Data for this study derive from the SEARCH Study and the SEARCH Nutrition Ancillary Study. The parent SEARCH study [17,18] ascertained cases of diabetes diagnosed when subjects were < 20 years of age, starting in 2002 and continuing through to the present. Participants with newly diagnosed diabetes in 2002–2005 were invited to participate in a baseline research visit (mean diabetes duration at visit 10.5 months) and two follow-up visits approximately 12 and 24 months after their baseline visit. At each research visit, fasting blood samples were obtained from metabolically stable participants (defined as no episode of diabetic ketoacidosis during the previous month), physical measurements were conducted and questionnaires were administered. For SEARCH participants with Type 1 diabetes diagnosed in 2002–2005, SEARCH Nutrition Ancillary Study collected additional nutritional data, including plasma measures of 25(OH)D (nmol/l) obtained from frozen samples (stored at −80 °C) collected at the baseline and first follow-up SEARCH visits. Both studies were reviewed and approved annually by the local Institutional Review Board(s) that had jurisdiction over the local study population and complied with the Health Insurance Portability and Accountability Act. All participants provided informed consent and/or assent.

Fasting blood samples were analysed to measure diabetes autoantibodies, HbA_1c and lipids. Glutamic acid decarboxylase 65 (GAD65) and insulinoma-associated 2 (IA-2) diabetes autoantibodies were analysed using a standardized protocol and a common serum calibrator developed by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)-sponsored standardization group [19]. Cut-off values for positivity were GAD65 ≥ 33 NIDDKU/ml and IA-2 5 ≥ NIDDKU/ml for IA-2 [19]. HbA_1c was measured in whole blood with an automated non-porous ion-exchange high-performance liquid chromatography system (model G-7; Tosoh Bioscience Inc., San Francisco, CA, USA). Lipid measurements, including triglycerides, were performed using Roche reagent on a Roche Modular-P Autoanalyzer (Roche Diagnostics, Indianapolis, IN, USA). Human leucocyte antigen class II genotyping (HLA DR/DQ) was performed with a PCR-based sequence-specific oligonucleotide probe system in the laboratories of the University of Washington (Seattle, WA, USA) and Roche Molecular Systems Pleasanton, CA, USA).

BMI (kg/m²) z-scores from measured height and weight were derived for age and gender using the US Centers for Disease Control and Prevention (CDC) National Center for Health Statistics growth charts [20]. Waist circumference was measured using the National Health and Nutrition Examination Survey (NHANES) III protocol [21].

A surrogate measure of insulin resistance, appropriate for youth with Type 1 diabetes, was derived from the following equation: Insulin sensitivity score = exp[4.64725 − 0.02032*(waist circumference, cm) − 0.09779 × (HbA_1c, − 0.00235 × (triglyceride, mg/dl)]. A detailed description of the development and validation of this equation has been published elsewhere [22]. Briefly, a subset of participants from SEARCH(n = 85; ages 12–19 years) underwent a 3-h euglycaemic–hyperinsulinaemic clamp to measure glucose disposal rate. Multivariate linear regression was used to estimate a surrogate measure insulin sensitivity score, with waist circumference explaining 60% of the variance in measured glucose disposal rate. The formula was reproduced and validated in youth with diabetes (Type 1 and Type 2) and healthy control subjects. Consistent with previous SEARCH analysis [3], the insulin sensitivity score was dichotomized to define insulin resistance (< 8.15) and insulin sensitivity (≥ 8.15).

Self-reported race and ethnicity were collected using 2000 US Census questions and, for this analysis, categorized as non-Hispanic white, African-American and ‘other’ [23]. Information about treatment regimen, including type(s) of insulin, total daily insulin dose, frequency of insulin injections or use of continuous subcutaneous insulin infusion (insulin pump) were also collected.

For participants ages ≥ 10 years, SEARCH ascertained physical activity status and pubertal development via questionnaire [24]. Moderate to vigorous physical activity was defined as ≥ 3 days/week of any activity that either tones or makes one sweat. Pubertal development was self-reported using the standard technique described by Marshall and Tanner [25,26]. For a sensitivity analysis, participants who were aged < 10 years were assumed to be prepubertal.

25(OH)D was measured using the direct, competitive chemiluminescence immunoassay developed by Dia Sorin (Stillwater, MN, USA) [universal naming convention (UNC) detectable range, 5–320 nmol/l; intra-assay coefficient of variation, 11.0%], based on a linkage between specific vitamin D antibody coated magnetic particles and an isoluminol derivative. This method uses an antibody as a primary binding agent and measures 25(OH)D₂ and 25(OH)D₃ [27]. There were n = 12 at baseline and n = 3 at follow-up, with 25(OH)D values below the detectable limit and set to 4.9 nmol/l for this analysis. For descriptive analysis, 25(OH)D was categorized as: (1) risk of deficiency (< 30.0); (2) risk of inadequacy (30.0–49.9); (3) sufficient (50.0–125.0); and (4) possibly harmful (> 125.0) [28]. For multivariate analysis, 25(OH)D was utilized as a continuous variable.

‘Type 1 diabetes’ was defined by physicians’ report of ‘Type 1’, ‘Type 1a’ or ‘Type 1b’ diabetes. Cross-sectional analysis included data from individuals with Type 1 diabetes who had measurements of 25(OH)D and an insulin sensitivity score at baseline, resulting in an analytic sample of 1426 individuals. For longitudinal analyses, participants from the cross-sectional analytic sample who had insulin resistance data at the 12- and/or 24-month follow-up visits were selected for inclusion. Given our interest in determining the association between 25(OH)D and incident insulin resistance, participants who were insulin resistant at baseline were excluded, resulting in a longitudinal sample of 735 participants. For participants who had measures of insulin resistance from two follow-up visits, data from their 24-month follow-up visit was used.

Analyses were performed using SAS (version 9.2; SAS Institute, Cary, NC, USA), with P < 0.05 indicating significance, except where otherwise stated. Comparisons of baseline percentages and mean values by participant characteristics were examined using χ²-tests and analysis of variance (ANOVA). Trends in exposure and outcomes were examined using paired t-tests and multiple linear regression in the sub-population of individuals who had measurements at both baseline and follow-up and were not insulin resistance at baseline.

To determine the association of 25(OH)D with insulin resistance at baseline, a series of multivariate logistic regression models were fitted, first adjusted for diabetes duration only, then for covariates shown to be associated with 25(OH)D and/or insulin resistance (age, gender, race/ethnicity, HLA genotype, insulin regimen, clinical site and season of visit). Although pubertal status and physical activity may also be confounders, these data were collected in participants aged ≥ 10 years only and we wanted to retain the full, larger sample for primary analyses. Thus, pubertal status and physical activity were included in subsequent separate models for the subset of individuals with these measures. Further, previous research suggests that adiposity may be a mediator and/or confounder of the association between 25(OH)D and insulin resistance; consequently, additional analyses were conducted including BMI z-score in the full, larger sample. Finally, we had an a priori interest in determining whether the association between 25(OH)D and insulin resistance varied by disease duration, age, HLA genotype, race/ethnicity and BMI z-score; thus, we examined effect modification using separate interaction terms and likelihood ratio tests (P < 0.1). To facilitate clinical interpretation, odds ratios and 95% confidence intervals were obtained by comparing a difference of 40 nmol/l in 25(OH)D, which reflects a comparison between being at risk for 25(OH)D deficiency (25 nmol/l) and being sufficient for 25(OH)D (65 nmol/l).

Using the longitudinal sample to determine if baseline 25(OH)D is associated with incident insulin resistance, we used a similar series of multivariate logistic regression models to that of the cross-sectional models, but also included the change in time-varying covariates. As with the cross-sectional models, the estimated odds ratio and 95% confidence intervals reflect the odds of incident insulin resistance, given a 40 nmol/l change in 25(OH)D.

As the use of a dichotomous cut point to define insulin resistance would reduce our power to identify a small association, we repeated all multivariate analysis using linear regression to determine the relation between 25(OH)D and the continuous insulin sensitivity score. Finally, for a sensitivity analysis, all multivariate analyses were repeated in the subsample of participants who had physician-diagnosed Type 1 diabetes and had at least one instance of positive diabetes autoantibodies (GAD65 or IA-2: n = 1211).

Forty-nine per cent of individuals were at risk of deficiency or at risk of insufficiency for vitamin D [25(OH)D < 50.0 nmol/l]. Compared with those who were vitamin D sufficient, these participants were more likely to be non-white, older at baseline and to have had their research visit during the winter months (Table 1; all P < 0.05). They also tended to have lower insulin sensitivity scores, and higher BMI z-scores, triglycerides and waist circumference (all P < 0.05). Individuals with insulin resistance were older, had lower 25(OH)D, longer diabetes duration and higher BMI z-score, HbA_1c, triglycerides and waist circumference (Table 2; all P < 0.05). Further, the prevalence of insulin resistance was significantly higher among race/ethnic minorities than non-Hispanic white people.

Multivariate analysis revealed that the cross-sectional association between 25(OH)D and insulin resistance did not vary by age at baseline, diabetes duration, HLA genotype and BMI z-score (each interaction P ≥ 0.1); thus, the interaction terms were not included in any of the models.

Higher 25(OH)D was significantly inversely associated with insulin resistance (Table 3; model 2), both before and after adjustment for diabetes duration, age, race/ethnicity, gender, insulin regimen, HLA genotype, clinical site and season of clinical visit. Participants who had sufficient levels of 25(OH)D (65 nmol/l) had an adjusted odds ratio of insulin resistance that was 0.70 times that of participants who were at risk of 25(OH)D deficiency (25 nmol/l) (95% CI 0.57– 0.85). Despite smaller sample sizes, because of restriction by design to participants ages ≥ 10 years, separate adjustment for pubertal status (model 3) and physical activity (model 4) yielded similar, statistically significant associations. However, as expected, adjustment for BMI z-score (model 5) in the larger cross-sectional sample resulted in an attenuation of the association (odds ratio 0.81, 95% CI 0.64–1.03). Sensitivity analyses among participants with at least one instance of positive diabetes autoantibodies resulted in similar findings (results not presented).

In parallel analysis using multivariate, linear regression, a 40-unit increase in 25(OH)D was associated with a higher insulin sensitivity score (model 2 covariate adjustment: β-coefficient = 0.50, 95% CI 0.33–0.68, P < 0.001). The association was attenuated but remained significant after further adjustment for BMI z-score (β-coefficient = 0.26, 95% CI 0.11–0.40, P = 0.004).

The longitudinal sample was similar to the cross-sectional sample at baseline (see Table 1) with respect to gender, HLA risk group, clinical site and season of visit (all P ≥ 0.05). However, the longitudinal sample had a higher percentage of non-Hispanic white youth (81.4%), was younger (10.3 ± 3.6 years), had higher 25(OH)D (63.0 ± 35.6 nmol/l), insulin sensitivity scores (12.0 ± 2.4) and shorter disease duration (9.0 ± 5.0 months) (all P < 0.05) than the cross-sectional sample. For individuals who had insulin sensitivity scores at both the baseline and follow-up examinations, insulin sensitivity scores declined significantly, from mean 12.0 ± 2.4 at baseline to 9.9 ± 2.8 at the follow-up visit (P < 0.01), suggesting worsening of insulin sensitivity with Type 1 diabetes progression.

Higher 25(OH)D at baseline was associated with reduced risk for incident insulin resistance at follow-up after adjustment for diabetes duration and time between research visit odds ratio 0.69, 95% CI 0.63–1.14) (Table 4). However, this association was no longer significant after adjustment for other potential confounders. Sensitivity analyses among participants with at least one instance of positive diabetes autoantibodies, resulted in similar findings (results not presented).

In parallel analysis using multivariate, linear regression, a 40-unit increase in baseline 25(OH)D appeared to be associated with higher insulin sensitivity score in follow-up (model 2 covariate adjustment: β-coefficient = 0.12, 95% CI –0.04 to 0.27), but was not significant P = 0.15).