Article Text

Download PDFPDF

A cross-sectional study of vitamin D and insulin resistance in children
  1. Andrea Kelly1,2,
  2. Lee J Brooks2,3,
  3. Shayne Dougherty1,
  4. Dean C Carlow4,5,
  5. Babette S Zemel2,6
  1. 1Division of Endocrinology/Diabetes, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA
  2. 2Department of Pediatrics, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania, USA
  3. 3Sleep Center, Division Pulmonary Medicine, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA
  4. 4Department of Pathology and Laboratory Medicine, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA
  5. 5Department of Pathology and Laboratory Medicine, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania, USA
  6. 6Division of Gastroenterology/Nutrition, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA
  1. Correspondence to Dr Andrea Kelly, Division of Endocrinology/Diabetes, 1130 Northwest Tower, The Children's Hospital of Philadelphia, 34th & Civic Center Boulevard, Philadelphia, PA 19104, USA; kellya{at}


Objective Vitamin D deficiency is common and has been associated with several non-bone/calcium related outcomes. The objective was to determine the association between serum 25-hydroxyvitamin D (25-OH-D) and fasting glucose, insulin and insulin sensitivity in obese and non-obese children.

Patients/setting/design Cross-sectional study of 85 children aged 4–18 years recruited from the local Philadelphia community and Sleep Center.

Main outcomes measures Fasting blood glucose, insulin and 25-OH-D were measured. Insulin resistance was calculated using homeostasis model assessment (HOMA). Body mass index standard deviation scores (BMI-Z) and pubertal stage were determined. Multivariable linear regression was used to determine factors associated with decreased 25-OH-D and to determine the association of vitamin D with HOMA.

Results Median 25-OH-D was 52 nmol/l (IQR 34–76). 26% of subjects were vitamin D sufficient (25-OH-D ≥75 nmol/l), 27% had intermediate values (50–75 nmol/l) and 47% were insufficient (25–50 nmol/l) or frankly deficient (<25 nmol/l). In the multivariable model, older age, higher BMI-Z and African–American race were all negatively associated with 25-OH-D; summer was positively associated with 25-OH-D. Lower 25-OH-D was associated with higher fasting blood glucose, insulin and HOMA after adjustment for puberty and BMI-Z.

Conclusion Low 25-OH-D, common in the paediatric population at risk for diabetes (older children, African–Americans, children with increasing BMI-Z) is associated with worse insulin resistance.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Functional studies of calcium absorption and parathyroid hormone in adults have revealed that 25-hydroxyvitamin D concentrations (25-OH-D) greater than 75 nmol/l are needed for positive calcium balance.1,,3 This optimal range is much higher than previously recognised. Recently, the Institute of Medicine's 2010 consensus report Dietary Intakes for Calcium and Vitamin D challenged this threshold and set an optimal 25-OH-D level as ≥50 nmol/l based upon bone health outcomes.4 The Endocrine Society, National Osteoporosis Foundation, American Society of Bone and Mineral Research and other sponsoring institutions are in the midst of revising guidelines; their recommendations for optimal 25-OH-D have yet to be published. In children, a scarcity of research has prevented an evidence-based definition of ‘optimal’ 25-OH-D concentration. In 2008 the American Academy of Pediatrics, responding to the re-emergence of rickets, updated its guidelines for vitamin D supplementation and recommended a serum 25-OH-D >50 nmol/l, a threshold based upon the prevention of rickets.5 It could not, however, recommend a threshold (if one exists) for maximising bone health or addressing other vitamin D-related outcomes. Nonetheless, numerous reports indicate that serum 25-OH-D values considered by some experts representative of vitamin D deficiency/insufficiency in adults are a common phenomenon in children.6,,10

What is already known on this topic

  • Vitamin D deficiency is common in adults and has been associated with a number of non-bone outcomes, including cardiovascular disease.

  • In children, optimal vitamin D status is not well define d and currently is based upon prevention of rickets.

  • Proximal, functional, non-bone outcomes are needed in order to define optimal vitamin D status in children.

What this study adds

  • This study confirms that obesity is a risk factor for vitamin D deficiency in children.

  • This study suggests vitamin D deficiency/insufficiency is associated with worse insulin resistance in children even after adjustment for obesity and puberty.

  • This study suggests insulin sensitivity may be one outcome by which to begin to define optimal vitamin D status.

Further highlighting the need to address vitamin D requirements and optimal vitamin D status in children are adult studies associating low vitamin D with increased risk of cancers,11 12 autoimmune diseases13 14 and infections.15 Additionally, the role of vitamin D in type 2 diabetes has recently been reviewed.16 Vitamin D promotes insulin secretion in adults.17 Low vitamin D has been associated with insulin resistance,18 19 and in some studies vitamin D supplementation has been associated with improved insulin secretion and glucose tolerance.20

These associations have not been fully explored in children. Two studies failed to confirm an association between low vitamin D status and insulin resistance in obese children,21 22 but had several limitations. The first did not adjust for the impact of puberty and BMI on insulin sensitivity21; the other did not achieve a substantial increase in serum 25-OH-D with a mean serum 25-OH-D after weight loss of 40 nmol/l.22 More recently, an association of low vitamin D with increased risk for fasting hyperglycaemia and metabolic syndrome (defined as fasting hyperglycaemia, increased waist circumference, low high-density lipoprotein, increased triglycerides, increased blood pressure) was found in adolescents in the 2001–2004 National Health and Nutrition Examination Survey (NHANES).23

Given the lack of consensus regarding the definition of optimal vitamin D status in children, identifying functional outcomes by which to define vitamin D status is important. One such outcome is insulin sensitivity. Thus, this study examined the prevalence of vitamin D deficiency and the association of vitamin D and insulin resistance in a population consisting of normal weight and obese children based upon the hypothesis that vitamin D status is associated with worse insulin resistance even after adjustment for important mediators of insulin resistance: overweight/obesity and puberty.

Patients and methods

Study group

Obese and non-obese children aged 4–18 years were recruited (1) from the primary care practices affiliated with The Children's Hospital of Philadelphia, (2) through regional newspaper advertisement and (3) the Sleep Center at The Children's Hospital of Philadelphia. With the exception of children with mild asthma, children with significant, chronic medical conditions such as genetic syndromes, neurological disease or diabetes were excluded. Children receiving medications that could affect metabolic functions such as oral glucocorticoids were also excluded.

The protocol was approved by the Institutional Review Board at The Children's Hospital of Philadelphia. Informed consent was obtained from young adult participants aged 18 years and from the parents/guardians of participants <18 years of age. Assent was obtained from participants 7–17 years of age.

Anthropometry and pubertal development

Weight was measured to the nearest 0.1 kg using a digital scale (Scale-Tronix, White Plains, New York, USA). Height was measured to the nearest 0.1 cm using a stadiometer (Holtain, Crymych, UK). Age- and gender-specific standard deviation scores (Z scores) for body mass index (BMI-Z) were calculated using current reference data for the USA.24

Pubertal status was ascertained using a validated self-assessment questionnaire25 26 to categorise Tanner stages of pubic hair distribution and genital development for boys and breast development for girls.27 Subjects were categorised as either (1) prepubertal, defined as Tanner stage 1, or (2) pubertal, defined as Tanner stage 2 or greater. In females, breast stage was used if there was a discrepancy between breast and pubic hair staging. In males, pubic hair was used if there was a discrepancy between genitalia and pubic hair staging.

Metabolic studies

After an overnight fast, blood was drawn for fasting glucose, insulin and 25-OH-D. Blood glucose was measured using the Rapidpoint blood gas analyser (Siemens Healthcare Diagnostics, Deerfield, Illinois, USA). Insulin and 25-OH-D samples were batched. Insulin assays were completed in The Children's Hospital of Philadelphia Clinical and Translational Research Center Biochemistry Core Laboratory using an ELISA diagnostic kit (catalogue no. 08-10-1113-99; ALPCO, Salem, New Hampshire). Homeostasis model assessment (HOMA), a measure of insulin sensitivity, was calculated as (fasting blood glucose (mmol/l)×insulin (μU/ml))/22.5.28

The entire 25-OH-D assay was performed in The Children's Hospital of Philadelphia Clinical Laboratory using high performance liquid chromatography coupled with tandem mass spectrometry (HPLC-MS/MS) based on the procedure of Maunsell et al29 with modifications. Instrumental analysis was performed on an Applied Biosystems API 4000 LC-MS/MS instrument equipped with a Perkin-Elmer Series 200 autosampler (Perkin-Elmer, Waltham, Massachusetts, USA) and two Perkin-Elmer Micro LC pumps controlled from a computer system using Analyst software (Applied Biosystems, Darmstadt, Germany). The instrument was operated with a TurboIonSpray interface in the positive ion mode. 25-OH-D2, 25-OH-D3 and [2H6]25-OH-D3 were monitored in the multiple reaction mode using the following transitions: 401>383 for 25-OH-D3, 413>395 for 25-OH-D2 and 407>389 for [2H6]25-OH-D3. Quantification was performed using Analyst software and the following integrated peak area ratios: 25-OH-D2/[2H6]25-OH-D3 and 25-OH-D3/[2H6]25-OH-D3. The assay gave a linear response from 3.3 to 338 nmol/l for both 25-OH-vitamin D2 and 25-OH-vitamin D3. The limit of quantitation (defined as the concentration of analyte which gave a signal to noise ratio ≥10) was 3.2 nmol/l for both compounds. The interassay variation was measured for both compounds by measuring the metabolite concentrations of three spiked serum specimens on each of 38 different days. The coefficients of variation (CVs) for 25-OH-D2 were 10.0%, 9.0% and 7.3% at 45, 87 and 250 nmol/l, respectively, and for 25-OH-D3 were 4.2%, 4.9% and 4.8% at 52, 107 and 150 nmol/l, respectively.


For the purposes of analyses, we defined seasons by months: winter (December–February), spring (March–May), summer (June–August) and fall (September–November). Study subjects were categorised into four vitamin D status groups used in a previously published paediatric study10: deficiency <25 nmol/l; insufficiency 25 to <50 nmol/l; intermediate 50 to <75 nmol/l; and sufficiency ≥75 nmol/l. These cut-offs are arbitrary; the intermediate category reflects a group of children with less than optimal 25-OH-D by some adult criteria,30 but who are in a range considered acceptable by paediatric standards,5 despite the identification of elevated parathyroid hormone in children with 25-OH-D in this range.9 10 These cut-offs also ensured that adequate numbers of study subjects were in each category for the purposes of statistical analysis and to allow exploration of threshold effects.

Statistical analyses

Means, standard deviations and ranges were used to summarise continuous variables that were normally distributed. Medians, minima and maxima or the IQR were used to summarise continuous variables that were not normally distributed. The Shapiro–Wilk W test was used to test for normality; p<0.05 was considered evidence for non-normality. To compare continuous measures among more than two groups, one-way analysis of variance with Bonferroni correction was used; for skewed data, these results were corroborated with the Kruskal–Wallis test. To compare proportions among groups, the χ2 tests for heterogeneity and trend and logistic regression were used.

The primary outcomes were 25-OH-D and HOMA. A multivariable approach was used to assess effects expected to be associated with 25-OH-D concentrations in children: age, sex, race, BMI-Z and season (winter vs not winter). A similar approach was used to assess the effects of 25-OH-D status on glucose, insulin and HOMA, adjusting for factors known to affect these metabolic outcomes: sex, race, puberty and BMI-Z. Analyses of HOMA were also performed using 25-OH-D as a continuous variable. These separate approaches were used to help determine whether a threshold effect of 25-OH-D upon HOMA was apparent since ‘normal’ 25-OH-D is currently defined by calcium/bone related outcomes5 and not non-traditional outcomes.

To improve fit, non-normally distributed data were transformed using the STATA software command ‘ladder’ which searches a subset of the ladder of powers for a transform that converts the variable of interest into a normally distributed variable. The normality of the transformed variables was tested using tests for skewness and kurtosis; transformed variables with χ2>0.05 were chosen. The fit of each model was assessed through the adjusted R2 value. The regression models were assessed further through graphical checks, the Shapiro–Wilk test of normality of the residuals, and the Cook–Weisberg test for heteroscedasticity (test for unequal variance of dependent variable across data) using Stata 9 (StataCorp, College Station, Texas, USA). For the final analysis of HOMA and vitamin D, bootstrap analysis31 with 999 repetitions was performed because the assumption of equal variance was not met. Bias-corrected and accelerated CIs were reported. Two-sided tests of hypotheses were used, and p<0.05 was considered statistically significant.


Study group

A total of 92 children were recruited; 32 were tested in winter, 29 in spring, 28 in summer and 9 in fall. HOMA was available in 85 of these 92 children. Data were missing completely at random because of inadequate blood sample, failure to fast (affecting glucose and insulin) or haemolysis (lowering insulin).

As shown in table 1, the mean age of the final cohort (38F/47M) was 11±4 years (mean±1 SD). The group consisted primarily of non-Hispanic white (n=41) and African–American (n=38) children. Thirty-four children were prepubertal. Nine per cent of participants were overweight, defined as BMI-Z between 85% and 95% and another 57% were obese, defined as BMI-Z>95%.

Table 1

Characteristics of study group with homeostasis model assessment (n=85)

Vitamin D status

In general, 25-OH-D was in the intermediate range, median 52 nmol/l (IQR=42) (table 1). Only 26% of the children were vitamin D sufficient (25-OH-D ≥75 nmol/l); 27% had intermediate values, 32% were insufficient and 15% were vitamin D deficient (table 2). The risk of vitamin D deficiency/insufficiency was more common in winter (21/32, p=0.005) and spring (16/29, p=0.045) versus summer (8/28); vitamin D deficiency/insufficiency in fall (3/9) was not significantly different from summer (p=0.79). Vitamin D insufficiency and vitamin D deficiency were associated with older age and higher BMI-Z. Vitamin D deficiency and insufficiency were common with African–American race (p=0.001). Sex was not associated with vitamin D status (p=0.12). Similar findings were observed when 25-OH-D was treated as a continuous variable in linear regression models (table 3).

Table 2

Differences in baseline characteristics and fasting biochemical data* by vitamin D status

Table 3

Results of multivariable linear regression for associations of subject characteristics with serum 25-OH-D concentration* (n=92)

Vitamin D and metabolic outcomes

Both vitamin D deficiency and insufficiency were associated with higher fasting blood glucose even after adjustment for BMI-Z. Similarly, both vitamin D deficiency and insufficiency were associated with higher fasting insulin and higher HOMA after adjustment for BMI-Z and puberty. No associations between sex and HOMA (p=0.238), race and HOMA (p=0.181), or winter season and HOMA (p=0.637) were found following adjustment for puberty and BMI-Z. Age was not a predictor when puberty, a known modulator of insulin sensitivity, was included the model. No interactions between vitamin D status and BMI-Z, puberty or race were observed. Similar associations were found when 25-OH-D concentration was treated as a continuous variable (table 4).

Table 4

Results of multivariable linear regression for association of serum 25-OH-D concentration with HOMA*,† after adjustment for other covariates (n=85)


Recapitulating recent literature,6,,10 this study identified suboptimal vitamin D status in over 70% of paediatric subjects using adult criteria; 16% had frank vitamin D deficiency. Risk factors for lower vitamin D included African–American race, higher BMI-Z, increasing age and winter season. Highlighting potential non-traditional roles of vitamin D, lower vitamin D was associated with higher fasting blood glucose, consistent with recent associations in a large cohort of adolescents.23 In addition, low vitamin D was associated with higher fasting insulin and worsening insulin resistance, even after adjustment for BMI-Z and puberty, important determinants of insulin sensitivity.

Although these findings were based upon cross-sectional data, they suggest that vitamin D supplementation might have a beneficial effect in children with vitamin D insufficiency since (1) obesity is associated with vitamin D deficiency and (2) the number of obese, insulin resistant children with metabolic syndrome and frank type 2 diabetes is increasing at alarming rates.32 33 However, intervention studies are needed to assess the effect of vitamin D repletion in obese children who are vitamin D insufficient.

We found that HOMA was approximately 0.5 less in vitamin D sufficient versus vitamin D deficient children. To provide perspective on the potential impact of improved vitamin D status upon insulin sensitivity, one might consider the results of the Diabetes Prevention Program, a 1-year intervention study of over 3000 overweight/obese adults at high risk for developing diabetes. Median HOMA improved by approximately 0.9 units in the metformin intervention arm and by approximately 1.7 units in the lifestyle intervention arm; both treatment arms were associated with decreased diabetes risk.34 The possibility that correction of hypovitaminosis D may be an additive strategy in type 2 diabetes prevention is attractive.

Using NHANES adult data, Scragg et al found decreasing insulin sensitivity with decreasing 25-OH-D concentrations in Mexican–American subjects and a trend towards this same relationship in non-Hispanic white subjects (p=0.058) but not in non-Hispanic black subjects.35 In a UK population-based study, 25-OH-D concentrations were inversely associated with HbA1C even after adjustment for BMI; this association was magnified in more obese cohorts.36 In contrast, 25-OH-D was not associated with insulin sensitivity in two obese paediatric populations.37 One reason for the disparity between our results and these paediatric studies may rely on our inclusion of normal weight children, although even with restriction to BMI-Z ≥95%, the relationship between vitamin D and HOMA persisted in our study (data not shown).

As the non-classic roles of vitamin D gain wider appreciation, a direct role in insulin sensitisation is worth considering; after all, the vitamin D receptor and 1α-hydroxylase (the enzyme responsible for vitamin D activation) are widely expressed in human tissue including muscle and adipose tissue. For instance, vitamin D has been linked to muscle integrity.38 Moreover, vitamin D supplementation in healthy children has been shown to increase lean body mass,39 and serum 25-OH-D has been associated with muscle power and force in adolescent girls.40 Thus, low 25-OH-D could compromise muscle mass and function, contributing to decreased physical activity and impaired peripheral glucose disposal, potential venues for decreased insulin sensitisation.

A number of study limitations are worth mentioning. The study population was derived from the local community and from a population with suspected sleep disorder, a group that may not represent the general population. It may, however, be representative of the obese population given the prevalence of snoring and obstructive sleep apnoea in obese individuals.41 As with other association studies, additional limitations involve unmeasured confounders, specifically for vitamin D status and obesity status. Multivitamin use, vitamin D supplementation and calcium intake were not documented, and 25-OH-D could be lower in obese children because of lower intake. The intakes of our study population are not expected to be significantly greater than in a population of similar age, sex and racial distribution derived from the same geographical area and in whom average vitamin D intake was approximately 200 IU daily and calcium intake was in the 750–800 mg/day range,10 although lower vitamin D intake was recently reported in a small group of obese children (n=21) versus non-obese children (n=20) (218 vs 339 IU daily).42 In adults, sequestration in adipose tissue has been suggested as the mechanism for lower 25-OH-D in obesity.43 With respect to insulin sensitivity and obesity, 25-OH-D concentrations may be a function of sunshine exposure and, hence, only be a surrogate for physical activity, a factor known to improve insulin sensitivity. Leisure activity was examined in NHANES; the relationship between vitamin D and insulin sensitivity persisted.35 It is possible that the differences in sensitivity were due to greater obesity or an inherent increased insulin resistance among the African–American population and that 25-OH-D is just a surrogate of African–American race. Our analyses did not suggest race was the underlying mechanism for the differences. BMI-Z was used as a surrogate of fat mass, but BMI-Z is a measure of excess weight, not necessarily excess fat. Nonetheless, BMI-Z has been found to correlate well with direct measures of body fat and to be accurate at classifying children who were overfat.44 Finally, while no significant interactions between season and race, age or BMI-Z were found in modelling the association between these variables and 25-OH-D, this lack of significance may be a reflection of sample size and may not accurately depict these relationships.


Vitamin D deficiency is prevalent among children and adolescents, but the long-term burden of vitamin D deficiency may not be limited to bone health. Based upon the current results, vitamin D status is an independent predictor of insulin resistance after adjustment for puberty and BMI-Z, important mediators of insulin resistance. Given the obesity epidemic among children and the recognition that many adult diseases have their origins in childhood, the possibility that low vitamin D promotes insulin resistance holds promise for a simple intervention by which to decrease the risk of type 2 diabetes.



  • Funding This work was supported by K-23-RR021973 (AK) and Institutional Clinical and Translational Science Award Research Grant (UL1-RR-024134) from the National Center for Research Resources.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of The Children's Hospital of Philadelphia.

  • Provenance and peer review Not commissioned; externally peer reviewed.