Diabet. Med. 29, e142–e150 (2012)
Aims To systematically review the evidence for the effect of vitamin D supplementation on glycaemia, insulin resistance, progression to diabetes and complications of diabetes.
Methods Systematic review and meta-analysis. We searched databases including MEDLINE, EMBASE and the Cochrane Library for randomized controlled trials comparing vitamin D or analogues with placebo. We extracted data on fasting glucose, glycaemic control, insulin resistance, insulin/C-peptide levels, micro- and macrovascular outcomes and progression from non-diabetes to diabetes. Studies were assessed independently by two reviewers according to a pre-specified protocol.
Results Fifteen trials were included in the systematic review. Trial reporting was of moderate, variable quality. Combining all studies, no significant improvement was seen in fasting glucose, HbA1c or insulin resistance in those treated with vitamin D compared with placebo. For patients with diabetes or impaired glucose tolerance, meta-analysis showed a small effect on fasting glucose (−0.32 mmol/l, 95% CI −0.57 to −0.07) and a small improvement in insulin resistance (standard mean difference −0.25, 95% CI −0.48 to −0.03). No effect was seen on glycated haemoglobin in patients with diabetes and no differences were seen for any outcome in patients with normal fasting glucose. Insufficient data were available to draw conclusions regarding micro- or macrovascular events; two trials failed to show a reduction in new cases of diabetes in patients treated with vitamin D.
Conclusions There is currently insufficient evidence of beneficial effect to recommend vitamin D supplementation as a means of improving glycaemia or insulin resistance in patients with diabetes, normal fasting glucose or impaired glucose tolerance.
Low 25-hydroxyvitamin D (25OHD) levels are commonly found in patients with both Type 1 and Type 2 diabetes mellitus [1–4]. In patients with established diabetes mellitus and in the general population, low 25OHD levels are associated with higher fasting glucose and higher levels of glycated haemoglobin [5,6]. Such associations are not only found in cross-sectional studies; low 25OHD levels are associated with a higher probability of future diagnosis of diabetes mellitus or metabolic syndrome in prospective population studies [7–9].
There are several mechanisms whereby vitamin D might alter glucose metabolism. Vitamin D is known to have anti-inflammatory and immunomodulatory effects . This could influence the autoimmue pathology of Type 1 diabetes, and could ameliorate low-grade chronic inflammation that has been implicated in insulin resistance in Type 2 diabetes . Vitamin D may also stimulate insulin release by pancreatic β-cells [12,13]. Elevated parathyroid hormone levels, consequent on low vitamin D levels, have also been implicated in impaired insulin release from pancreatic β-cells .
In patients with diabetes, low vitamin D levels predict future macrovascular events [2,4], although evidence for prediction of microvascular events is much less strong . The macrovascular association may be attributable to effects on blood pressure , rennin–angiotensin system activity , endothelial function , vascular endothelial growth factor  or chronic inflammation. Doubt remains as to whether low vitamin D levels are causal or merely a marker of worse disease or worse health status.
Results from previous supplementation studies have been conflicting. A previous systematic review examining the effect of vitamin D supplementation  did not find sufficient evidence to draw conclusions as to the effect of supplementation on glycaemic indices. Several new studies have been published in the 4 years since this review, thus we present an up-to-date analysis of the effect of vitamin D supplementation on markers of glycaemia, insulin resistance and their sequelae.
We performed a systematic review of the literature using a pre-specified protocol according to the guidelines of the Cochrane Collaboration. The objective was to systematically review the evidence that vitamin D can improve insulin resistance and glycaemic control in patients with frank diabetes, patients with impaired glucose tolerance and patients with normal glucose tolerance. We also assessed the evidence that vitamin D supplementation can reduce progression to diabetes.
We included randomized controlled trials in the following groups: vitamin D supplementation vs. placebo, vitamin D and calcium supplementation vs. calcium supplementation alone, or vitamin D and calcium supplementation vs. placebo. Calcium was thus permitted as a co-supplement with vitamin D or as a supplement in both arms. Studies including patients with diabetes other than Type 1 or Type 2 diabetes, patients who were already on vitamin D supplementation, patients with end-stage renal failure on dialysis or studies which included patients with primary hyperparathyroidism were excluded.
Data sources and study search
We searched MEDLINE, Cumulative Index to Nursing and Allied Health (CINAHL), EMBASE and the Cochrane library. Databases were searched from the earliest available date to end of March 2011. Grey literature was searched as recommended in the current Cochrane Collaboration guidelines, using Google, and unpublished trials were sought using http://www.controlled-trials.com. Search terms used were [vitamin D OR vitamin D2 OR vitamin D3 OR cholecalciferol OR ergocalciferol OR al(ph/f)acalcidol OR paricalcitol OR doxercalciferol] AND [diabetes OR insulin resistance OR hyperglycaemia OR glucose OR glyc(a)emia OR retinopathy OR nephropathy OR peripheral vascular disease OR myocardial infarction OR cerebrovascular accident OR stroke OR beta cell]. We also hand searched reference lists of all eligible articles as well as of any previous review articles. No language restrictions were imposed; no restrictions were placed on participant age or sex.
The supplements included were vitamin D2 (ergocalciferol), vitamin D3 (cholecalciferol), 1,25 dihydoxycholecalciferol (calcitriol), 1-alpha calcidol, paracalcitol, doxecalciferol. Calcium supplements were included in the intervention group or in the comparator group.
We collected information on the following outcomes: change in insulin resistance, change in C-peptide levels, change in HbA1c, change in fasting glucose, development of microvascular complications, (retinopathy, nephropathy, neuropathy) and development of macrovascular outcomes (myocardial infarction, stroke, coronary revascularization, peripheral vascular disease). We also collected data on age, sex, type and dose of vitamin D, therapy duration, study location, together with the inclusion and exclusion criteria of each study.
Data extraction and quality assessment
Data extraction and study quality assessment were performed independently by two reviewers (PSG and MDW) using a standard proforma and discrepancies were resolved by consensus. For studies lacking a reported standard deviation of change in outcome between baseline and follow-up, we derived standard deviation of change as the mean of the baseline and follow-up standard deviations for each treatment group; similar methods have previously been used successfully in meta-analyses of blood pressure trials . Study quality and bias risk was assessed via predefined categories: quality of allocation concealment, potential for selection bias, quality of blinding, intention-to-treat analysis and comparability of groups. Funnel plots were derived for change in glucose (the commonest outcome) depicted as mean difference vs. 1 − standard error of mean difference.
Data analysis and synthesis
Tabulated data on study characteristics were prepared. Meta-analyses were performed using RevMan software version 5 (Cochrane Collaboration). Data were combined using a random-effects model as some heterogeneity of outcome was expected. The I2 statistic was calculated as an index of heterogeneity between studies, with 95% confidence intervals calculated using MIX 1.7 software (http://www.meta-analysis-made-easy.com). For each variable, mean change between baseline and follow-up for each group were compared in the meta-analysis calculation. Data for mean differences in fasting plasma glucose and HbA1c were combined; for insulin resistance and insulin and C-peptide results, data were combined using standardized mean difference (effect size) as different studies used different measures of insulin resistance. Standardized mean difference was calculated as [change in outcome/pooled standard deviation of outcome]. C-peptide is released in a 1:1 ratio with insulin, and was thus included as a substitute marker for insulin when insulin levels were not available. Data on change in insulin and C-peptide are highly correlated with insulin resistance , and were thus combined in meta-analysis with measures of insulin resistance using standardized mean difference. We pre-specified analysis by two subgroups: studies enrolling patients with normal fasting glucose at baseline (no history of diabetes mellitus, impaired fasting glucose or impaired glucose tolerance) and those with abnormal glucose tolerance (defined as either impaired glucose tolerance on glucose tolerance test or impaired fasting glucose, or meeting criteria for diagnosis of diabetes mellitus). These entities were combined as they represent parts of a spectrum of pathophysiological derangement.
Selection of trials
The initial search strategy found 341 abstracts. Thirty-six abstracts proceeded to detailed evaluation of the complete report. Fifteen studies were included in the systematic review (Fig. 1). The reasons for exclusion included lack of randomization, patients on dialysis and absence of a control group. Study details are given in Table 1.
|Study||Place of study||n||Study population||Mean age (years)||Male (%)||Baseline 25OHD level (nmol/l)||Control||Supplement (1)||Supplement (2)||Duration of supplementation|
|Nilas and Christiansen, 1984 ||Denmark||128||Post-menopausal women||55||0||NA||Placebo + calcium (500 mg)||Vitamin D3 (2000 IU) + calcium (500mg)||1-alphacalcidol (0.25 mcg) + calcium (500 mg)||2 years|
|Ljunghall et al., 1987 ||Uppsala, Sweden||65||Middle-aged men with impaired glucose tolerance||(61–65)||100||95||Placebo||1-alphacalcidol (0.75 mcg)||12 weeks|
|Orwoll et al., 1994 ||Bethesda, MD, USA||20||Type 2 diabetes mellitus||61||NA||35||Placebo||Calcitriol (1 mcg)||2 months|
|Fliser et al., 1997 ||Germany||16||Healthy males||26||100||NA||Placebo||Calcitriol (1.5 mg)||7 days|
|Major et al., 2007 ||Canada||63||Healthy, overweight women||43||0||NA||Placebo||Calcium 600 mg + vitamin D 400 IU||15 weeks|
|Pittas et al., 2007 ||Boston, MA, USA||314||222 with normal fasting glucose; 92 with impaired fasting glucose||71||52||76||Placebo||Calcium (500 mg) + vitamin D (700 IU)||3 years|
|Sugden et al., 2008 ||Dundee, UK||34||Type 2 diabetes mellitus||64||53||38||Placebo||Ergocalciferol (100 000 IU)||8 weeks|
|de Boer et al., 2008 ||USA||33 951||Healthy post-menopausal women||62||0||47||Placebo||Calcium (1000 mg) + vitamin D3 (400 IU)||7 years|
|Avenell et al., 2009 ||England and Scotland||5292||Over 70 + record of osteoporosis fracture + Caucasian||77||15||38||Placebo||Vitamin D3 (800 IU) + calcium (1000 mg)||24–62 months|
|Li et al., 2009 ||China||35||Latent autoimmune diabetes||40||77||62||Insulin + placebo||Insulin + alpha calcidol (0.25 mcg)||12 months|
|Jorde and Figenschau, 2009 ||Norway||32||Type 2 diabetes mellitus||56||56||59||Placebo||Cholecalciferol (40 000 IU)||6 months|
|Nagpal et al., 2009 ||New Delhi, India||65||Healthy, centrally obese, without diabetes||44||100||33||Placebo||Vitamin D3 (120 000 IU)||6 weeks|
|Von Hurst et al., 2010 ||Auckland||81||South Asian women with insulin resistance||42||0||20||Placebo||Cholecalciferol (4000 IU)||6 months|
|Witham et al., 2010 ||Dundee, UK||102||Type 2 diabetes mellitus||66||40||45||Placebo||Vitamin D3 (100 000 IU)||Vitamin D3 (200 000 IU)||16 weeks|
|De Zeeuw et al., 2010 ||Europe and USA||281||Type 2 diabetes mellitus with nephropathy||64||70||41||Placebo||1 μg paricalcitol||2 μg paricalcitol||24 weeks|
Study quality and bias
The quality of study design and reporting was variable (Table 2). Random allocation was a prerequisite for inclusion in the systematic review, but many study reports contained incomplete or no description of the method of random allocation. Eight out of 15 studies reported methods likely to ensure allocation concealment. Ten out of 15 stated a clear number and reason for withdrawals, with five study reports not mentioning withdrawals. Four studies reported a clear analysis on intention to treat; 10 analyses were possibly on intention to treat but lacked an unambiguous description. Funnel plots examining fasting glucose (the most commonly measured outcome) for the normal fasting glucose and abnormal glucose tolerance subgroups revealed no asymmetry to suggest publication bias (see also Supporting Information, Figures S1 and S2). We used individual patient level data from two studies [3,22] to confirm that the standard deviation of change in fasting glucose and HbA1c imputed from the start and finish standard deviation was similar to or greater than the real standard deviation for each treatment group, thus confirming that this approach did not overstate the precision of effect estimates.
|Authors||Quality of allocation concealment||Potential for selection bias||Intention to treat||Blinding—patients||Blinding—healthcare professionals||Blinding—outcome assessors||Comparability of groups|
|Nilas and Christiansen, 1984 ||U||+||U||+||+||+||+|
|Ljunghall et al., 1987 ||U||+||U||U||U||U||+|
|Orwoll et al., 1994 ||U||–||U||+||+||+||+|
|Fliser et al., 1997 ||U||–||U||+||+||+||+|
|Major et al., 2007 ||U||+||U||+||+||+||+|
|Pittas et al., 2007 ||+||+||U||+||+||+||+|
|Sugden et al., 2008 ||+||+||U||+||+||+||+|
|de Boer et al., 2008 ||+||–||+||+||+||+||+|
|Avenell et al., 2009 ||+||U||+||+||+||+||+|
|Li et al., 2009 ||U||–||U||U||U||U||+|
|Jorde and Figenschau, 2009 ||U||+||U||+||U||U||+|
|Nagpal et al., 2009 ||+||+||+||+||+||+||+|
|Von Hurst et al., 2010 ||+||+||–||+||+||+||+|
|Witham et al., 2010 ||+||+||U||+||+||+||+|
|De Zeeuw et al., 2010 ||+||+||+||+||+||+||+|
Eight studies reported fasting glucose at baseline and follow-up. Four of these eight studies enrolled participants with normal fasting glucose; the remainder enrolled those with abnormal glucose tolerance. Data from these eight studies (n = 1005) were combined using meta-analysis (Fig. 2). In the group with normal fasting glucose, there was no significant decrease in fasting glucose with vitamin D supplementation (0.01 mmol/l, 95% CI −0.06 to 0.09); results showed homogeneity (I2 = 0%, 95% CI 0–28%). In those with abnormal glucose tolerance, there was a small but significant reduction in fasting glucose in the vitamin D group compared with placebo (−0.32 mmol/l, 95% CI −0.57 to −0.07, P = 0.01), with no significant heterogeneity (I2 = 0%, 95% CI 0–67%). Omitting studies using 1-alpha hydroxylated vitamin D compounds [23,24] did not change these results (between-group difference 0.01 mmol/l, 95% CI −0.07 to 0.09 for the group with normal fasting glucose; −0.35 mmol/l, 95% CI −0.61 to −0.08 for the group with abnormal glucose tolerance); similarly, omitting the largest study  did not change the point estimate in the group with abnormal glucose tolerance (−0.30 mmol/l, 95% CI −0.90 to 0.30). Omitting studies where calcium was administered [25–27] similarly did not change the outcome in either the group with normal fasting glucose (difference = 0.01 mmol/l, 95% CI −0.21 to 0.23) or in the group with abnormal glucose tolerance (−0.30 mmol/l, 95% CI −0.90 to 0.30).
Longer-term glycaemic control
Four studies examined HbA1c as an outcome, all in patients with either impaired glucose tolerance or diabetes mellitus (Fig. 3). Vitamin D supplementation did not reduce HbA1c compared with placebo in these studies (between-group difference 0.3 mmol/mol, 95% CI −2.0 to 2.5 (0.03%, 95% CI −0.18 to 0.23%); I2 = 0%, 95% CI 0–0%); omission of one study using 1-alpha hydroxylated vitamin D derivatives  showed a similar effect [between-group difference 0.3 mmol/mol, 95% CI −2.2 to 2.7 (0.03%, 95% CI −0.20 to 0.25%)].
In six studies examining patients with abnormal glucose tolerance, data on insulin resistance [using homeostasis model assessment of insulin resistance (HOMA-IR) models] or fasting insulin/C-peptide levels could be combined in meta-analysis (Fig. 4). The standard mean difference favouring vitamin D treatment was 0.25 (95% CI 0.03−0.48, P = 0.03) without significant heterogeneity (I2 = 0%, 95% CI 0–74%), with a slightly lower effect size when omitting the largest study  (0.16, 95% CI −0.11 to 0.42); this omitted study was also the only study in this subgroup in which calcium was co-administered. These results were in contrast to those with normal fasting glucose, where no significant difference was seen (effect size 0.02, 95% CI −0.13 to 0.17. I2 = 0%; 95% CI 0–0%); this lack of effect persisted after removing studies co-administering calcium (effect size 0.00, 95% CI −0.49 to 0.48). In a group of healthy volunteers, insulin sensitivity, as measured by glucose disposal during euglycaemic clamp, was similar after both vitamin D and placebo treatment .
Endothelial function data
Two studies, both in patients with Type 2 diabetes mellitus, measured the effect of vitamin D on endothelial function, as measured by flow-mediated dilatation of the brachial artery. One study showed a significant improvement in endothelial function 8 weeks after 100 000 units of vitamin D2 (+2.35% vs. 0.06%; P = 0.05) . The other study found no significant change in endothelial function at 8 weeks with either 100 000 units (−0.8%) or 200 000 units (−1.5%) of vitamin D3 compared with placebo (−0.2%) .
One study examined the effect of paricalcitol on diabetic nephropathy ; no significant reduction in geometric mean urinary albumin/creatinine ratio was seen compared with placebo for either a 1-μg dose (−11% vs. placebo, P = 0.23) or a 2-μg dose (−18% vs. placebo, P = 0.053).
Insufficient data were available on macrovascular outcomes to perform analysis as most studies did not report this information.
Progression to new diagnosis of diabetes
Two studies examined the effect of vitamin D supplementation on the incidence of newly diagnosed diabetes cases. The MRC RECORD trial showed no significant reduction in new cases of diabetes diagnosed by self-report of diagnosis or anti-diabetic medication (adjusted odds ratio 1.11, 95% CI 0.77–1.62, P = 0.57) . The Women’s Health Initiative trial  also found no reduction in incident cases of diabetes in the vitamin D-treated group (hazard ratio 1.01, 95% CI 0.94–1.10).
Summary of evidence
The results of this systematic review suggest a weak effect of vitamin D supplementation in reducing fasting glucose and improving insulin resistance in patients with Type 2 diabetes or impaired glucose tolerance. No effect on these variables was noted in patients with normal glucose tolerance. The magnitude of reduction in fasting glucose seen in this analysis is small and, given that HbA1c, a marker of longer-term glycaemic control, did not change in patients with impaired glucose tolerance or Type 2 diabetes is of debatable clinical significance. Certainly, an effect of this magnitude is unlikely to be important for individual patients; however, it is unclear whether such a change would merit action at a population level. Given the lack of large, long-term vitamin D supplementation studies examining micro- and macrovascular outcomes, we chose to focus on surrogate measures, i.e. glycaemic control and insulin resistance, which were examined in less depth by the previous systematic review . Such outcomes remain important to both patients and clinicians in making treatment decisions in diabetes.
The quality of the included studies was moderate, with incomplete reporting; in part because of the age of many of the studies, which were published before current trial reporting standards were established. Most studies enrolled small numbers of patients.
Despite a thorough search strategy, including grey literature and trials databases, unavailable studies may exist which have not been included. We found no evidence from funnel plots to support significant publication bias, but the numbers of included studies are low, limiting the ability of such plots to detect potential bias. Our predominantly negative findings suggest that unpublished studies (which also tend to be negative) would be very unlikely to alter our conclusions to support efficacy of the intervention. A variety of target populations have been studied, mostly in small studies, and for many of these studies (e.g. those enrolling patients with normal glucose handling) it is perhaps unsurprising that no effect on glycaemic control was seen. The dose and duration of vitamin D supplementation used may not have been optimal; most studies used doses of < 2000 IU vitamin D per day, and doses as high as 5000 IU per day may be required to elevate serum 25OHD levels above the 75-nmol/l level that some commentators regard as optimum for health [31,32]. It should be noted, however, that this recommendation is derived from observational studies and evidence from supplementation studies that such a target is desirable is lacking. The small number of studies did not allow us to meta-regress baseline 25OHD levels against the degree of improvement in glycaemic indices, or to examine whether higher baseline glucose or HbA1c values may have been associated with greater improvements with vitamin D therapy. Calcium has independent associations with glucose handling, insulin secretion and vascular health [33,34], thus disentangling the effect of calcium from vitamin D in studies giving both is difficult. Only one randomized controlled trial examined the impact of vitamin D and/or calcium on microvascular outcomes in diabetes , which is the major aim of improving glycaemic variables.
We found insufficient evidence to recommend vitamin D supplementation as a way of either preventing new onset Type 2 diabetes or for improving glycaemic control in patients with Type 2 diabetes or impaired glucose tolerance. Vitamin D supplementation may have a role in modifying other aspects of the metabolic and cardiovascular derangements that accompany Type 2 diabetes , including hypertension and endothelial dysfunction, but larger, longer studies focusing on micro- and macrovascular outcomes are required, rather than continuing to focus on the surrogate measures of glucose and glycaemic control.
MDW has received grant funding from Diabetes UK, Scottish Government, Heart Research UK, ME Research UK, Tenovus Scotland and Chest Heart and Stroke Scotland to support research on vitamin D. PSG and ERP have nothing to declare.
No external funding was used to perform this study. MDW is funded by a Chief Scientist Office, Scottish Government Clinician Scientist award.