The Effect of Vitamin D Supplementation on Insulin Sensitivity: A Systematic Review and Meta-analysis

The International Diabetes Federation (1) has reported that the worldwide diabetes prevalence in the adult population reached 8.8% (424.9 million people) in 2017. The majority (87%–91%) of these diabetes cases concerns type 2 diabetes. Type 2 diabetes has become a major public health concern due to its complications. It decreases quality of life and increases mortality risk (2). Additionally, ∼7.3% of the adult population (352.1 million people) suffers from prediabetes (1). Prediabetes is defined by increased fasting glucose or impaired glucose tolerance (postprandial hyperglycemia), which are both strongly associated with obesity (3). Moreover, postprandial hyperglycemia is a major risk factor for developing type 2 diabetes (4).

Obesity, prediabetes, and type 2 diabetes are often characterized by low circulating levels of vitamin D (vitamin D deficiency) (5). Indeed, based on cross sectional studies, vitamin D deficiency has been linked with impaired insulin sensitivity in humans (6). Furthermore, people with low vitamin D levels (7) may be at greater risk to develop type 2 diabetes. Of interest, the vitamin D receptor, which mediates the function of vitamin D, is also expressed in insulin-sensitive tissues (including adipose tissue, muscle, and pancreas) (8–10). Therefore, currently nonskeletal functions of vitamin D in insulin-sensitive organs are being investigated.

Several plausible mechanisms have been proposed to explain a potential role of vitamin D in improving insulin sensitivity (6,11). In adipose tissue, vitamin D may affect lipid metabolism (12) and may reduce inflammation (13). Vitamin D may affect pancreatic insulin secretion via protection of β-cells from local inflammation (14). Vitamin D may also affect insulin secretion, which is mediated by a calcium-dependent mechanism (15). In addition, it has been demonstrated that vitamin D affects skeletal muscle metabolism, insulin sensitivity, and lipid composition (16,17). Thus, increasing circulating vitamin D concentration might be expected to have beneficial effects on tissue energy and substrate metabolism, thereby contributing to an improvement of whole-body insulin sensitivity.

In recent years, several meta-analyses (18–23) have been conducted to investigate the effect of vitamin D supplementation on whole-body insulin sensitivity. However, the findings from these reports are inconsistent. From the six meta-analyses published between 2015 and 2018, four studies (18–21) indicated insufficient evidence/no effect on glycemic control and insulin resistance, whereas two studies (22,23) indicated sufficient evidence for an improvement of glycemic control and insulin sensitivity. Moreover, these meta-analyses have been focusing only on glucose homeostasis under fasting conditions by studying parameters such as HOMA of insulin resistance (HOMA-IR), fasting glucose, fasting insulin, and HbA_1c as main outcome measures.

Of importance, the skeletal muscle is considered an important organ in peripheral insulin sensitivity, as it affects 70%–90% of total glucose disposal under postprandial conditions (nonfasting conditions) (24,25). Studies have shown that vitamin D may affect skeletal muscle substrate metabolism and insulin sensitivity (16,17). Thus, increasing circulating vitamin D concentration by means of supplementation might be expected to have beneficial effects during postprandial conditions by improving skeletal muscle glucose handling/insulin sensitivity. Therefore, we undertook a systematic review and meta-analysis of randomized controlled trials (RCTs) to delineate the impact of vitamin D supplementation on insulin sensitivity derived from the gold standard hyperinsulinemic-euglycemic clamp or from multisampled oral (OGTT) or intravenous (IVGTT) glucose tolerance tests as well as the postprandial glucose concentrations (OGTT) in humans with or at risk for insulin resistance.

We conducted our systematic review and meta-analysis according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (26). The protocol for our systematic review was registered at the International Prospective Register of Systematic Reviews, PROSPERO (reg. no. CRD42018092961). The main outcome of interest was the change in insulin sensitivity, derived from a gold standard hyperinsulinemic-euglycemic clamp or in the form of the Matsuda index derived from OGTT and/or insulin sensitivity index derived from IVGTT, subsequent to vitamin D administration in individuals with or at risk for insulin resistance. The secondary outcome was the change in 2-h postprandial glucose concentration and/or area under the curve (AUC) for glucose during an OGTT.

A comprehensive literature search (PubMed/MEDLINE [Medical Literature Analyses and Retrieval System Online], Cochrane Library, Web of Science, Embase database [Ovid], and CINAHL [Cumulative Index to Nursing and Allied Health Literature]) was performed to identify articles from 1980 until 31 December 2018. The main keywords used were as follows: overweight, obesity, prediabetes, type 2 diabetes mellitus, vitamin D, insulin sensitivity, insulin resistance, blood glucose. These keywords were combined with Boolean operators (e.g., OR, AND, NOT), and all fields or MeSH (Medical Subject Heading) terms. This set of search terms was slightly modified when we searched in every database due to a different system and technical limitations (Supplementary Table 1).

Eligible studies met the PICOS (Patients/participants, Intervention, Comparison/control group, Outcome, and Study design) criteria: 1) study was an RCT; 2) study population consisted of individuals with elevated (risk for) insulin resistance (overweight, obesity, prediabetes, polycystic ovary syndrome [PCOS], and type 2 diabetes without complications); 3) interventions were vitamin D supplementation versus the appropriate placebo; 5) vitamin D supplementation dose was daily, weekly, or monthly; 6) trial length was ≥2 months; 7) serum 25-hydroxyvitamin D [25(OH)D] level was measured; 8) insulin sensitivity was measured by Matsuda index derived from an OGTT and/or insulin sensitivity index derived from IVGTT or by a hyperinsulinemic-euglycemic clamp at the beginning and at the end of the trial; and 9) study was published in English.

Exclusion criteria were as follows: 1) non–clinical trial studies; 2) studies without outcome of insulin sensitivity derived from multisampled OGTT or hyperinsulinemic-euglycemic clamp; 3) study populations with end-stage renal disease (kidney disease), cancers, gestational diabetes mellitus (GDM), nonalcoholic steatohepatitis, cardiovascular diseases complications, and infectious diseases; 4) intervention periods of <2 months; 5) vitamin D supplementation provided as a single dose; and 6) study performed in children or adolescents (<18 years old).

Following the search, duplicates were removed. Titles and abstracts were screened by two authors (A.P. and M.A.v.B.). Final study selection, based on the inclusion criteria, was done by two authors (A.P. and M.A.v.B.) and approved by another author (E.E.B.). Any disagreements between the authors were resolved through discussion with the fourth author (J.W.E.J.).

Data were extracted by two authors (A.P. and M.A.v.B.). Data extracted from each study included the following items: first author, reference, year of publication, country of study, study design, inclusion criteria, sample size, form of vitamin D, dose and frequency of vitamin D supplementation, any cosupplementation (e.g., diet or calcium), treatment control group, duration of supplementation, participants’ characteristics (n, sex [% male], age, BMI, ethnicities), comorbidities, baseline and follow-up serum 25(OH)D levels, and outcome measures (e.g., M value or glucose infusion rate [GIR] or insulin-mediated glucose uptake derived from clamp, Matsuda index, insulin sensitivity index, 2-h glucose, or AUC_glucose from an OGTT) (Table 1). The corresponding author of an eligible publication was contacted by e-mail for additional data when relevant.

Any further necessary calculations on study data, such as converting measurement units or calculating SD, were conducted by the first author (A.P.) and checked by another author (M.A.v.B.). Serum 25(OH)D levels were collated in nanomoles per liter; a multiplication factor of 2.456 was used to convert 25(OH)D levels from nanograms per milliliter to nanomoles per liter. Two-h glucose tolerance test glucose concentrations (2-h glucose) were collated in millimoles per liter; we used a multiplication factor of 0.0555 to convert glucose levels from milligrams per deciliter millimoles per liter as appropriate (27).

The quality of selected RCTs was assessed independently by two authors (A.P. and M.A.v.B.) using the risk-of-bias (RoB) checklist from the Cochrane Collaboration (28). The quality assessments of the checklist included 1) bias from randomization process (random sequence generation and allocation concealment), 2) bias due to deviations from intended interventions (blinding of participants and personnel), 3) bias due to missing outcome data, 4) bias from measurements of the outcome, 5) bias from selection of reported result, and other sources of bias. Each criterion could be answered in three ways: “low risk” (adequate information), “unclear risk” (if there was unclear information), and “high risk” (if there was high concern). Study quality and the risk of bias in the eligible RCTs were systematically assessed using software Review Manager 5.3 (the Nordic Cochrane Centre, the Cochrane Collaboration, Copenhagen, Denmark, 2014).

To calculate the effect size of each study, we used the mean change and SD of the outcome measures from baseline to the end of the intervention in the control and intervention groups (29). When the outcome measure was reported as mean and 95% CI or mean and SEM, values were estimated using Review Manager 5.3 software (30). If the outcome measures were reported in median, range, or 25th–75th percentiles, mean and SD values were estimated using formulas published by Wan et al. (31). If the outcome measures were only reported in figures, we used software to estimate the value. SDs of the mean difference were estimated using the following formula:

Because the pretest-posttest correlation coefficients (r) were not reported in studies, an r value of 0.5 was assumed throughout this meta-analysis (30).

If a study included more than two intervention groups (e.g., two different doses of vitamin D), which were compared with one placebo (control) group, the number of subjects of the control group was divided by the number of comparisons. If the outcome measurement was performed at multiple time points after the intervention period, we only used first time point after the intervention (29).

The effect size is reported as standardized mean difference with its 95% CI (26,32). Standardized mean difference is used as a summary statistic in meta-analysis when the studies all assess the same outcome but measure it in a variety of ways (in this study, for example, all studies measured insulin sensitivity but use different methodology, i.e., clamp or Matsuda index or insulin sensitivity index). A random-effects model was used to estimate outcomes. Heterogeneity was assessed using the I² statistic, indicating what proportion of the variation in observed effects across studies is due to the variation in true effects, with values >60% indicating substantial heterogeneity. A P value < 0.05 is considered statistically significant (29). The analysis was performed and generated using Review Manager 5.3 software. The PRISMA checklist was used as a guide for checking the quality of our systematic review (Supplementary Table 2).

Publication bias was analyzed by visual inspection of the funnel plots.

If the heterogeneity was >60%, additional analyses were conducted. Meta-regression analysis and sensitivity analysis using the leave-one-out method (removing one study each time and repeating the analysis) were applied to gain insight into the source of the heterogeneity (29).

In total, 18 RCTs (33–50) published between 2011 and 2018 with 1,220 participants were included in the current meta-analysis (Fig. 1). Characteristics of the studies are summarized in Table 1. The studies were conducted in the U.S. (n = 5), Canada (n = 1), Iran (n = 1), India (n = 1), Malaysia (n = 1), Australia (n = 2), the Netherlands (n = 1), Denmark (n = 1), Norway (n = 1), Finland (n = 1), Sweden (n = 1), Austria (n = 1), and Italy (n = 1). Most studies included both men and women, except two studies that included only women (43,45) and two other studies that included only men (44,50). Among the 18 studies, 8 enrolled several ethnic populations (multiethnicity) (33,36–38,40,46–48), 2 enrolled individuals with African American ethnicity (34,44), 2 enrolled Asians (43,45), 1 enrolled participants from Middle Eastern countries (35), and 5 did not provide information about ethnicity (39,41,42,49,50). Sixteen trials were in participants with overweight, obesity, prediabetes, type 2 diabetes, or metabolic syndrome, and two of these trials also included apparently healthy participants (40,50), while participants in the two other trials had PCOS (45) or a history of GDM (43). Studies used different cutoffs for baseline vitamin D levels: nine studies (34,37–40,45–47) recruited individuals with vitamin D levels <50 nmol/L, and the others (33,35,36,41,42,44,48–50) included individuals with vitamin D levels <75 nmol/L at the beginning of trial (Table 1). Mean serum 25(OH)D concentrations at baseline varied from 19.2 nmol/L (45) to 59.9 nmol/L (33) in vitamin D–supplemented groups and 16.9 nmol/L (45) to 61.4 (33) in placebo groups. The mean oral dose of vitamin D was 4,608 IU per day (range 1,200 [37] to 12,695 [36] per day), with the majority of studies providing daily doses ≤4,000 IU (33,34,37–39,41,43,45,46,48–50). The duration of vitamin D supplementation ranged from 2 (42) to 12 (44) months.

There were 18 studies with sufficient data to be included in the meta-analysis to estimate the overall effect of vitamin D supplementation on insulin sensitivity derived from a clamp (39,42,46,47,49) or indices of insulin sensitivity derived from an OGTT or IVGTT (33–38,40,41,43–45,48,50). The total number of included individuals was 1,220; of these, 612 received vitamin D supplementation and 608 received placebo. From the 13 studies that used the Matsuda index derived from OGTT and/or insulin sensitivity index from IVGTT (33–38,40,41,43–45,48,50), none reported a significant change (increase or decrease) in insulin sensitivity after vitamin D supplementation compared with placebo. Furthermore, among those five studies in which the outcome of insulin sensitivity was derived from a clamp (39,42,46,47,49), only one study (Cefalo et al. [49]) reported a statistically significant improvement of insulin sensitivity.

Based on a random-effects meta-analysis, comparing the standardized mean difference of the change in insulin sensitivity from baseline between the vitamin D–supplemented and placebo groups, the overall effect on insulin sensitivity was not significant after vitamin D supplementation (standardized mean difference −0.01, 95% CI −0.12 to 0.10, P = 0.87; I²= 0%, P = 0.60) (Fig. 2).

There were 12 studies that reported 2-h glucose concentrations (6 studies) (33,34,36,41–43) or AUC_glucose (6 studies) (35,37,45,48–50) as an outcome measure. A meta-analysis of these data including 745 individuals (n = 379 treated with vitamin D and n = 366 with placebo) was performed to compare the mean change in 2-h glucose and/or AUC_glucose between the beginning and the end of study. Vitamin D supplementation had no effect on postprandial glucose, with a standardized mean difference of 0.09 (95% CI −0.08 to 0.25, P = 0.29; I² = 18%, P = 0.26) (Fig. 3A).

All 18 trials (n = 1,223; 613 individuals treated with vitamin D and 610 individuals with placebo) reported the change in serum vitamin D level after the intervention. There was a significant increase in the serum vitamin D level in the vitamin D supplementation groups compared with the control groups (standardized mean difference 2.25, 95% CI 1.90–2.60, P < 0.001) (Fig. 3B) with considerable heterogeneity (I² = 81%, P < 0.001).

Most included studies had a low risk of bias according to randomization process (random sequence generation and allocation concealment), blinding participants (comparability of intervention groups and placebo [control] groups), and incomplete outcome data (clear description of dropout/withdrawal/attrition rate). The blinding of outcome assessment was not clearly mentioned in some of the studies (34–36,41,44,45,48,49), which resulted in unclear risk of detection bias. The overall quality assessment of the included trials in this meta-analysis is shown in Supplementary Fig. 1.

The visual inspection of funnel plots of changes in insulin sensitivity and postprandial glucose did not suggest potential publication bias. In addition, the funnel plot of serum vitamin D changes in this meta-analysis following vitamin D supplementation also did not indicate publication bias (Supplementary Fig. 2).

Regarding the change in serum vitamin D levels following supplementation, there was a considerable heterogeneity (I² = 81%). A meta-regression analysis of dose or duration with standardized mean difference of serum vitamin D levels showed that the dose of vitamin D supplementation may partly explain the heterogeneity (Supplementary Table 3). Further sensitivity analysis by leave-one-out analysis shows that no single study is responsible for the heterogeneity of changes in serum vitamin D levels.

We conducted a systematic review and meta-analysis of 18 RCTs to determine the effect of vitamin D supplementation on insulin sensitivity in individuals with or at risk for insulin resistance. In this meta-analysis, vitamin D supplementation had no significant effect on insulin sensitivity derived from a hyperinsulinemic-euglycemic clamp or the Matsuda index derived from an OGTT and/or insulin sensitivity index from IVGTT. Additionally, no significant changes in 2-h glucose or AUC_glucose during an OGTT were observed following vitamin D supplementation.

In our analysis, study populations were often characterized by the presence of overweight or obesity (33–39,41–44,46–49), accompanied by baseline serum concentrations of 25(OH)D <50 nmol/L (33–44,46–50) or <75 mmol (35,36,41,42,44,49,50). Indeed, it has been described that human obesity often coincides with low circulating 25(OH)D levels (5). Several mechanisms, such as uptake/sequestration of vitamin D within adipose tissue (51) and volumetric dilution due to increased body volume (52), may explain the link between obesity and low vitamin D levels. Furthermore, a comprehensive systematic review by Autier et al. (53) suggests that the low serum vitamin D levels could also be due to the chronic low-grade inflammation, which is an important characteristic of individuals with obesity and type 2 diabetes. Thus, vitamin D supplementation in human overweight/obesity could be expected to increase serum 25(OH)D levels.

In this meta-analysis, we showed that vitamin D supplementation increased the mean 25(OH)D level of treatment groups in all studies, although with considerable heterogeneity. Additional meta-regression analysis showed that some of the heterogeneity is explained by dose but not by treatment duration. Nevertheless, in our meta-analysis, improving 25(OH)D concentrations did not translate into an improvement in insulin sensitivity and glucose metabolism. In the present meta-analysis, we included only studies using standardized methodologies (nonfasting measures of insulin sensitivity), by multisampled OGTTs or IVGTTs (33–38,40,41,43–45,48,50) or the gold standard hyperinsulinemic-euglycemic clamp (39,42,46,47,49). Our meta-analysis, which shows no effect of vitamin D supplementation on peripheral insulin sensitivity and postprandial glucose handling, supports several other findings of studies that used fasting indices of insulin sensitivity (e.g., HOMA-IR) (18,19).

It has been shown that postprandial glycemia is strongly associated with obesity (3) and predicts cardiovascular events among insulin resistant individuals (4,54). Wood et al. (55) and Manson et al. (56). have shown that vitamin D supplementation did not result in the prevention of cardiovascular events. The most recent work by Pittas et al. (57) showed that vitamin D supplementation (4,000 IU) did not result in a significantly lower risk of diabetes, assessed by 2-h OGTT in 2,423 individuals. The skeletal muscle is a key organ for peripheral insulin sensitivity, as it is responsible for the majority of glucose disposal during postprandial conditions (24,58). Although effects of vitamin D on glucose responsiveness, insulin receptor substrate (59) and insulin sensitivity (17), and muscle mitochondria biogenesis (16) as well as lipid metabolism (17) have been documented at the transcriptional and posttranslational level in vitro, our analysis showing no effect on postprandial glucose concentrations suggests that these beneficial effects do not translate into beneficial effects at the functional level.

Studies have shown that vitamin D may affect insulin secretion (15,33) rather than insulin sensitivity per se. Vitamin D may contribute to normalization of extracellular calcium, ensuring normal calcium flux through cell membranes, and regulate the β-cell calcium pool (15). Nevertheless, whether there is sufficient evidence that vitamin D supplementation may improve insulin secretion in the postprandial condition (e.g., disposition index) needs further investigation.

Furthermore, there is only one trial included in this meta-analysis, by Cefalo et al. (49), that reported a significant increase of insulin sensitivity derived from a clamp after vitamin D supplementation. Of note, in that study vitamin D supplementation was combined with a low-calorie diet intervention, yet the sample size of this study was relatively small. Therefore, whether combining vitamin D supplementation with diet restriction may improve insulin sensitivity in humans with or at risk for insulin resistance needs further investigation in a well-controlled study.

In interpretation of the results, several factors should be taken into consideration. The optimal time of intervention and dose that are necessary to evaluate the effects of vitamin D supplementation on parameters related to glucose metabolism and insulin sensitivity are not well established. We showed that neither dose nor treatment duration was associated with changes in insulin sensitivity and postprandial glucose. It also should be noted that the results could be affected by other factors such as variation in age, season (19), and ethnicity (23) as well as tissue-specific metabolism and insulin resistance (60). A recent meta-analysis in populations with type 2 diabetes by Li et al. (23) has suggested that vitamin D supplementation may have more beneficial effects in individuals with Middle Eastern ethnicity as compared with other ethnicities, but some of the included studies on which this suggestion was based were of questionable quality. In this meta-analysis, we could not perform subgroup analysis based on ethnicity because 8 out of 18 trials (33,36–38,40,46–48) were conducted in multiethnic populations, whereas 5 other studies (39,41,42,49,50) did not report any information about ethnicity. In addition, the heterogeneity regarding the change in serum vitamin D levels may also be partly explained by compliance that was achieved during intervention, which we were unable to ascertain in more detail. Another limitation might be that the number of individuals who participated in the included studies was relatively small, possibly due to the invasive and expensive character of the measurement (e.g., clamp).

Despite the fact that the number of studies using clamps, the gold standard for the measurement of insulin sensitivity, was relatively small (only 5 studies), the outcome on insulin sensitivity derived from clamps was similar to the outcome based on the results from OGTTs and IVGTTs (13 studies). In addition, three studies that met our inclusion criteria and used a hyperinsulinemic-euglycemic clamp (61–63) were published in 2019—after the window for being included in the meta-analysis. After repeating of the meta-analysis with in total 21 studies (3 studies in 2019 and the original 18 studies), the outcome did not change (Supplementary Fig. 3A–C). In conclusion, this systematic review provides no evidence that supplementation with vitamin D has a beneficial effect on peripheral insulin sensitivity, as determined by hyperinsulinemic-euglycemic clamp, the Matsuda or insulin sensitivity index, and postprandial glucose concentrations after an OGTT in people with or at risk for insulin resistance.

Funding. A.P. was supported by Indonesia Endowment Fund for Education (LPDP) during the preparation of the manuscript.

The funder had no role in the preparation, study design and analysis, or publication of the manuscript.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.P., E.E.B., J.W.E.J., and M.A.v.B designed the study. A.P. and M.A.v.B. searched databases and performed the selection of studies. A.P. analyzed the data. M.A.v.B checked the data analysis. A.P. wrote the manuscript. M.A.v.B., J.W.E.J., and E.E.B critically evaluated the review and meta-analysis, commented on it, and approved the last version. All authors reviewed and approved the final manuscript. E.E.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.