Effects of Vitamin D Supplementation on Glucose and Insulin Homeostasis and Incident Diabetes among Nondiabetic Adults: A Meta-Analysis of Randomized Controlled Trials

Aims Emerging evidence has suggested a mechanistic link from vitamin D metabolism to glucose and insulin homeostasis. This study is aimed at specifically quantifying the direct effects of vitamin D supplementation on indexes of glucose and insulin homeostasis as well as incidence of type 2 diabetes (T2D) among nondiabetic adults. Methods We systematically searched randomized controlled trials (RCTs) of vitamin D supplementation in nondiabetic adults in PubMed, EMBASE, and CENTRAL. Random-effects meta-analysis was conducted to pool the estimates. Results Our meta-analysis included 47 RCTs involving 44,161 nondiabetic individuals with a median trial duration of 4 months and a median dose of 4000 IU/d. Vitamin D supplementation significantly reduced fasting glucose by 0.11 mmol/L, fasting insulin by 1.47 mIU/L, and HOMA-IR by 0.32 while increasing total 25 (OH) D levels by 40.14 nmol/L. We found no significant effects of vitamin D supplementation on insulin secretion or beta cell function indexes. Based on the data from six trials involving 39,633 participants and 2533 incident T2D cases, vitamin D supplementation was not associated with the risk of incident diabetes compared to placebo (pooled relative risk: 1.01, 95% confidence interval: 0.93 to 1.08). Conclusions Our meta-analysis found that vitamin D supplementation might improve glucose and insulin metabolism without affecting the risk of T2D among nondiabetic adults.


Introduction
Because type 2 diabetes (T2D) has become an important public health problem worldwide, its prevention has become imperative [1,2]. Over the past decade, a large body of evidence from both observational and experimental studies has clearly suggested vitamin D's nonskeletal effects, especially those on individual or combined metabolic syndrome parameters such as adiposity, blood pressure, lipid metabolism, glucose intolerance, insulin resistance and secretion, and other metabolic abnormalities [2][3][4][5]. Epidemiological studies have linked low vitamin D levels to the pathogenesis of diabetes [6] and also supported the favorable effects of adequate vitamin D intake on reducing the risk of T2D [7,8]. Experimental studies have provided evidence for the direct beneficial effects of vitamin D supplementation on glucose and insulin homeostasis as well as other metabolic abnormalities in patients with diabetes [9][10][11]. However, those trials focused mainly on the treatment or adjuvant therapy effects of vitamin D on the progression of diabetes rather than on T2D onset.
Several studies have assessed associations between vitamin D and serum indexes of pancreatic β-cell function and insulin resistance that reflect the pathogenesis of T2D, including the quantitative insulin sensitivity check index (QUICKI), fasting insulin, and HbA1c [12,13]. Some but not all trials have identified beneficial effects on insulin secretion, glucose homeostasis, and insulin resistance among nondiabetic individuals [14][15][16][17][18][19]. However, these trials are limited by small sample size, short intervention period, nonrandomized treatment allocation, and the lack of objective assessment of vitamin D status and insulin or glucose homeostasis [14][15][16][17][18][19]. In particular, few trials have focused mainly on the effects of vitamin D on the primary prevention of T2D. Although two systematic review and meta-analyses have been conducted to address the effects of vitamin D supplementation on glucose homeostasis and insulin resistance [10,20], the results remain uncertain due to the inclusion of 10 or fewer randomized controlled trials (RCTs). Given the strong scientific premise and emerging evidence from new trials, there is a need for a meta-analysis to maximize statistical power to reliably quantify the direct effects of vitamin D supplementation on glucose and insulin homeostasis and incidence of T2D among nondiabetic adults.

Search Strategy.
PubMed database, EMBASE, and Cochrane Central Register of Controlled Trials (CENTRAL) were searched for relevant articles from inception to November 21, 2017. The following search terms were used: ("vitamin D" OR "vitamin D2" OR "vitamin D3" OR cholecalciferol OR ergocalciferol OR alphacalcidol OR alfacalcidol OR calcitriol OR paricalcitol OR doxerocalciferol OR "25 (OH)D") AND (pre-diabet * OR diabet * OR T2DM OR TIIDM or "Type 2 DM" OR "Type II DM" OR insulin OR Homa * OR OGTT OR "impaired fasting glucose" OR IFG OR "fasting plasma glucose" OR FPG OR hba1c OR "beta cell") AND (random * or RCT). No restrictions on language were imposed. In addition, the reference lists of retrieved papers and recent reviews were screened.

Study Selection and Outcomes of Interest.
Studies were included if they met the following criteria: (1) randomized controlled clinical trials, (2) adults (≥18 years) without diabetes, (3) oral vitamin D (vitamin D2 or vitamin D3) supplementation with or without calcium, (4) placebo or no treatment, with or without calcium as a comparator, and (5) reporting at least one of our outcomes of interest. Specifically, the primary outcomes were incidence of diabetes and changes in blood 25-hydroxy vitamin D (25 (OH) D) levels between baseline and posttrial. The secondary outcomes included indexes of glucose intolerance, insulin resistance, and insulin secretion, such as HbA1c (%), fasting glucose, fasting insulin, insulin sensitivity (QUICKI, homeostatic model assessment-insulin resistance (HOMA-IR, HOMA2-IR), homeostasis model assessment of insulin sensitivity (HOMA2%S)), β-cell function (homeostasis model assessment of β-cell function (HOMA-B, HOMA-%B, or HOMA2-%B)), and disposition index (DI). Studies of participants with diabetes, women with gestational diabetes mellitus, and patients with cancer, cardiovascular disease, chronic kidney disease, or undergoing dialysis were excluded.

Data Extraction and Quality Assessment.
The following data were extracted from the studies included using a standardized data collection form: first author, publication year, study location, and basic characteristic information of participants, including the total number, age, sex, BMI, and healthy status. Baseline level of serum 25 (OH) D, the dose and type of vitamin D supplementation, choice of control, duration of follow-up, and outcomes were also extracted.
The Cochrane Collaboration's risk-of-bias method was used to assess the methodological quality of random sequence generation, allocation concealment, blinding methods, handling of incomplete outcome data, and selective reporting. Each domain was rated as low, unclear, or high. Two blinded authors performed the literature search, study selection, data extraction, and quality assessment. Disagreements were resolved by consensus.

Statistical Analysis.
We performed meta-analyses using Der Simonian and Laird's random-effects model with inverse-variance (standard error) weighting of individual study results when data could be combined. The pooled risk ratios (RR) with their 95% confidence intervals (CIs) were calculated for diabetes risk as a binary outcome, and the pooled weighted mean differences (WMDs) with their 95% CIs were calculated for biomarkers as continuous variables. Between-study heterogeneity was calculated using I 2 statistics. The percentage of I 2 around 25% (I 2 = 25), 50% (I 2 = 50), and 75% (I 2 = 75) indicates low, medium, and high heterogeneity, respectively. To explore potential effect modifiers, we performed subgroup analyses stratified by vitamin D dose (≤1000 IU/d vs. >1000 and ≤4000 IU/d vs. >4000 IU/d) and trial duration (<3 months vs. ≥3 and <12 months vs. ≥12 months). Furthermore, we tested for a possible nonlinear dose-and time-response effect of vitamin D supplementation on each biomarker using restricted cubic spline regression analyses. For primary outcomes, a visual inspection of funnel plot and Egger's test were performed to explore potential publication bias. All analyses were performed using STATA software (version 12.0, StataCorp, College Station, TX). All statistical significance was defined as two-sided (α < 0 05) unless specified otherwise. Most RCTs (>75%) displayed low risk of bias in terms of incomplete outcome data and selective reporting. However, the extent of random sequence generation, allocation concealment, or blinding methods was unclear or at high risk of bias for one-third to half of the included RCTs.  Table 1). There was no significant difference among the subgroups stratified by duration (p for trend = 0.70) (Supplementary Table 2). Some evidence of publication bias was detected based on Egger's test (p = 0 01) and visual inspection of funnel plot (Supplementary Figure 4B). There was high heterogeneity in the meta-analyses (I 2 > 75%). Table 1 and Supplementary Table 2 Figure 6) compared to placebo. Our further subgroup analysis showed that vitamin D supplementation might decrease fasting insulin levels in a dose-(p for trend = 0.08) and durationresponse manner (p for trend = 0.08). There were no significant differences between vitamin D supplementation and placebo groups in terms of changes in HbA1c (WMD: −0.04%, 95% CI: −0.07 to 0.00%) and 2 h glucose (WMD: −0.06 mmol/L, 95% CI: −0.47 to 0.35 mmol/L).
In addition, there were no significant differences in 2 h insulin levels between vitamin D supplementation and      Figure 9E).

3.7.
Effect on the Incidence of T2D. Six trials encompassing 39,633 participants and 2533 incident T2D cases were included [19,[21][22][23][24][25]. Our meta-analysis found that vitamin D supplementation (median dose of 5714 and median duration of 12 months) had no significant effect on the risk of T2D (RR: 1.01, 95% CI: 0.93 to 1.08) ( Figure 2). Our subgroup analysis showed that high-dose vitamin D supplementation (>4000 IU/d) did not decrease the risk of T2D (RR: 0.72, 95% CI: 0.37 to 1.37). No evidence of publication bias was detected based on Egger's test (p = 0 44) and visual inspection of funnel plot (Supplementary Figure 4A). One trial that evaluated the association between vitamin D supplementation and risk of diabetes or impaired fasting glucose among adults found that 700 IU/d vitamin D supplementation for 3 years did not prevent the development of diabetes or impaired fasting glucose [26]. Of the 7 trials involving participants with prediabetes [16,17,19,21,22,25,27], four trials [19,21,22,25] and three trials [21,22,25] reported the outcomes of progression of prediabetes to diabetes and its reversal from prediabetes to normoglycemia, respectively. Vitamin D supplementation was not significantly associated with the increased risk of prediabetes progression to diabetes (RR, 0.83; 95% CI, 0.57 to 1.20) or its reversal to normoglycemia (RR, 1.20; 95% CI, 0.83 to 1.74) (Supplementary Figure 10); however, additional  trials are needed to achieve the necessary power to address such impact.

Discussion
Our meta-analysis of RCTs found that vitamin D supplementation had no significant effect on incident diabetes. Vitamin D supplementation significantly decreased fasting levels of glucose, insulin, and HOMA-IR. Overall, our study provided evidence suggesting that vitamin D supplementation with a dose > 4000 IU/d may be sufficient to improve glucose and insulin homeostasis indices among nondiabetic participants. It should be noted that we found a decrease rather than an increase in HOMA-B, which should be considered an index of beta-cell "function" rather than "activity" [28]. Given the significant heterogeneity and relatively small sample sizes in available RCTs, additional large-scale and long-term randomized controlled trials in nondiabetic participants with vitamin D insufficiency are warranted to assess the efficacy of vitamin D supplementation on the primary prevention of T2D. Several potential mechanisms have been proposed to explain the possible role of vitamin D supplementation in regulating the metabolism of glucose and insulin. A large body of literature has suggested that optimal vitamin D homeostasis is essential for both insulin sensitivity and secretion, which are fundamental to the pathogenesis of T2D. The widely accepted view was that vitamin D directly stimulates insulin receptors to enhance insulin sensitivity and insulin responsiveness for glucose transport [29,30]. The effect of vitamin D on insulin resistance may be indirect, e.g., through beneficial effects on adiposity. Another explanation has elucidated that vitamin D affects the release of insulin via binding to the vitamin D receptor in β-cells and vitamin Ddependent calcium-binding proteins [31,32]. Vitamin D may also indirectly affect calcium-dependent insulin secretion via regulation of calcium transport through β-cells [33]. However, the exact mechanisms remain unclear [30].
A majority of studies have investigated the association between vitamin D and glucose and insulin homeostasis among the patients with diabetes, while few studies have paid attention to this association among nondiabetic adults. One meta-analysis based on RCT data in nondiabetic populations suggested that vitamin D supplementation affected only fasting glucose in a subgroup with mean baseline HbA1c ≥ 8% but had no significant effect on HbA1c or HOMA-IR [9]. Similar to our present results, another meta-analysis showed that vitamin D had no effect on HOMA-IR but decreased fasting glucose and the level of HbA1c in prediabetes [10]. Moreover, we found a significant decrease in fasting insulin among the participants taking vitamin D supplementation especially among those taking a high dose (≥4000 IU/d), which suggests that a high dose of vitamin D supplementation might improve the insulin resistance. However, it should be noted that a greater reduction in fasting insulin was observed in the trials with shorter follow-up (<3 months) than those with longer follow-up (≥12 months), which might be a chance finding.
Vitamin D supplementation elevated the levels of serum 25 (OH) D and ameliorated the deficiency (or even normalized the levels) of vitamin D (serum 25 (OH) D < 30 nmol/ L). Although some studies had shown that a dose of 10,000 IU/d vitamin D improved endothelial function and blood pressure [34][35][36], the recommended dietary allowances (RDAs) of vitamin D were 600 IU/d for ages 1~70 years and 800 IU/d for ages 71 years and older, corresponding to a serum 25 (OH) D level of at least 20 ng/mL (50 nmol/liter) according to a report from the Institute of Medicine (IOM) [37]. Among the RCTs included in our study, the intervention dose of vitamin D was over the RDA. In addition, some experts thought that the prevalence of vitamin D inadequacy had been overestimated [37] and suggested that long-term vitamin D supplementation might have an adverse effect on health [38]. It should be noted that current recommendations for vitamin D supplementation (200-600 IU/d) are inadequate to achieve optimal serum 25 (OH) D levels (>90 nmol/L). Thus, there is not yet any consensus as to whether the general population needs further supplementation of vitamin D, and a uniform guideline for vitamin D supplementation is still an issue worthy of discussion. However, it is widely accepted that patients with vitamin D deficiency should supplement vitamin D intake [37,39]. Our subgroup analysis showed that vitamin D supplementation of 4000 IU/day might decrease fasting glucose, insulin, and HOMA-IR and increase QUICKI, which supports the beneficial effects of high-dose vitamin D supplementation on improving glucose metabolism.
In a review of studies of serum 25 (OH) D in relation to bone mineral density, lower extremity function, dental health, and risk of falls, fractures, and colorectal cancer, Boonen et al. found that optimal levels of serum 25 (OH) D were 90-100 nmol/L for all endpoints [40]. The average older individual requires an oral vitamin D 3 intake of at least 800-1000 IU/d (20-25 μg) to achieve a serum 25 (OH) D of 75 nmol/L [40]. The largest randomized trial, the Women's Health Initiative Clinical Trial of 33,951 initially nondiabetic postmenopausal women, did not observe any effect from daily intake of 1000 mg elemental calcium plus 400 IU vitamin D 3 on risk of incident diabetes over 7 years of follow-up [23]. However, 400 IU vitamin D 3 daily may have been too low to confer a clinical benefit. In particular, median levels of serum 25 (OH) D were raised from 42.3 to 54.1 nmol/L (roughly 12 nmol/L), which is lower than the optimal value of 75 nmol/L. Our meta-analysis found that serum 25 (OH) D was increased up to 82.22 nmol/L posttreatment among patients taking a median dose of 4000 IU/ d vitamin D supplementation for a median duration of 3 months. Vitamin D supplementation > 4000 IU/d may be an option to improve glucose and insulin homeostasis indices among nondiabetic participants.
One main advantage of our meta-analysis was the inclusion of a large number of eligible RCTs, which enhanced its reliability and maximized statistical power. Another advantage was that we specifically assessed the dose-dependent effect of vitamin D supplementation. However, some limitations of our meta-analysis merit consideration. First, there was high between-trial heterogeneity in most subgroup analyses, which invalidated the effect estimates and limited the generalization of our findings to all populations. Second, the subjects of the RCTs included came from different countries and had different lifestyles and genetic backgrounds; such characteristics may modulate the effects of vitamin D supplementation. Third, available but limited RCT data revealed discernible effects of vitamin D supplementation on some but not all commonly used indices of glucose and insulin metabolisms with statistical significance. In particular, we lacked sufficient data to test the study hypothesis that vitamin D supplementation delays or even prevents the development of T2D among nondiabetic individuals. Finally, there was some evidence of publication bias in our metaanalysis. We cannot completely rule out the possibility that this affected the significance of our results.

Conclusions
In conclusion, our meta-analysis found no effect of vitamin D supplementation on incidence of T2D but suggested a possible dose-response effect of vitamin D supplementation on improving glucose and insulin metabolism among nondiabetic adults, indicating a possible benefit of taking highdose vitamin D supplements for primary prevention of T2D. Future well-designed trials are warranted to confirm our findings and validate optimal vitamin D dosage.

Data Availability
The data used to support the findings of this study are included within the article and the supplementary information file.

Disclosure
The funding source had no role in the conception and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation of the manuscript. Huilin Tang and Deming Li are listed as co-first author.