- IR
-
insulin resistance
- MetS
-
metabolic syndrome
- 25(OH)D
-
25-hydroxy vitamin D
- RR
-
relative risk
- T2D
-
Type 2 diabetes
Vitamin D plays a critical role in the regulation of plasma Ca concentration via effects on intestinal absorption and bone metabolism( Reference DeLuca 1 ). In addition, it has also been reported to influence glucose regulation via effects on insulin secretion and action( Reference Lee, Clark and Gill 2 ). Vitamin D insufficiency, typically assessed by circulating blood levels of 25-hydroxy vitamin D (25(OH)D), has long been suspected as a risk factor for Type 1 diabetes( Reference Hypponen, Laara and Reunanen 3 ). Ecological evidence has shown higher rates of metabolic disorders including diabetes and hypertension( Reference Rostand 4 , Reference Pittas, Chung and Trikalinos 5 ) with increasing distance from the equator, suggesting possible associations of vitamin D insufficiency in areas with less sunlight. More recently, there is accumulating evidence to suggest that altered vitamin D and Ca homoeostasis may play a role in the development of Type 2 diabetes (T2D)( Reference Pittas, Sun and Manson 6 – Reference Liu, Meigs and Pittas 8 ). Although several studies have reported a protective association of vitamin D and T2D, the current findings are not consistent. A previous meta-analysis of eight observational studies showed that vitamin D intake (as measured by self-reported dietary intake) or status (as measured by blood 25(OH)D concentration) was associated with decreased risk of T2D, while pooled analysis of seven clinical trials of vitamin D supplementation did not show an effect on incident diabetes or measures of glycaemia( Reference Mitri, Muraru and Pittas 9 ).
To further clarify the inconsistencies around the associations of vitamin D with T2D, we conducted an updated literature-based meta-analysis. The present analyses differ from the previous reviews in several important ways. First, it exclusively includes studies where vitamin D status was assessed at baseline (measured as circulating blood 25(OH)D concentrations and/or dietary consumption) in apparently healthy adult participants, as opposed to vitamin D supplementations involving adults with wide-range of pre-existing clinical conditions including metabolic diseases. Second, it systematically combines the most up-to-date published data thus far to further clarify the contradictory evidence accumulated. Third, it records several clinically relevant study-level sub-groups to perform detailed exploration of potential sources of heterogeneity than previously reported. Additionally, for a comprehensive assessment, we have reviewed and quantified, where appropriate, the associations for related metabolic outcomes including metabolic syndrome (MetS) and measures of insulin resistance (IR).
Methods
Data sources and search strategy
We performed a systematic review and meta-analysis of prospective studies that evaluated the associations between baseline vitamin D status and incident T2D, MetS or IR in adults, using the Meta-analysis of Observational Studies in Epidemiology guidelines( Reference Stroup, Berlin and Morton 10 ). We searched PubMed, EMBASE and Web of Science electronic databases for prospective (cohort or nested case–control) population studies that evaluated relevant associations between vitamin D levels and metabolic outcomes. The computer-based searches combined free and MeSH search terms and a combination of key words related to Vitamin D (e.g., ‘25-hydroxyvitamin D’ and ‘cholecalciferol’) and T2D (e.g., ‘diabetes mellitus’), MetS (e.g., ‘IR syndrome’ and ‘syndrome X’) and IR (‘HOMA index’ and ‘HOMA-IR’). There were no language restrictions and publications were searched up to 10 August 2012. Additionally, reference lists of all retrieved articles were scanned to identify any other relevant studies. Data were abstracted, where available, on study, publication date, geographical location, population source, time of baseline survey, sample population, study design, sample source (plasma/serum), nature of sample (fresh or frozen and storage temperature), assay type and source, case definition, sample size, numbers of events, mean age, sex, duration of follow-up, adjustment for confounders related to sun exposure (such as season and latitude of location) and summary statistics. Studies were included if they had at least 1 year of prospective follow-up, with vitamin D measured at baseline, and recruited participants from approximately general populations (i.e., did not select participants on the basis of pre-diabetes or pre-existing T2D at baseline or participants with conditions that affect vitamin D metabolism such as chronic kidney disease). Studies in which the outcome was Type 1 diabetes were excluded because its pathophysiology is somewhat different from that of T2D( Reference Mitri, Muraru and Pittas 9 ). In addition, the search also included studies in which vitamin D status was assessed by blood 25(OH)D concentrations, dietary intake of vitamin D, or 25(OH)D score predicted from dietary vitamin D intake. We excluded studies using vitamin D supplementation as an exposure from our main meta-analysis and studies that were not based on adult populations. Literature search and data extraction were conducted in parallel by two authors. Each article was assessed using the inclusion criteria above and any disagreement regarding eligibility of an article was discussed and agreement reached by consensus and by involving a third author.
Statistical analysis
Hazard ratios, risk ratios and OR were assumed to approximate the same measure of relative risk (RR) across studies. RR with 95% CI were used as the common measures of association. To allow consistent comparisons across included studies, all individual risk estimates were standardised to involve comparisons between the top third v. bottom third of the population's baseline distribution of vitamin D values by using methods previously described( Reference Danesh, Collins and Appleby 11 ). Briefly, log risk estimates were transformed assuming a normal distribution, with the comparison between top and bottom thirds being equivalent to 2·18 times the log RR for a 1 sd increase (or equivalently, as 2·18/2·54 times the log risk ratio for a comparison of extreme quarters). When a study published more than one estimate of the association between vitamin D and risk of a metabolic outcome according to subgroups (e.g., by gender or smoking status), a within-study summary estimate was obtained. Summary RR were pooled using a random-effects meta-analysis( Reference DerSimonian and Laird 12 ). Statistical heterogeneity across studies was quantified using Cochran x 2 and the I 2 statistics, respectively( Reference Higgins, Thompson and Deeks 13 , Reference Higgins and Thompson 14 ). To investigate possible sources of heterogeneity, where appropriate, the analyses were stratified by geographical locations, study design, population source, sample type, type of vitamin D exposure (i.e., whether blood levels or dietary intake), duration of follow-up, number of events, levels of adjustment and adjustment for seasonal variations or latitude of study location. We assessed the potential for publication bias through formal tests, namely Begg's funnel plots( Reference Begg and Mazumdar 15 ) and Egger's regression symmetry test( Reference Egger, Davey Smith and Schneider 16 ). All statistical tests were two-sided and used a significance level of P < 0·05. All analyses were conducted using Stata version 12 (Stata Corp, College Station, Texas, USA).
Results
Literature search
A total of 2656 citations were retrieved from the electronic search (Fig. 1). After screening of the titles and abstracts and more detailed evaluations of full texts, we identified a total of seventeen articles (based on eighteen unique prospective studies in aggregate, comprising 210 107 participants and a total of 15 899 metabolic events). Table 1 summarises the general characteristics of all included studies that examined baseline vitamin D status in relationship with various metabolic outcomes. Thirteen of the included articles( Reference Pittas, Sun and Manson 6 , Reference Liu, Meigs and Pittas 8 , Reference Gagnon, Lu and Magliano 17 – Reference Husemoen, Thuesen and Fenger 27 ) consisting of fourteen prospective studies reported on vitamin D status and risk of diabetes (190 626 participants and 9399 incident diabetes outcomes). All included T2D studies were based on prospective cohort analyses carried out in Europe (Denmark, Finland, Germany, Norway and Sweden)( Reference Knekt, Laaksonen and Mattila 18 , Reference Thorand and Zierer 20 , Reference Deleskog, Hilding and Brismar 22 , Reference Grimnes, Emaus and Joakimsen 23 , Reference Husemoen, Thuesen and Fenger 27 ); North America (United States and Canada)( Reference Pittas, Sun and Manson 6 , Reference Liu, Meigs and Pittas 8 , Reference Kayaniyil, Retnakaran and Harris 21 , Reference Robinson, Manson and Larson 24 – Reference Pittas, Dawson-Hughes and Li 26 ); Australia;( Reference Kirii, Mizoue and Iso 19 , Reference Gagnon, Lu and Magliano 28 ) and Asia (Japan), with participants’ ages ranging from 30 to 79 years. Duration of follow-up for incident T2D endpoints ranged from 3 to 22 years, with a median follow-up of 10 years. All studies assessed vitamin D status only once at baseline with ten studies measuring serum/plasma 25(OH)D concentration( Reference Pittas, Sun and Manson 6 , Reference Knekt, Laaksonen and Mattila 18 , Reference Thorand and Zierer 20 – Reference Robinson, Manson and Larson 24 , Reference Husemoen, Thuesen and Fenger 27 , Reference Gagnon, Lu and Magliano 28 ), three studies reporting on dietary vitamin D intake( Reference Kirii, Mizoue and Iso 19 , Reference Liu, Song and Ford 25 , Reference Pittas, Dawson-Hughes and Li 26 ) and one study calculating a predicted 25(OH)D score from dietary vitamin D intake( Reference Liu, Meigs and Pittas 8 ). Ascertainment of diabetes outcomes were mostly by validated self-report or by medical record data. Overall seven studies adjusted for age, risk factors for diabetes and seasonality/latitude of study location( Reference Pittas, Sun and Manson 6 , Reference Thorand and Zierer 20 – Reference Deleskog, Hilding and Brismar 22 , Reference Robinson, Manson and Larson 24 , Reference Husemoen, Thuesen and Fenger 27 , Reference Gagnon, Lu and Magliano 28 ), another six studies adjusted for age and diabetes risk factors( Reference Knekt, Laaksonen and Mattila 18 , Reference Kirii, Mizoue and Iso 19 , Reference Grimnes, Emaus and Joakimsen 23 , Reference Liu, Song and Ford 25 , Reference Pittas, Dawson-Hughes and Li 26 ) while one study adjusted for age only( Reference Liu, Meigs and Pittas 8 ).
Fig. 1. Search strategy for the studies included in current review.
Table 1. General characteristics of the prospective studies included in the review
AusDiab, Australian Diabetes, Obesity and Lifestyle Study; CARDIA, Coronary Artery Risk Development in Young Adults Study; CHD, FHS, Framingham Heart Study; FMCHES, Finnish Mobile Clinic Health Examination Survey; JPHC, Japan Public Health Centre-based Prospective Study; M-FHS, Mini-Finland Health Survey; MONICA/KORA Augsburg, Monitoring of Trends and Determinants in Cardiovascular Disease/Cooperative Health Research in the Region of Augsburg; MRCEPS, Medical Research Council Ely Prospective Study; MS, Metabolic Syndrome; NR, PROMISE, PROspective Metabolism and ISlet cell Evaluation Cohort Study; SDPP, Stockholm Diabetes Prevention Program; WHI, Women's Health Initiative; WHS, Women's Health Study.
* Also assessed insulin resistance outcomes.
† This is the number of incident cases of Type 2 diabetes. MS was derived as a continuous risk z-score in this study.
In addition, we identified four prospective cohort studies( Reference Gagnon, Lu and Magliano 17 , Reference Liu, Song and Ford 25 , Reference Forouhi, Luan and Cooper 29 , Reference Fung, Steffen and Zhou 30 ) that evaluated baseline vitamin D status and MetS outcomes. Of these, two studies measured serum 25(OH)D concentrations( Reference Gagnon, Lu and Magliano 17 , Reference Forouhi, Luan and Cooper 29 ) and two studies reported on dietary intake of vitamin D( Reference Liu, Song and Ford 25 , Reference Fung, Steffen and Zhou 30 ). One study defined MetS using a continuous MetS risk z score using the mean of five continuous indices of obesity( Reference Forouhi, Luan and Cooper 29 ). In addition, we also identified two prospective studies reporting on vitamin D status and homeostasis model of assessment-insulin resistance (HOMA-IR) index( Reference Kayaniyil, Retnakaran and Harris 31 , Reference Forouhi, Luan and Cooper 32 ).
Vitamin D status and risk of Type 2 diabetes
The pooled RR for incident T2D for individuals in the top third v. those in the bottom third of total baseline vitamin D levels for the fourteen available studies, typically adjusted for age, diabetes risk factors and seasonality and using a random-effects meta-analysis was 0·81 (95% CI, 0·71, 0·92; Fig. 2). Studies that measured blood 25(OH)D concentrations had a pooled RR of 0·75 (95% CI 0·62, 0·91) v. RR 0·90 (95% CI 0·79, 1·03) for the three studies that assessed dietary vitamin D status (Fig. 3). There was evidence of moderate heterogeneity between the findings of the fourteen studies (I 2 = 67%; 95% CI, 42, 81; P < 0·001) with little of the difference explained by several study-level sub-groups such as geographical location, study design, population source, sample type, type of vitamin D exposure, duration of follow-up, number of T2D events and whether studies controlled for seasonality or latitude of location (Fig. 4). There was, however, evidence of publication bias as assessed by funnel plots and Egger's test (P = 0·006; Supplementary Fig. 1).
Fig. 2. Associations of vitamin D with Type 2 diabetes and other metabolic outcomes. Study acronyms and references are provided in Table 1. The summary estimate presented was calculated using a random effects model; using a fixed effects model was 0·96 (95% CI 0·94, 0·99) for diabetes, 0·86 (95% CI 0·80, 0·92) for metabolic syndrome (MS), and 0·92 (95% CI 0·83, 1·01) for insulin resistance outcomes. †Level of adjustment: +, adjusted for age and sex; + + , adjusted for diabetes/MS risk factors; + + +, adjusted for diabetes/MS risk factors plus seasonality/latitude. These studies quantified insulin resistance using homeostasis model of assessment-insulin resistance, a continuous variable.
Fig. 3. Association between vitamin D and Type 2 diabetes, measured as circulating blood and dietary exposures. Study acronyms and references are provided in Table 1. The summary estimate presented was calculated using a random effects model; using a fixed effects model was 0·97 (95% CI 0·94, 1·00) for 25-hydroxyvitamin D and 0·91 (95% CI 0·81, 1·02) for dietary vitamin D intake. †Level of adjustment: +, adjusted for age and sex; + + , adjusted for diabetes/metabolic syndrome (MS) risk factors; + + +, adjusted for diabetes/MS risk factors plus seasonality/latitude.
Fig. 4. Association of vitamin D with Type 2 diabetes, according to various study-level characteristics. The summary estimate presented was calculated using a random effects model; size of data markers is proportional to the inverse of the variance of the relative ratio. +, Adjusted for age and sex; + + , adjusted for diabetes risk factors; + + +, adjusted for diabetes risk factors plus seasonality/latitude.
Vitamin D status and risk of metabolic syndrome and insulin resistance
The combined RR for MetS outcome in a comparison of individuals in the top third v. those in the bottom third of baseline vitamin D levels, adjusted for age, MetS risk factors and seasonality was 0·86 (95% CI 0·80, 0·92), based on a random-effects model (Fig. 2). There was no evidence of heterogeneity among the contributing studies. The corresponding pooled RR for IR (i.e., HOMA-IR) was 0·84 (95% CI 0·64, 1·12). RR for MetS outcomes remained materially unchanged when they were grouped according to several study characteristics (Supplementary Fig. 2). There was no apparent evidence of publication bias as confirmed by funnel plots and Egger's test (P = 0·18; Supplementary Fig. 3). No sub-group analyses were performed for IR outcome as only two studies had relevant information.
Supplementary material
The supplementary material for this article can be found at http://www.journals.cambridge.org/pns
Acknowledgements
This research received no grant from any funding agency in the public, commercial or not-for-profit sectors. All authors declare no conflicts of interest. H. K., S. K. and R. C. conducted the literature search and did the analyses. H. K. and R. C. drafted the manuscript. O. H. F. provided critical comments during analyses and in the manuscript. The authors would like to acknowledge the contribution of Anna Deleskog, Dept of Molecular Medicine and Surgery, Karolinska Institutet, Karolinska University Hospital, M1:03, SE-17176 Stockholm, Sweden, for providing additional information to be included in the review.
Comment
We conducted an updated review and meta-analysis to clarify the nature and magnitude of the associations of vitamin D status with subsequent risks of developing T2D and other metabolic outcomes in apparently healthy adult populations. The pooled (albeit variably adjusted) results show that individuals within the top third of vitamin D levels (measured as both 25(OH)D levels and self-reported dietary intake of vitamin D) had an approximately 19% lower risk of developing diabetes compared with those in the bottom third. We found substantial heterogeneity among contributing studies which could not be completely accounted for by the study level characteristics assessed. Additionally, individuals within the top third of vitamin D levels had about 14% lower risk of developing MetS compared with those in the lowest third. Pooled results from the two studies with available information on IR outcome also demonstrated inverse, however, non-significant, association for higher vitamin D.
The findings of this study may have several mechanistic explanations. The role of vitamin D in metabolic disorders such as T2D is supported by the pleiotropic effects of this compound and its involvements in several important physiological pathways. For instance, vitamin D is involved in pancreatic β-cell dysfunction( Reference Labriji-Mestaghanmi, Billaudel and Garnier 33 – Reference Chertow, Sivitz and Baranetsky 37 ) and IR( Reference Scragg, Sowers and Bell 7 , Reference Labriji-Mestaghanmi, Billaudel and Garnier 33 , Reference Chiu, Chu and Go 34 ), the primary underlying defects characterising T2D. It has also been reported that direct action of 1,25-dihydroxyvitamin D, may lead to increased expression of insulin receptor and enhanced insulin responsiveness for GLUT( Reference Maestro, Davila and Carranza 38 , Reference Maestro, Campion and Davila 39 ). Additionally, elucidation of the supposed indirect effects of vitamin D (via regulating Ca homeostasis mechanisms) will further help in understating its mechanistic links to diabetes and related metabolic disorders.
The strengths and potential limitations of this review and meta-analyses deserve careful consideration. Our updated meta-analysis of the associations between vitamin D status and T2D updated the previous literature( Reference Mitri, Muraru and Pittas 9 , Reference Parker, Hashmi and Dutton 40 – Reference Renzaho, Halliday and Nowson 42 ) by including several recently published large-scale prospective studies. We have employed standardised RR from all individual studies to allow a consistent combination of estimates across studies. Our meta-analyses consisted of studies that had exclusively recruited participants from general populations, thereby potentially reducing any ‘reverse-causation’ effects of clinically-evident pre-existing disease on vitamin D levels. The included studies were carried out in several different geographical locations across Europe, Asia and North America, increasing the likelihood that these results can be generalised to white and mixed populations. Further studies, however, are still needed in other ethnic populations (such as cohorts involving participants from the African continent). We were unable to examine the associations with IR as information was available on only two studies (1013 participants and eighty-four events). Neither were we able to perform consistent multivariate adjustments by combining models with the same set of potential confounders owing to our reliance on published data with variable degrees of adjustments across eligible studies. Despite attempts to provide results in a consistent manner by contacting various study authors, there remained some heterogeneity among the available studies which requires further investigation. Additionally, current evidence from prospective studies on the association between vitamin D status and risk of diabetes was limited by the use of vitamin D intake or a single measurement of 25(OH)D concentrations as a surrogate for the ‘usual’ levels of vitamin D( Reference Mitri, Muraru and Pittas 9 – Reference Danesh, Collins and Appleby 11 ). A single baseline measure of circulating or dietary vitamin D may not be able to take into account the within-individual variations of vitamin D levels across seasons, as evident in our sub-group analysis. Studies are, therefore, needed with serial measurements of vitamin D status to adjust for its variability while quantifying the associations. Furthermore, studies where vitamin D status was solely measured by dietary intake are unlikely to capture overall vitamin D status in their participants as sunlight exposure is the principal source of this vitamin, and such intake assessments may also be affected by misclassification biases that are inherent to self-reported dietary instruments( Reference Willett 43 ).
Although this updated meta-analysis involved aggregate data from >200 000 participants with >16 000 incident metabolic events, it was primarily based on summary results abstracted from the published reports, thereby precluding us from undertaking more in-depth analyses (e.g., assessments of optimum levels, shapes of association, regression dilution analyses, etc.). Such detailed analyses, under a broader range of circumstances, may require collaborative pooling of individual patient-level data from prospective studies in order to yield the most consistent and precise estimates( Reference Collaboration 44 ). Finally, as our meta-analysis involves exclusively prospective studies, findings of this review are not able to establish causality, which requires robust evidence from the intervention studies. However, currently available randomised controlled trials typically involve participants with pre-existing or at high risk of metabolic disorders, are generally short-duration or underpowered and report inconsistent results( Reference Mitri, Muraru and Pittas 9 ). Therefore, carefully designed intervention studies assessing optimum dosage, long-term safety and independent effects of vitamin D supplementations on metabolic outcomes, involving particularly healthy individuals, are warranted.
In conclusion, the results of this review, based on long-term prospective studies involving apparently healthy adults, confer significant inverse associations of baseline vitamin D status with incident T2D and MetS. Interventions aimed at maintaining adequate levels of vitamin D in addition to preventing deficiency may be a useful preventive measure for metabolic disorders. Nonetheless, reliable evidence from randomised supplementation trials with adequate power, scientific rigour and participants without pre-existing metabolic diseases, will be essential to help confirm the observational findings and inform clinical guidelines.