, former smoking status did not change the results, nor did adjustment for serum PTH in the subgroup where this was measured (data not shown). There was no statistical interaction between sex and quartile of serum 25(OH)D in relation to Type 2 DM (P = 0.65), and the results were, as expected, similar when the analyses were performed stratified by sex (data not shown).
Table 2. The hazard ratios for developing Type 2 diabetes mellitus during 11 years of follow-up in relation to serum 25(OH)D quartile at baseline in the Tromsø Study 1994–95 in 4157 non-smokers
| ||Cox regression, model 1||Cox regression, model 2||Cox regression, model 3|
|HR||95% CI||P-value||HR||95% CI||P-value||HR||95% CI||P-value|
|Quartiles* of serum 25(OH)D (events/person years at risk)|
| Quartile 4 (ref) (34/10 201)||1||—||—||1||—||—||1||—||—|
| Quartile 3 (34/10 160) ||1.00||0.62–1.61||1.00||0.95||0.59–1.53||0.85||0.94||0.59–1.51||0.81|
| Quartile 2 (51/10 035)||1.50||0.97–2.31||0.07||1.29||0.83–1.99||0.26||1.27||0.82–1.97||0.28|
| Quartile 1 (64/9827)||1.89||1.25–2.88||< 0.01||1.40||0.91–2.14||0.12||1.37||0.89–2.10||0.15|
| Sex (female = 0, male = 1)||1.31||0.97–1.75||0.08||1.53||1.14–2.07||< 0.01||1.61||1.16–2.24||< 0.01|
| BMI (kg/m2)||—||—||—||1.17||1.14–1.21||< 0.01||1.17||1.13–1.20||< 0.01|
| Physical activity score ||—||—||—||—||—||—||0.91||0.84–0.97||< 0.01|
|Former smoking (0 = no, 1 = yes)||—||—||—||—||—||—||1.12||0.82–1.54||0.48|
Table 3. The hazard ratios for developing Type 2 diabetes mellitus during 11 years of follow-up in relation to serum 25(OH)D quartile at baseline in the Tromsø Study 1994–95 in 1962 smokers
| ||Cox regression, model 1*||Cox regression, model 2*||Cox regression, model 3*|
|HR||95% CI||P-value||HR||95% CI||P-value||HR||95% CI||P-value|
|Quartiles† of serum 25(OH)D (events/person years at risk)|
| Quartile 4 (ref) (8/4705)||1||—||—||1||—||—||1||—||—|
| Quartile 3 (16/4859)||1.79||0.77–4.19||0.18||1.50||0.66–3.61||0.32||1.55||0.66–3.64||0.31|
| Quartile 2 (19/4441)||2.33||1.02–5.35||0.05||1.84||0.80–4.23||0.15||1.76||0.76–4.05||0.19|
| Quartile 1 (21/4395)||2.68||1.18–6.08||0.02||1.50||0.64–3.55||0.36||1.47||0.62–3.48||0.38|
| Sex (female = 0, male = 1)||2.68||1.55–4.62||< 0.01||2.73||1.57–4.73||< 0.01||2.97||1.71–5.20||< 0.01|
| BMI (kg/m2)||—||—||—||1.25||1.19–1.32||< 0.01||1.24||1.18–1.31||< 0.01|
| Physical activity score ||—||—||—||—||—||—||0.89||0.78–1.01||0.06|
Exclusion of the participants with diabetes having unknown year of diagnosis did not change the results substantially (data not shown). When performing logistic regression, the risk estimates (odds ratios) and significance levels were closely similar to what we found in the Cox regression model presented in the tables (data not shown).
Using serum 25(OH)D as a continuous variable with adjustment for month of blood sampling, age- and sex-adjusted analyses again revealed a 14 and 15% reduction in incident Type 2 DM per 10 nmol/l increase in serum 25(OH)D in non-smokers and smokers, respectively (Table 4). However, after additional adjustment for BMI, the estimates were again reduced and no longer significant (Table 4).
Table 4. The hazard ratios for developing Type 2 diabetes mellitus during 11 years of follow-up in relation to serum 25(OH)D at baseline in the Tromsø Study 1994–95 in 4157 non-smokers and 1962 smokers; Cox regression model with serum 25(OH)D as continuous variable
| ||HR (95% CI) for developing Type 2 DM, non-smokers||HR (95% CI) for developing Type 2 DM, smokers†|
|Serum 25(OH)D (per 10 nmol/l)||0.89 (0.81–0.97)*||0.86 (0.75–0.98)*|
|+ Month of blood sampling||0.86 (0.78–0.94)**||0.83 (0.73–0.96)*|
|+ Age (years)||0.87 (0.78–0.96)**||0.84 (0.73–0.96)*|
|+ Sex (female = 0, male = 1)||0.86 (0.78–0.95)**||0.85 (0.74–0.98)*|
|+ Physical activity score||0.87 (0.78–0.96)**||0.86 (0.75–0.98)*|
|+ BMI (kg/m2)||0.95 (0.86–1.05)||0.96 (0.83–1.12)|
When we stratified the participants into quartiles of BMI, we found that for both smokers and non-smokers, the hazard ratio for developing Type 2 DM decreased significantly with increasing serum 25(OH)D in the lowest (leanest) BMI quartile, while there was no effect in the other BMI quartiles (Table 5). Additional adjustment for BMI did not change the results (data not shown).
Table 5. The hazard ratios for developing Type 2 diabetes mellitus during 11 years of follow-up in relation to serum 25(OH)D at baseline in the Tromsø Study 1994–95 in 4157 non-smokers and 1962 smokers, stratified by BMI quartiles
| ||BMI quartile|
|BMI range (kg/m2) ||11.9–23.1 ||23.2–25.4||25.5–28.2||28.3–54.7|
| Serum 25(OH)D||0.67 (0.42–1.07)||1.06 (0.83–1.36)||1.01 (0.83–1.36)||0.94 (0.83–1.06)|
| + Month of blood sampling||0.62 (0.40–0.97)*||1.05 (0.82–1.36)||0.95 (0.78–1.17)||0.92 (0.81–1.05)|
| + Age (years)||0.65 (0.42–1.02)||1.07 (0.83–1.36)||0.96 (0.78–1.18)||0.92 (0.82–1.05)|
| + Sex (female = 0, male = 1)||0.54 (0.33–0.90)*||1.05 (0.81–1.34)||0.95 (0.78–1.17)||0.91 (0.80–1.04)|
| + Physical activity score||0.56 (0.34–0.92)*||1.06 (0.83–1.36)||0.96 (0.78–1.18)||0.91 (0.80–1.04)|
| Events/subjects||6/705||5/524 ||21/405||32/328|
| Serum 25(OH)D||0.64 (0.42–0.98)*||0.89 (0.56–1.42)||1.07 (0.84–1.34)||0.87 (0.71–1.07)|
| + Month of blood sampling||0.61 (0.39–0.96)*||0.94 (0.59–1.49)||1.09 (0.84–1.41)||0.83 (0.67–1.02)|
| + Age (years)||0.60 (0.38–0.94)*||0.96 (0.60–1.52)||1.08 (0.84–1.41)||0.83 (0.67–1.03)|
| + Sex (female = 0, male = 1)||0.62 (0.39–0.98)*||0.96 (0.61–1.52)||1.08 (0.84–1.45)||0.81 (0.64–1.02)|
| + Physical activity score||0.62 (0.39–0.98)*||0.99 (0.61–1.58)||1.09 (0.84–1.42)||0.83 (0.65–1.04)|
- Top of page
- Subjects and methods
- Competing interests
In this prospective observational study, we found an increased 11 year risk of Type 2 DM for both smoking and non-smoking participants with serum 25(OH)D concentrations in the lower quartiles and, conversely, the risk of incident Type 2 DM during follow-up decreased with increasing serum 25(OH)D concentrations. These associations were attenuated, and no longer significant, after adjustment for BMI. However, stratified analyses within BMI quartiles still showed a significant decrease in risk of developing Type 2 DM with increasing serum 25(OH)D concentration in the lowest BMI quartile in both smokers and non-smokers.
An advantage of our study is the inclusion of a large number of both men and women in a wide age range. A high attendance rate of nearly 80% at both visits of the Tromsø Study 1994–95 minimized the risk of selection bias and thereby increased the external validity, at least among Caucasians. Attendance rate was lower in the youngest and oldest age groups, as only 52 and 54% of the invited participants in the age ranges 25–34 and 75–84 years, respectively, attended the second visit of the study. However, previous reports from the Tromsø Study have not found any significant differences regarding self-reported health between participants who only attended the first visit and participants attending both the first and second visit .
The diabetes registration was thorough, and included also non-pharmacologically treated participants with Type 2 DM. However, we must expect some misclassification of diabetes because of the limitation of the registry for validation through hospital medical records. Since the local hospital provides the only laboratory in the municipality, the diabetes registration committee also had HbA1c measurements taken by the family doctors available when validating the possible cases, thus increasing the quality of the registry.
The use of serum 25(OH)D as a predictor reflects both nutritional and cutaneous sources of vitamin D, and is therefore a better marker of available vitamin D in the body than self-reported intake of vitamin D . The immunometric method used for analyses of vitamin D does not recognize ergocalciferol (25(OH)D2) . This is not a concern in this study, as food fortification and over-the-counter vitamin D supplements for adults in Norway consist of cholecalciferol (vitamin D3) exclusively. Of greater concern is that compounds from cigarette smoking seem to interfere with the assay in a dose-dependent way , which makes the serum 25(OH)D measurements in smokers uncertain. Although attempts were made to adjust for this bias, the results in smokers must be interpreted with caution. Another limitation of the study was the relatively high serum 25(OH)D concentration in the population, which might have hidden a possible effect of low serum 25(OH)D concentrations on risk of diabetes. We also had only one serum 25(OH)D measurement available, and although the tracking of serum 25(OH)D has been shown to be of the same magnitude as for other risk factors, such as serum lipids and blood pressure , repeated serum 25(OH)D measurements would have strengthened the study.
This study does not confirm the findings from a Finnish nested case–control study with 17 years of follow-up, where the authors reported an adjusted odds ratio for developing Type 2 DM of 0.28 [95% confidence interval (CI) 0.10–0.81] in men when comparing the highest vs. the lowest quartile of baseline serum 25(OH)D . In women, there was, however, no association (odds ratio 1.14; 95% CI 0.60–2.17) . Their diabetes registration was based on prescription registers only, and thus did not include diet-controlled diabetes. Compared with the Finnish study, our cohort was older and had higher serum 25(OH)D concentrations (a mean serum 25(OH)D concentration of 34.5 nmol/l in the first quartile compared with 22.3 nmol/l in the Finnish study). Their follow-up time was also longer (17 vs. 11 years), and the number of cases higher (412 vs. 183). However, in a previous work that included only one of the two cohorts used in the nested case–control study, the results were similar to ours, as the lower risk of Type 2 DM in the highest serum 25(OH)D quartile was attenuated and no longer significant after adjustment for BMI . Our loss of significance after adjustment for BMI might therefore be due to lack of power.
The inverse relationship between serum 25(OH)D and BMI is well known and consistently described, and it is generally believed to be due to sequestration and/or storage of vitamin D and its metabolites in fat tissue . It has been argued that vitamin D deficiency may be the cause and not the result of overweight  and, in that case, it might be reasonable not to adjust for BMI, as the BMI could be an intermediate step between vitamin D and diabetes and thus, not a confounder. However, two recently published randomized controlled trials with vitamin D supplementation in overweight patients do not support a causal effect of vitamin D on weight [28,29]. Our results showing that serum 25(OH)D was associated with decreased risk of Type 2 DM in the leanest quartile suggest that although much of the association between serum 25OH)D and subsequent Type 2 DM is mediated through BMI, there might still be an impact of 25(OH)D independent of BMI. This is consistent with the results of two minor studies where insulin sensitivity improved after vitamin D supplementation without any concomitant change in BMI [30,31]. One could also speculate that the well-known association between increased BMI and elevated risk of Type 2 DM is so strong that it excludes the possibility of finding an effect of serum 25(OH)D in the higher BMI quartiles. However, as the number of events was low in the leanest BMI quartiles, these results must be interpreted with caution.
Several epidemiological studies have reported an inverse cross-sectional relationship persisting after adjustment for BMI between serum 25(OH)D concentrations and prevalent diabetes , metabolic syndrome [9,10,32–34], fasting glucose [9–12] and insulin resistance [10,13–16]. Also, in the Ely Study, including 524 British non-diabetic men and women, baseline serum 25(OH)D was inversely associated with 10 year risk of hyperglycaemia and insulin resistance independent of baseline BMI . However, stratified analyses by high/low insulin-like growth factor binding protein 1 showed that the association between baseline serum 25(OH)D and hyperglycaemia was only significant in participants with lower insulin-like growth factor binding protein 1 . No association between serum 25(OH)D and metabolic syndrome was found in the Rancho Bernardo Study, where an inverse association between serum 25(OH)D and glucose was described in men (n = 410) only, and not in women (n = 660) .
In the few intervention studies performed, which are mostly small, short-term and, in most cases, without control groups, the results are inconsistent regarding insulin sensitivity and secretion after treatment with various forms of vitamin D with or without calcium [7,36–40]. A larger study, including 314 participants, reported improvement in fasting glucose concentrations and homeostasis model assessment of insulin resistance scores in participants randomized to 500 mg calcium and 700 IU cholecalciferol vs. placebo, but only in women with baseline impaired fasting glucose . In the Women’s Health Initiative trial, there was no effect of a daily dose of 1000 mg calcium and 400 IU vitamin D vs. placebo regarding 7 years risk of incident diabetes in a total of 33 951 women , nor did the RECORD study show any beneficial effect of intake of 800 IU vitamin D for 24–62 months compared with placebo on incident Type 2 DM in 5292 elderly participants . Those two studies were originally designed for skeletal outcomes, and the dosages used might be too low to reach a substantial increase in serum 25(OH)D concentration. Smaller controlled trials, using high-dosage vitamin D supplementation among diabetic patients, have so far shown no effect on glycaemic control or insulin sensitivity ; however, in high-risk patients, a significant decrease in insulin resistance has been reported [30,31].
How can these inconsistent results be explained? Epidemiological studies on vitamin D are challenging. One well-known problem is the use of different laboratory methods for serum 25(OH)D measurements, with no universal gold standard. Another problem is how to handle the seasonality of serum 25(O)D, as it is strongly influenced by UVB radiation from sunlight. Most studies, like ours, have only one serum 25(OH)D measurement available per participant. In many studies, the whole cohort is stratified into quartiles of vitamin D, and the authors thereafter try to account for seasonal variation by adjusting for season or month of blood sampling. However, a recently published study using simulated models reported that this method might result in a bias away from null . This could be avoided by using season- or month-specific quartiles, thereby putting more weight on the relative serum 25(OH)D concentrations instead of the absolute concentrations . This is debatable, as we cannot assume that the seasonal variation will be the same for all participants. Also, we do not know what is the most important for different health outcomes, the mean serum 25(OH)D concentration over time or the length of time with serum 25(OH)D concentrations below a critical threshold. The results of the studies may also differ due to different populations included, with different serum 25(OH)D concentrations and diabetes risk profiles. Finally, residual confounding might explain the positive results in a number of observational studies, as vitamin D status or supplementation might be considered as a marker of a health-conscious lifestyle, so that other, not measured, factors are the real protective ones.
In conclusion, we have demonstrated that lower baseline serum 25(OH)D concentrations are associated with a higher 11 year risk of Type 2 DM in a prospective population-based study; however, this finding was no longer significant after adjustment for BMI. Epidemiological studies in this particular field are challenging for a number of reasons, and long-term randomized controlled trials, as well as prospective follow-up studies from a young age, using repeated serum 25(OH)D measurements, are needed to assess the possible role of vitamin D status over time and of long-term vitamin D supplementation in the prevention of Type 2 DM.