Vitamin D Supplementation, Glycemic Control, and Insulin Resistance in Prediabetics

A Meta-Analysis

Naghmeh Mirhosseini; Hassanali Vatanparast; Mohsen Mazidi; Samantha M. Kimball

Disclosures

J Endo Soc. 2018;2(7):687-709. 

In This Article

Results

Search Results and Study Selection Process

We identified a total of 1553 citations using the search keywords. After removing duplicates, 230 records remained. After screening via titles and abstracts, 123 articles remained for further evaluation. We excluded 95 articles for the following reasons: the population included persons with diabetes, children, or adolescents, or occurred during pregnancy; the duration of supplementation was less than 2 months; supplementation was provided on a monthly basis or as a large bolus dose; and studies were not placebo controlled. Studies with insufficient information, after unsuccessful attempt to obtain the information through communication with the authors, were excluded.[54–59] Twenty-eight RCTs met our eligibility criteria and were included in the meta-analysis. Details of the search process and study selection are illustrated in Figure 1.

Figure 1.

Study selection flow diagram. PRISMA flow diagram of search results following study section procedure assessing vitamin D supplementation and glycemic control among RCTs of adult population.

Risk of Bias Assessment

There was no risk of selection bias because all included studies were reported to be randomized and the allocation was sufficiently concealed. There was a lack of information on blinding of patients and personnel (n = 2) and blinding of outcome assessment (n = 2). However, all evaluated studies had a low risk of bias according to random allocation concealment, comparability of intervention groups, clear definition of inclusion/exclusion criteria, and the description of dropout/withdrawal. Bias due to attrition was a concern in one trial.[60] Intention-to-treat analyses were conducted in 15 studies. An overview of the quality of bias assessment for each study is presented in Table 1.

Characteristics of the Included Studies

The characteristics of the included studies are provided in Table 2. Included studies were published between 2007 and 2017 from different countries, including the United States (seven studies), Norway (three studies), Iran (six studies), and one study each from Canada, the Netherlands, Japan, Finland, Scotland, India, Malaysia, the United Kingdom, Australia, Austria, Denmark, and Germany. The sample size varied from 23[60] to 511.[61] Participants in six studies were females only[62–66] and males only in one study.[67] The mean age of participants ranged from 26 years[60] to 71 years.[21] The duration of follow-up across studies ranged from 2 months[64] to 5 years,[68] with a median follow-up of 22 weeks [interquartile range (IQR): 14 to 48 weeks].

Oral daily doses of vitamin D3 ranged from 10.5 μg/d (420 IU/d)[69] to 175 μg/d (7000 IU/d),[70] with 10 studies providing daily doses over 100 μg/d (4000 IU/d).[60,63,64,70–76] In 12 studies, vitamin D supplementation was provided on a weekly basis with a dose range of 500 to 2222 μg/wk (20,000 to 88,880 IU/wk), roughly equivalent to a daily dose of 72 to 318 μg (2860 to 12,700 IU/d).[43,61,63,64,68,71,73–78] The average serum 25(OH)D concentrations at baseline varied from 25 nmol/L[79,80] to 76 nmol/L,[65] with a median of 38 nmol/L (IQR: 31 to 54 nmol/L) in vitamin D–supplemented groups and 41 nmol/L (IQR: 34 to 56 nmol/L) in the placebo groups. Thirteen studies recruited subjects who were vitamin D deficient [serum 25(OH)D <50 nmol/L] or insufficient [serum 25(OH)D <75 nmol/L] at the beginning of the trial.[42,43,63,64,67,70,71,73,74,76,79–81] Cosupplementation of vitamin D and calcium occurred in four studies, which assessed vitamin D plus calcium against placebo.[21,72,73,79] A determination of prediabetes was an inclusion criterion in 11 RCTs.[15,61,67,68,71–74,79,81,82] Being overweight or obese (BMI ≥25 kg/m2) was one of the inclusion criteria in 15 RCTs.[21,42,60–62,64,67,70–73,75,79,81,82]

There were 26 studies with acceptable methodological quality and low dropout rate (<20%). These studies included a large number of participants in the intervention groups (n = 1924), an average vitamin D supplementation dose of 88 μg/d (3500 IU/d), and a median of 22 weeks (around 6 months) for the length of supplementation, a duration long enough to detect changes in measured outcomes. Each of these features improved the statistical strength of this study. Analyzed together, these data provide a diverse population.

Pooled Estimate of the Effect of Vitamin D on Serum 25(OH)D Level

All included trials measured the effect of vitamin D supplementation on serum 25(OH)D concentrations. All trials showed a significant increase in serum 25(OH)D levels in vitamin D–supplemented groups, with the overall median serum 25(OH)D concentration ≥86 nmol/L at follow-up (mean ± SD: 91 ± 25 nmol/L), compared with the placebo group (49 ± 20 nmol/L). In seven studies,[64,69,76,78–80,83] follow-up serum 25(OH)D concentrations were below 86 nmol/L, which may be related to low supplementation dose and/or high BMI. The substantial increase of serum 25(OH)D concentration in two studies[43,71] was likely related to the high dose of supplementation [300 μg/d (12,000 IU/d) and 143 μg/d (5700 IU/d)]. Overall, serum 25(OH)D concentrations in the treated arms significantly improved by 45.1 nmol/L (95% CI: 41.3 to 48.9; P < 0.001, I 2 = 97.4%; Figure 2).

Figure 2.

Forest plot of mean change from baseline in serum 25(OH)D concentrations (nmol/L) between vitamin D supplementation and control.

Pooled Estimate of the Effect of Vitamin D on Glycemic Control

HbA1c. Sixteen studies examined HbA1c as an outcome. Of the individual studies, seven reported reduced HbA1c with vitamin D supplementation,[62,67,69,71,73,82,83] and nine reported null results.[15,42,43,61,68,72,74,79,81] Vitamin D supplementation and improved vitamin D status reduced HbA1c significantly, compared with placebo, across all studies by –0.48% (95% CI: –0.79 to –0.18; P = 0.002, I 2 = 92.1%; Figure 3).

Figure 3.

Forest plot of mean change from baseline in HbA1c (%) between vitamin D supplementation and control.

FPG. The effect of vitamin D on FPG was reported in 25 RCTs. Of these studies, eight reported reduced FPG with vitamin D supplementation,[21,61,63,69,72,73,78,82] four found an increase,[42,43,62,70] and 13 studies reported no change in FPG at follow-up.[60,64–66,68,71,74–77,80,81,83] The combined data show vitamin D supplementation reduced FPG by –0.46 mmol/L (95% CI: –0.74 to –0.19; P = 0.001, I 2 = 92.4%; Figure 4).

Figure 4.

Forest plot of mean change from baseline in FPG (mmol/L) between vitamin D supplementation and control.

Pooled Estimate of the Effect of Vitamin D on Insulin Resistance and Glucose Tolerance

Fasting HOMA-IR. The influence of vitamin D supplementation on insulin resistance (using HOMA-IR) was evaluated in 20 studies. In the individual studies, eight reported a lowering effect of vitamin D on HOMA-IR,[21,63,64,68,74,76,78,80] one study found an increase,[72] and 11 found no effect in the vitamin D–supplemented group.[43,60–62,66,69–71,73,79,81] Vitamin D supplementation was found to reduce HOMA-IR across all studies by –0.39 (95% CI: –0.68 to –0.11; P = 0.007, I 2 = 91.3%; Figure 5).

Figure 5.

Forest plot of mean change from baseline in HOMA-IR between vitamin D supplementation and control.

2HPG. Ten studies contributed data on the effect of vitamin D supplementation on 2HPG. Of these studies, two reported positive effects of vitamin D supplementation,[73,82] one reported negative effects,[81] and seven reported null results.[60–62,68,71,72,79] Combined data demonstrate that vitamin D supplementation did not significantly effect 2HPG, but a nonsignificant trend was found for decreased 2HPG by –0.13 mmol/L (95% CI: –0.34 to 0.08; P = 0.2, I 2 = 69.1%; Figure 6).

Figure 6.

Forest plot of mean change from baseline in plasma glucose after 2HPG (mmol/L) between vitamin D supplementation and control.

Sensitivity analysis. In the leave-one-out sensitivity analyses, the pooled effect estimates remained similar across all studies for HOMA-IR, 2HPG, and serum 25(OH)D concentration. These results confirm that the significant difference between the studied groups reflects the overall effect of all included studies. For HbA1c, after excluding one study,[62] the effect size decreased from –0.48% (P = 0.002, I2 = 92%) to –0.31% (P = 0.01, I2 = 86%), and the effect of vitamin D supplementation on HbA1c was significant. For FPG, after excluding the study by Sun et al.,[69] the effect size decreased from –0.46 mmol/L (P = 0.001, I2 = 92%) to –0.36 mmol/L (P = 0.005, I2 = 90%), and after excluding the study by Pittas et al.,[21] the effect size decreased to –0.36 mmol/L (P = 0.004, I 2 = 89%). In both situations, the effect of vitamin D supplementation on reduced FPG remained significant.

Publication bias. For HbA1c, visual inspection of funnel plot asymmetry demonstrated a potential publication bias for the comparison of HbA1c percentage between vitamin D–administered groups and placebo groups [Figure 7(a)]. The presence of a publication bias also was suggested by Egger's linear regression (intercept = –5.13, SE = 2.33; 95% CI = –10.15 to –0.12, t = –2.2, two-tailed P = 0.04). After adjusting the effect size for potential publication bias using the "trim and fill" method, four potentially missing studies were imputed in the funnel plot and the effect size increased from –0.48% (95% CI: –0.79 to –0.18) to –0.71% (95% CI: –1.02 to –0.39) [Figure 7(b)].

Figure 7.

(a) Funnel plot of SE by standardized mean difference for HbA1c, detailing publication bias in the studies selected for analyses. Closed circles represent observed published studies. (b) "Trim and fill" method to impute for potentially missing studies for HbA1c. Four potentially missing studies were imputed in funnel plot. Closed circles represent observed published studies. Squares with circle inside represent imputed studies.

For FPG, the funnel plot was asymmetric [Supplemental Figure 1(a)] though Egger's linear regression (intercept = –3.74, SE = 1.88; 95% CI = –7.64 to 0.15, t = –1.99, two-tailed P = 0.059) did not indicate a potential bias. Using the "trim and fill" correction method, the effect size was adjusted for potential publication bias and six potentially missing studies were imputed in the funnel plot. The effect size increased from –0.46 mmol/L (95% CI: –0.74 to 0.19) to –0.72 mmol/L (95% CI: –1.02 to –0.42) [Supplemental Figure 1(b)].

Figure S1.

(a) Funnel plot of standard error by standardized mean difference for FPG, detailing publication bias in the studies selected for analyses. Closed circles represent observed published studies. (b) Trim and fill method to impute for potentially missing studies for FPG, six potentially missing studies were imputed in funnel plot. Closed circles represent observed published studies. Squares with circle inside represent imputed studies.

For HOMA-IR, funnel plot asymmetry indicated a potential publication bias in determining the effect size of vitamin D supplementation on HOMA-IR changes, compared with placebo group [Supplemental Figure 2(a)]. The presence of a publication bias was not confirmed by Egger's linear regression (intercept = –1.09, SE = 2.18; 95% CI = –5.67 to 3.49, t = –0.5, two-tailed P = 0.6). Using the "trim and fill" method to adjust the effect size for potential publication bias, five potentially missing studies were imputed in the funnel plot and the effect size increased from –0.39 (95% CI: –0.68 to –0.11) to –0.62 (95% CI: –0.92 to –0.32) [Supplemental Figure 2(b)].

Figure S2.

(a) Funnel plot of standard error by standardized mean difference for HOMA-IR, detailing publication bias in the studies selected for analyses. Closed circles represent observed published studies. (b) Trim and fill method to impute for potentially missing studies for HOMA-IR, five potentially missing studies were imputed in funnel plot. Closed circles represent observed published studies. Squares with circle inside represent imputed studies.

Visually inspected funnel plot symmetry did not indicate any potential publication bias for the comparison of 2HPG levels between the vitamin D–supplemented group and placebo groups [Supplemental Figure 3(a)]. Moreover, the Egger's linear regression (intercept = –1.19, SE = 1.48; 95% CI: –4.6 to 2.2, t = –0.81, two-tailed P = 0.4) did not detect any publication bias. Using the "trim and fill" correction and adjusting the effect size for potential publication bias, no potentially missing study was imputed in the funnel plot and the effect size remained the same (effect size: –0.13 mmol/L, 95% CI: –0.34 to 0.08) [Supplemental Figure 3(b)].

Figure S3.

(a) Funnel plot of standard error by standardized mean difference for 2HPG, detailing publication bias in the studies selected for analyses. Closed circles represent observed published studies. (b) Trim and fill method to impute for potentially missing studies for 2HPG, no potentially missing studies was imputed in funnel plot. Closed circles represent observed published studies. Squares with circle inside represent imputed studies.

For serum 25(OH)D levels, the funnel plot was asymmetric [Supplemental Figure 4(a)] and Egger's linear regression (intercept = 3.61, SE = 1.76; 95% CI = –0.005 to 7.23, t = 2.05, two-tailed P = 0.05) indicated a potential bias. Using the "trim and fill" correction method, the effect size was adjusted for potential publication bias and no potentially missing studies were imputed in the funnel plot. The effect size remained the same at 45.1 nmol/L (95% CI: 41.3 to 48.9) [Supplemental Figure 4(b)].

Figure S4.

(a) Funnel plot of standard error by standardized mean difference for serum 25(OH)D, detailing publication bias in the studies selected for analyses. Closed circles represent observed published studies. (b) Trim and fill method to impute for potentially missing studies for serum 25(OH)D, no potentially missing studies was imputed in funnel plot. Closed circles represent observed published studies. Squares with circle inside represent imputed studies.

Subgroup Analysis

Effect of population characteristics. We conducted a subgroup analysis to examine the effect of vitamin D supplementation in prediabetes in comparison with populations that were overweight/obese but not prediabetic (Table 3). The lowering effect was observed in both groups with no significant difference in the change of HbA1c and FPG between prediabetics and overweight/obese participants. Both HbA1c and HOMA-IR showed a greater reduction over time among overweight/obese individuals compared with prediabetics (HbA1c: –0.98 ± 0.45 vs –0.29 ± 0.14, P = 0.1; HOMA-IR: –0.62 ± 0.23 vs –0.07 ± 0.16, P = 0.05). There were not enough studies to perform subgroup analyses on 2HPG.

Effect of combined vitamin D and calcium supplementation. Subgroup analysis was performed to determine if concomitant calcium supplementation influenced the effects of vitamin D (Table 3). There was no significant difference in the change of HbA1c and HOMA-IR when calcium was provided in combination with vitamin D compared with vitamin D alone (HbA1c: –1.05 ± 0.74 vs –0.53 ± 0.18, P = 0.2; HOMA-IR: –0.46 ± 0.5 vs –0.38 ± 0.1, P = 0.4), although coadministration of calcium showed greater reduction in HbA1c. However, FPG (–1.67 ± 0.5 vs –0.18 ± 0.1, P = 0.002) and 2HPG (–0.54 ± 0.3 vs 0.03 ± 0.06, P = 0.02) showed a greater reduction when vitamin D was provided in combination with calcium. Overall, combining calcium with vitamin D improved its effect on glycemic control.

Influence of age on the effect of vitamin D. Subgroup analysis was conducted to determine if age influenced outcomes by comparing studies in which mean participant age in each study was less or greater than 45 years (Table 3). HbA1c showed greater improvement in populations with a mean age younger than 45 years in comparison with older populations (–1.15 ± 0.6 vs –0.30 ± 0.1, P = 0.05). Greater reduction in FPG for populations older than 45 years was not statistically significant (–0.58 ± 0.20 vs –0.31 ± 0.24, P = 0.2). Changes in HOMA-IR and 2HPG did not differ significantly between the two compared age groups.

Influence of obesity on the effect of vitamin D. We compared the effect of vitamin D supplementation on glycemic measures between studies conducted in overweight/obese and nonobese populations (Table 3). There was no significant difference in the change of HbA1c, FPG, and HOMA-IR between obese and nonobese populations. We were not able to compare 2HPG as it was reported in only one study with a nonobese population in comparison with nine studies on obese populations.

Effect of baseline vitamin D status. Participants with vitamin D deficiency, mean serum 25(OH)D concentration <50 nmol/L, at the beginning of the intervention were compared with those who had mean serum 25(OH)D concentration ≥50 nmol/L (Table 3). Greater reductions were found within HbA1c and FPG levels when baseline mean serum 25(OH)D concentration was ≥50 nmol/L, whereas the lowering effect was significantly less in the subgroup with baseline mean 25(OH)D <50 nmol/L (HbA1c: –0.79 ± 0.25 vs –0.14 ± 0.13, P = 0.04; FPG: –0.69 ± 0.21 vs –0.11 ± 0.10, P = 0.05). The lowering effect of vitamin D on HOMA-IR and 2HPG did not differ based on serum 25(OH)D status at baseline.

Effect of serum 25(OH)D concentration at follow-up. Subgroup analysis was carried out based on the concentration of serum 25(OH)D achieved at follow-up, either below or above the median level (86 nmol/L) (Table 3). Vitamin D supplementation significantly decreased HbA1c (P = 0.05), FPG (P = 0.05), and HOMA-IR (P = 0.1) to a greater extent when serum 25(OH)D concentration achieved was above 86 nmol/L. In agreement with these results, we also found dose-response effects for all four parameters. With increased vitamin D supplementation dose, there was a greater reduction in HbA1c (y = –2926.2x + 2741.3, R2 = 0.06), FPG (y = –2686.4x + 3414.3, r 2 = 0.045), HOMA-IR (y = 876.75x + 3437.5, r 2 = 0.031), and 2HPG (y = –1779x + 3461.3, r 2 = 0.153).

Effect of length of intervention. We compared the effects of vitamin D in studies of short and long duration (<6 months vs ≥6 months, respectively) (Table 3). For HbA1c, we found that vitamin D supplementation for less than 6 months provided a larger effect size on HbA1c in comparison with long durations (–0.75 ± 0.33 vs –0.25 ± 0.11, P = 0.1). However, both FPG and HOMA-IR showed a greater but nonsignificant reduction with supplementation greater than 6 months compared with shorter durations (FPG: –0.64 ± 0.22 vs –0.32 ± 0.19, P = 0.1; HOMA-IR: –0.53 ± 0.18 vs –0.24 ± 0.27, P = 0.1). The duration of supplementation did not affect the reduction in 2HPG. There were three studies included in our meta-analysis with intervention durations of 2 months[64,76,83] that found a significant reduction in HbA1c,[83] a decreased trend in FPG,[64,76,83] and significant reduction in HOMA-IR[64,76] following vitamin D supplementation. Despite a shorter duration of intervention, compared with other trials, these studies provided higher vitamin D supplementation doses (2800 to 7100 IU/d) and included vitamin D–deficient populations at baseline.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....