Plasma 25-Hydroxyvitamin D and Progression to Diabetes in Patients at Risk for Diabetes: An ancillary analysis in the Diabetes Prevention Program

The DPP was a randomized, controlled clinical trial from 1996 to 2001 at 27 sites in the U.S., comparing the effects of intensive lifestyle intervention, metformin, or placebo on the development of diabetes in adults at high risk for the disease. The eligibility criteria, design, and methods of the DPP have been described in detail elsewhere (2). In brief, inclusion criteria included age ≥25 years, BMI ≥24 kg/m² (≥22 kg/m² in Asian Americans), fasting plasma glucose between 5.3 and 6.9 mmol/L (95–125 mg/dL) (≤6.9 mmol/L for American Indian sites), and plasma glucose between 7.8 and 11 mmol/L (140–199 mg/dL) after a 75-g oral glucose tolerance test. The primary exclusion was any medication known to alter glucose tolerance. The Institutional Review Board at each site approved the protocol, and all participants gave written informed consent. The Tufts University Institutional Review Board approved the present ancillary observational study.

All participants were given standard advice on healthy diet and physical activity before randomization to one of three arms: intensive program of lifestyle modification (aiming to achieve a weight reduction of at least 7% of initial body weight), standard lifestyle recommendations plus metformin, or standard lifestyle recommendations plus placebo (2). The present observational study was conducted among participants randomized to two arms, the intensive lifestyle (n = 1,079) and placebo (standard lifestyle, n = 1,082). The metformin arm was excluded to minimize the cost associated with measurement of 25-hydroxyvitamin D. Also, 122 participants were excluded because of lack of consent for ancillary studies (n = 120), or no available specimen for measurement of 25-hydroxyvitamin D at baseline or 6-month follow-up visits (n = 2). After exclusions, 2,039 participants were included in the initial multivariate analyses that included age, sex, and BMI as covariates, and 2,002 participants with complete data for all covariates were included in the additional multivariate analyses.

Plasma 25-hydroxyvitamin D concentration was measured in samples stored at −70°C from the baseline, 6-month, 1-year, 2-year, 3-year, and 4-year follow-up visits. Stability of vitamin D metabolites during transport and long-term freezing has been documented previously (25,26). Plasma 25-hydroxyvitamin D was measured at the Metabolic Laboratory at Tufts Medical Center by liquid chromatography, tandem mass spectrometry (LC/MS/MS) (Waters ACQUITY UPLC with TQD triple quadrupole mass spectrometer), certified through the National Institute of Standards and Technology (NIST) vitamin D quality assurance program (27). In the most recent testing, correlation with the NIST external standard for total 25-hydroxyvitamin D was 0.994.

The primary DPP outcome, development of diabetes, was assessed using strict laboratory criteria, according to the protocol based on oral glucose (75-g) tolerance testing performed annually and fasting plasma glucose performed semiannually or when symptoms consistent with hyperglycemia occurred (2,28). The diagnosis required confirmation by repeat testing. If a participant was started on a diabetes medication by their physician, they were asked to stop the medication and return for glucose testing to confirm the diagnosis of diabetes.

The self-reported level of leisure physical activity was assessed annually with the modifiable activity questionnaire and expressed as the average metabolic equivalent (MET-hours) per week for the previous year (2). Usual daily nutrient intake during the previous year was assessed at baseline and at the 1-year follow-up visits with the use of a modified version of the Block food-frequency questionnaire, and calcium intake was estimated as described previously (29). Data on vitamin D intake were not available. Standardized interviewer-administered questionnaires were used annually to obtain self-reported data on personal medical history, smoking, medications, alcohol use, and family medical history. Self-reported race/ethnicity was classified according to the 1990 U.S. Census questionnaire. Weight was measured using a standard calibrated scale, height was measured with a standard stadiometer, and BMI was calculated (kg/m²). Hypertension was defined as blood pressure >140/90 mmHg or use of antihypertensive medication. Fasting blood was obtained and processed according to standardized procedures. Measurement methods for hemoglobin A_1c, glucose, C-reactive protein, and creatinine have been described previously (2). To adjust for sun exposure variability at each DPP study site, we constructed an ultraviolet index for each site based on the National Weather Service data on the monthly means of ultraviolet index for each geographic location in 1997 (30).

Discrete-time proportional hazards models were used to assess the association between plasma 25-hydroxvitamin D (in tertiles) and incident diabetes to account for interval-censored data (data for diabetes status and other variables were available at 6-month intervals) and time-dependent covariates (25-hydroxyvitamin D, BMI, and physical activity). In multivariate models, we adjusted for potential confounders including DPP clinical site location (categorical variable, 1–27), age (years), sex (male or female), BMI (kg/m²), race (white, black, or other), family history of diabetes (yes or no), history of hypertension at baseline (yes or no), smoking status at baseline (never, past, or currently smoking), alcohol and calcium consumption (average of values self-reported at baseline and 1-year follow-up visits [g/day and mg/day, respectively]), plasma C-reactive protein (average of values at baseline, 6-month, and 1-year follow-up visits [mg/dL]), glomerular filtration rate using the modification of diet in renal disease estimation (31) (average of values at baseline, 6-month, and 1-year follow-up visits [mL/min/1.73 m²]), and physical activity (MET-hours per week). We also adjusted for ultraviolet radiation index at the participant’s study site (mean annual in 1997, 90 J/m²/h). Although adjustment for BMI and physical activity reflect the main effects of the intensive lifestyle intervention, to account for additional unmeasured effects of intervention, all analyses were adjusted for treatment assignment (lifestyle intervention, yes or no). The predictor (25-hydroxyvitamin D) and other variables (BMI and physical activity), whose value was measured at multiple time points, entered the analyses as time-varying “lagged” covariates. At each successive 6-month time point when the outcome (diabetes) was assessed, the value of these variables was calculated as the mean of the current and most recent available value prior to that visit. For the time-varying variables, if either the current or most recent value was missing, we imputed values using the nonmissing observation (current or most recent). If both current and most recent values were missing, then no value was imputed. Plasma 25-hydroxyvitamin D exhibited a normal distribution and entered the multivariate models on a continuous scale. For ease of interpretation, we present results after participants were categorized into three groups using tertiles (33.3rd and 66.7th percentiles) of plasma 25-hydroxyvitamin D concentration, as the mean of the values obtained at baseline and 6-month visits, to account for season differences. The hazard ratio (HR) of diabetes in each of the two highest tertiles was compared with the lowest tertile by extrapolating the per-unit change in estimated hazard from the multivariate model.

We repeated analyses using the cut points for plasma 25-hydroxyvitamin D (as the mean of the values obtained at baseline and 6-month visits), recommended for skeletal outcomes by the 2011 IOM report, as follows: <12 ng/mL (at risk for deficiency, reference group), 12–19.9 ng/mL (at risk for inadequacy), 20–29.9 ng/mL (adequacy), 30–49.9 ng/mL (potentially beneficial), and ≥50 ng/mL (potentially harmful) (32). Results are presented for categories of plasma 25-hydroxyvitamin D concentration as the mean of the values obtained at baseline and 6-month visits, based on cut points suggested by the IOM Dietary Reference Intakes for Calcium and Vitamin D report (24). The results of the HR of diabetes in each of the higher groups were compared with the lowest group by extrapolating the per-unit change in estimated hazard from the multivariate model. We also conducted secondary analyses by tertiles and IOM cut points using data only from the baseline visit to compare with the repeated measures analyses (33).

We performed subgroup analyses to examine potential effect modification by the following covariates: treatment arm (intensive lifestyle or placebo), baseline age (median cohort value, <50 or ≥50 years), sex (male or female), race/ethnicity (white or nonwhite), baseline BMI (overweight, <30 k/m², or obese, ≥30 k/m²), and baseline total calcium intake (median cohort value, <921 or ≥921 mg/day). We checked for the statistical significance of the interaction by using Wald χ² tests (34). Regression assumptions of proportional hazards and linearity were assessed for violations, and all P values were based on two-sided tests. Statistical analyses were performed using SAS version 9.2 (SAS, Cary, NC).

At baseline, the average age of the cohort was 51 years and average BMI was 34 kg/m² (Table 1). The racial distribution was diverse: ∼57% Caucasian and 21% African American, and the remaining participants were Hispanic American, Asian American or Pacific Islander, or Native American. The mean self-reported calcium intake was 1,106 mg/dL and plasma 25-hydroxyvitamin D concentration was 21.6 ng/mL. Based on the 2011 IOM report (24), 45% of the cohort had inadequate calcium intake at baseline and approximately half (49%) were at risk for deficiency or inadequacy for vitamin D (defined as 25-hydroxyvitamin D <20 ng/mL). Participants in the highest tertile of plasma 25-hydroxyvitamin D concentration were more likely to be older, male, Caucasian, and less overweight. They also had lower fasting plasma glucose, hemoglobin A_1c, C-reactive protein, and kidney function; reported higher physical activity and alcohol and calcium intake; and were more likely to reside in areas with higher annual ultraviolet index.

Participants were followed for an average of 2.7 years (range 0.5–4.5). On a continuous scale, for every 5 ng/mL increase in 25-hydroxyvitamin D, there was a 13% decrease in the risk of progression to diabetes (HR 0.87 [95% CI 0.82–0.92]; P value <0.0001 for every 5-unit increase in 25-hydroxyvitamin D). When we categorized participants into tertiles, there was a 39% lower risk among those in the highest tertile (median [IQR] 25-hydroxyvitamin D, 30.1 [27.0–34.5] ng/mL) compared with those in the lowest tertile (12.8 [10.4–14.9] ng/mL) after adjustment for age and sex (Table 2). Further multivariate adjustment for a variety of diabetes risk factors, including change in physical activity and BMI during the study duration, and randomization to the DPP intervention (lifestyle or placebo) attenuated the association, but it remained statistically significant (HR 0.72 [95% CI 0.56–0.90], comparing the third and first tertile). Additionally adjusting for ultraviolet radiation index did not change the association (data not shown). Using data from the baseline visit only, the association between 25-hydroxyvitamin D and diabetes was in the same direction but was not statistically significant (0.94 [0.75–1.17], comparing the third and first tertile).

In analyses by categories of plasma 25-hydroxyvitamin D using cut points recommended for skeletal outcomes by the 2011 IOM report (24), participants with 25-hydroxyvitamin D concentration >30 ng/mL had a lower risk compared with those <30 ng/mL, whereas those in the highest category had a 60% lower risk of developing diabetes compared with participants in the lowest category (HR 0.40 [95% CI 0.20–0.81], for 25-hydroxyvitamin D ≥50 mg/mL vs. <12 ng/mL) (Table 3). However, there were very few participants and diabetes cases in the highest category. Additionally adjusting for ultraviolet radiation index did not change the association (data not shown). Using data from the baseline visit only, the association between 25-hydroxyvitamin D by IOM cut points and diabetes was also the same direction but was not statistically significant (0.89 [0.47–1.65], for 25-hydroxyvitamin D ≥50 mg/mL vs. <12 ng/mL).

The inverse associations between plasma 25-hydroxyvitamin D and incident diabetes were generally consistent across all subgroups (Table 4). The association between 25-hydroxyvitamin D and diabetes appeared to be stronger among participants randomized to placebo compared with lifestyle (HR 0.70 [95% CI 0.52–0.94] vs. 0.80 [0.54–1.17], respectively); among those older than 50 years compared with those younger than 50 years (0.64 [0.46–0.90] vs. 0.81 [0.57–1.12]); among women versus men (0.68 [0.50–0.91] vs. 0.82 [0.55–1.19]); among obese versus nonobese (0.70 [0.52–0.93] vs. 0.78 [0.50–1.18], and among those with self-reported calcium intake ≥921 mg/dL versus <921 mg/dL (0.59 [0.42–0.82] vs. 0.84 [0.59–1.16]). However, the study had inadequate power to assess the significance of the association within subgroups, and the tests for interaction were not statistically significant in any of the strata analyzed (P for interactions >0.40) (Table 4).

In this unique cohort of U.S. adults at high risk for diabetes, higher plasma 25-hydroxyvitamin D concentrations, assessed repeatedly during the follow-up period, were associated with lower risk of diabetes, even after adjusting for weight loss and lifestyle interventions (dietary changes and increased physical activity) known to decrease diabetes risk.

Our results are consistent with those from other observational longitudinal studies that have reported on the association between vitamin D status and risk of developing type 2 diabetes (7–9,11–15,35,36). The association between higher vitamin D levels and lower risk of type 2 diabetes was statistically significant in seven published cohorts (7,9,11–13,15,36) and was in the same direction, albeit not statistically significant, in three other cohorts (8,9,35). In one cohort among older postmenopausal women, lower serum 25-hydroxyvitamin D was not associated with increased risk of developing type 2 diabetes (14).

Our study offers a methodological advantage over previous studies, which assessed vitamin D status (either by self-reported total vitamin D intake [7,35], predicted 25-hydroxyvitamin D score [11], or blood 25-hydroxyvitamin D concentration [9,12–14,36]) based on a single baseline measurement, which may not reflect long-term vitamin D status. Use of repeated measurements and cumulative averages of dietary variables has been shown to yield stronger associations than use of a single baseline measurement (33), and this was the case in our study. The need for repeated measurements is especially true with vitamin D, as single measurements of 25-hydroxyvitamin D may not reflect vitamin D status over time, owing to changes in dietary and other lifestyle changes, weight changes, sun exposure, and other relevant factors that may change during follow-up. Measuring 25-hydroxyvitamin D at multiple time points during follow-up allowed us to reduce measurement error and obtain an integrated measure that reflects long-term vitamin D status for each participant, which may be more relevant etiologically than the most remote (baseline) value (33). An alternative explanation of our results may be that plasma 25-hydroxyvitamin D is a marker of the DPP treatment effects from increased physical activity and weight loss; however, the association remained after we adjusted for the main effects of the DPP intervention (changes in physical activity and weight) during the study period. Although the analyses adjusted for these factors, residual confounding cannot be excluded. Our study has additional strengths, such as the inclusion of a large clinically relevant population at high risk for diabetes with a substantial proportion of nonwhite participants, which improves the external validity of our results because they are directly applicable to people at risk for diabetes who are in need for effective interventions to delay progression to diabetes. Also, the diagnosis of diabetes was rigorously evaluated and confirmed based on repeated laboratory measures, which is in contrast with other observational studies that ascertained diabetes by validated self-report (7,8,12,14,35), combination of self-report and laboratory measurement (11), or registry-based data (9,36). Furthermore, the method we used to measure 25-hydroxyvitamin D has been validated through the NIST vitamin D quality assurance program (27).

The results from small clinical trials and post hoc analyses of larger trials on the effect of vitamin D supplementation on glycemia or incident diabetes have been inconsistent (16–23,37–39). In these studies, vitamin D appears to have beneficial effects only in people at risk for diabetes (17,20,23), which is consistent with the findings of the current study. In contrast, vitamin D supplementation appears to have no effect among those with normal glucose tolerance, or a very small effect that would require a very large sample population to detect (17,18,22,37,39). The studies that have reported on the effect of vitamin D supplementation on glycemia in patients with established type 2 diabetes were underpowered to draw any firm conclusions.

The hypothesis that vitamin D may be relevant to prevention of type 2 diabetes is biologically plausible. Both insulin resistance and impaired pancreatic β-cell function have been reported with vitamin D insufficiency (4). Vitamin D may have a direct effect mediated by binding of the active form, 1,25(OH)₂D, to the vitamin D receptor expressed in pancreatic β-cells. The presence of the vitamin D response element in the human insulin gene promoter and transcriptional activation of the human insulin gene caused by 1,25(OH)₂D further support a direct effect of vitamin D on insulin synthesis and secretion. Finally, activation of vitamin D may occur within the β-cell by 25-OHD-1α-hydroxylase (CYP27B1), which is expressed in β-cells.

Although 25-hydroxyvitamin D is a well-established biomarker of exposure for vitamin D due to intake or biosynthesis, there is a lack of clarity regarding the validity of 25-hydroxyvitamin D as a marker of biological effects (24). The 2011 IOM Dietary Reference Intakes for Calcium and Vitamin D report concluded that adequate evidence for setting dietary reference intakes for vitamin D exists only in relation to skeletal outcomes and recommended a 25-hydroxyvitamin D level >20 ng/mL as sufficient. For nonskeletal outcomes, including diabetes, the IOM report concluded that available data are inconclusive and therefore not sufficient for dietary reference intakes development. The report also noted that levels >30 ng/mL are not consistently associated with increased benefit, whereas a level >50 ng/mL may be a “reason for concern.” Our results support the hypothesis that higher blood 25-hydroxyvitamin D (>30 ng/mL) is associated with lower risk of diabetes, and the observed inverse association was linear at all levels of 25-hydroxyvitamin D. Although our study included very few participants with very high (≥50 ng/mL) 25-hydroxyvitamin D levels, potentially limiting interpretation, this group appeared to have the lowest risk of progressing to diabetes, and our modeling of vitamin D levels as a continuous linear variable supports a benefit for this upper tail of the distribution.

In subgroup analyses, the inverse associations between plasma 25-hydroxyvitamin D and incident diabetes were consistent, across all subgroups. The association appeared to be stronger among participants randomized to placebo, those older than 50 years, obese people, women, and those with higher calcium intake. It is important to note that the tests for interaction were not statistically significant; therefore, no firm conclusions can be drawn from these subgroup analyses.

Of note, the observed inverse association between vitamin D status and incident diabetes did not appear to be affected by race (as a proxy for the altered vitamin D homeostasis in people with dark skin). These data suggest that although there are well-recognized differences in vitamin D metabolism among different race/ethnic groups (40), higher vitamin D status is associated with lower risk of diabetes among all people regardless of skin color.

In conclusion, higher vitamin D status, assessed by plasma 25-hydroxyvitamin D concentration measured repeatedly during follow-up, was associated with a lower risk of incident diabetes among people at high risk for the disease. If these results are confirmed in ongoing and planned randomized trials of vitamin D, they will have important public health implications because both of these interventions can be implemented easily and inexpensively to prevent progression of type 2 diabetes among those at high risk.

This work was supported by grants R01-DK-76092 and R01-DK-79003 (to A.G.P.) from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), the National Institutes of Health, and the National Institutes of Health Office of Dietary Supplements; Grant UL1RR025752 (to Tufts University) from the National Center for Research Resources; the U.S. Department of Agriculture Agreement 58-1950-9001 (to B.D.-H.); the Marilyn Fishman Grant for Diabetes Research (to J.M.) from the Endocrine Fellows Foundation; and Grant UO1-DK-48489 from the NIDDK to the DPP clinical centers and the Coordinating Center for the design and conduct of the DPP study. The Southwestern American Indian Centers were supported directly by the NIDDK, including its Intramural Research Program, and the Indian Health Service. The General Clinical Research Center Program, National Center for Research Resources, supported data collection at many of the clinical centers. Funding was also provided by the National Institute of Child Health and Human Development; the National Institute on Aging; the National Eye Institute; the National Heart, Lung, and Blood Institute; the Office of Research on Women’s Health; the National Center for Minority Health and Human Disease; the Centers for Disease Control and Prevention; the Indian Health Service; and the American Diabetes Association.

Lipha (Merck-Sante), Bristol-Myers Squibb, and Parke-Davis provided medication. LifeScan, Inc., Health O Meter, Hoechst Marion Roussel, Inc., Merck-Medco Managed Care, Inc., Merck and Co., Nike Sports Marketing, Slim Fast Foods Co., and Quaker Oats Co. donated materials, equipment, or medicines for concomitant conditions. McKesson BioServices Corp., Matthews Media Group, Inc., and the Henry M. Jackson Foundation provided support services under subcontract with the Coordinating Center. No other potential conflicts of interest relevant to this article were reported.

A.G.P. obtained funding, researched data, and wrote the manuscript. J.N. conducted analyses and reviewed and edited the manuscript. J.M., D.M.N., F.B.H., and B.D.-H. contributed to discussion and reviewed and edited the manuscript. W.H. researched data and reviewed and edited the manuscript. C.G. conducted measurements of vitamin D and reviewed and edited the manuscript. J.N. 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.

Parts of this study were presented in abstract form at the 71st Scientific Sessions of the American Diabetes Association, San Diego, California, 24–28 June 2011.

The investigators gratefully acknowledge the commitment and dedication of the participants of the DPP.