Introduction

Insulin resistance is one of the most common metabolic alterations related to obesity.1,2 It represents a key element of the metabolic syndrome and an important link between obesity and other metabolic as well as cardiovascular complications.3,4 Children with insulin resistance are also at risk for type 2 diabetes5 and a high proportion of cases apparently remain undiagnosed initially.6

Although reference data for body mass index (BMI) were published by organisations like International Obesity Taskforce, World Health Organization, Centers for Disease Control and Prevention and many others and these are widely used in paediatrics, there is still a lack of adequate reference data for markers of insulin resistance.7 Fasting insulin and homeostasis model assessment to quantify insulin resistance (HOMA-IR) have been suggested among others as surrogate markers for screening purposes in adults,8 as the gold standard method to measure insulin sensitivity (the hyperinsulinaemic euglycaemic clamp) is very labour- and time-intensive and thus not feasible in epidemiological research. Oral glucose tolerance testing, a procedure widely used to measure insulin response in clinical practice, is also not feasible in a setting-based field study like ours. Matthews et al.9 showed that estimates of insulin resistance from HOMA-IR correlated well with estimates from the clamp-technique. Glycosylated haemoglobin (HbA1c) is a standard marker for glycaemic control in diabetic patients but has also been proposed as a predictive marker of insulin resistance.8

Population-based data on insulin resistance in children are rare, especially for pre-pubertal children and from large-scale epidemiological studies. Several authors have shown data on the distribution of insulin, glucose and HOMA-IR values in paediatric populations, and some also suggested cut-off values for insulin and HOMA-IR.10, 11, 12, 13, 14, 15, 16, 17 However, all of these studies were limited to national study populations and sample sizes were mostly too small for statistical modelling of reference values. None of the studies included >1000 pre-pubertal children, except for a Mexican cross-section with about 2500 children in the age stratum of 6–10 years.13 All authors describe a pronounced age dependency of insulin and HOMA-IR, and to a lesser extent also of glucose values. Studies in adolescents show a peak of insulin and HOMA-IR values in puberty and again a slight decline towards adulthood.10,13,18 Comparability of results is limited due to the multitude of different laboratory procedures (for example, enzyme-linked immunosorbent assay, radioimmunoassay, immunoenzymometric assay, immunoelectrochemiluminometric assay) used for the determination of insulin. These different assays can show up to a twofold variation in insulin concentrations.19

The aim of this study is to present age- and sex-specific reference values for insulin, glucose and HOMA-IR based on a European population of normal weight pre-pubertal children from eight European countries who participated in at least one wave of the IDEFICS (‘identification and prevention of dietary- and lifestyle-induced health effects in children and infants’) surveys (2007–2010) and received standardised examinations and laboratory measurements.

Materials and Methods

Study population

A population-based prospective cohort study was one of the key elements of the IDEFICS project. All children of the defined age group who lived in the selected study regions and attended one of the participating pre- or primary schools were eligible for participation. Children and parents were approached via schools and preschools to ensure inclusion of all social groups. Written consent of parents and verbal assent of children were given separately to the different modules of the examination, that is, participants were free to refrain from single components like blood drawing.

The baseline survey in the school year 2007/2008 included 16 228 preschool and primary school children aged 2–9 years from eight European countries (Belgium, Cyprus, Estonia, Germany, Hungary, Italy, Spain and Sweden). Of these, 11 292 were followed up again after 2 years. In the follow-up survey (2009–2010), 2517 new children were additionally included.

Both waves of surveys comprised anthropometrical measurements and examinations of children as well as parental self-completion questionnaires on lifestyle habits and dietary intakes of children. The physical examination programme during the IDEFICS surveys covered standard anthropometric measures, that is, height, weight and circumferences of waist, hip, upper arm and neck, as well as skinfold thicknesses, and the measurement of blood pressure and pulse rate. BMI was calculated as weight (in kg) divided by height squared (in m) and classified according to the International Obesity Taskforce criteria.20

Biomarkers were analysed in blood, urine and saliva samples. Standardised procedures were used by all survey centres and a quality management system was established.21 Venous blood was collected after an overnight fast from 9185 of the baseline and 962 of the children newly recruited at follow-up. In addition, 1011 children who did not provide blood at baseline survey, provided blood at follow-up. The present analyses included children of both waves of surveys with available data for height, weight and the laboratory analyses of interest. If serum measurements of a child were available for both survey waves, preference was given to the baseline survey. Girls and boys below 3 and above 10.9 years of age were excluded from the study sample due to the small number of children in these age strata (n<100) to avoid instability of the statistical model. Children were also excluded if they had a diagnosis of diabetes or if they reported to be non-fasting at the time of blood withdrawal. The present analysis is thus based on 7074 children, as depicted in Figure 1. The background of the IDEFICS study, its research goals and instruments have been described elsewhere in detail.22

Figure 1
figure 1

Flow chart of children included in the analysis.

Full size image

Laboratory analyses

Children participating in the IDEFICS baseline survey were asked to provide fasting venous blood, morning urine and saliva samples. If consent was not given for venous blood withdrawal, capillary blood was taken with the consent of the parents and the children. Blood glucose was assessed on site at each study centre by point-of-care analysis using the Cholestech LDX analyser (Cholestech, Cholestech Corp., Hayward, CA, USA) either in venous or capillary blood. Precision and accuracy of this analyser were comparable to clinical diagnostic laboratory methods23 with a slight positive bias of glucose measurement.24 Pre-analytical sample processing of blood samples was done at the local survey centres or at local laboratories. Samples were then frozen (at −80 °C) and shipped to a central laboratory, certified by ISO 15189:2007, for later analysis of insulin, HbA1c and other biological markers. Insulin was analysed by an electrochemiluminescence immunoassay (Roche Modular System, Mannheim, Germany), HbA1c was analysed by high-performance liquid chromatography (AUTO-GA variant). Details on the biological sampling procedures can be obtained from an earlier publication.25 HOMA-IR was calculated as fasting insulin (μIU ml−1) × fasting glucose (mmol l−1)/22.5.

Statistical analyses

Percentile curves of insulin, glucose and HOMA-IR were calculated as a function of the covariate age stratified by sex using general additive model for location scale and shape (GAMLSS) method as an extension of the LMS method.26 The LMS method models three parameters: the skewness L, the median M and the coefficient of variation S. The skewness accounts for the deviation from a normal distribution using a Box–Cox transformation, the median of the outcome variable is modelled depending on one explanatory variable and the coefficient of variation accounts for the variation of data points around the mean and adjusts for non-uniform dispersion. The GAMLSS method is able to model more than one covariate and also other distributions that particularly include the kurtosis. We used the gamlss package (version 4.2–6) of the statistical software R (version 3.0.1).27 Different distributions were fitted to the observed distribution of insulin, glucose and HOMA-IR. Moreover, the influence of age on parameters of the considered distributions were modelled either as a constant, as a linear function, or as a cubic spline of the covariates. Goodness-of-fit was assessed by the Bayesian information criterion and by Q–Q plots to select the final model including the fitted distribution of insulin, glucose and HOMA-IR and the influence of covariates on distribution parameters. Worm plots were used as a diagnostic tool to assess whether adjustment for kurtosis was required.28 Finally, percentile curves for the 5th, 10th, 25th, 50th, 75th, 90th and 95th were calculated based on the model that showed the best fit.27,29

The final models for insulin and HOMA-IR for boys and girls considered a Box–Cox t (BCT) distribution modelling μ as a cubic spline depending on age, log(σ) as a linear function of age, and ν and τ as constants. With regard to glucose, a lognormal distribution was used for boys and girls considering μ as a linear function of age and log(σ) as a constant.

Results

The reference population is composed of 3434 girls and 3640 boys. One-year age groups (3.0–10.9 years of age) include from about 250 up to 1500 children. Characteristics of this population are presented as mean values (±s.d.) or median (25th, 75th percentile) by age group in Table 1 for the study population before and after restriction to normal weight children. Anthropometrical measures and glucose concentrations show only little differences between boys and girls, but insulin concentrations are higher in girls than in boys through all age groups, which is also reflected in slightly higher HOMA-IR values.

Table 1 Characteristics of study population by one-year age group, before and after restriction to normal weight children
Full size table

Percentiles of insulin, glucose and HOMA-IR are presented in half-year age groups in Table 2 and Figures 2, 3, 4, respectively. Concentrations of all three biomarkers clearly show a positive trend with age, which is also observed for the variance of insulin and HOMA-IR, whereas for glucose the variance is similar in all age groups. Insulin and HOMA-IR values are higher in girls than in boys for all age groups, whereas glucose values are slightly higher in boys.

Table 2 Percentiles of insulin (pmol l−1), glucose (mmol l−1) and HOMA-IR in normal weight children calculated with GAMLSS
Full size table
Figure 2
figure 2

Percentiles of fasting serum insulin (measured by an electrochemiluminescence immunoassay) from normal weight children of the IDEFICS cohort (2007–2010).

Full size image
Figure 3
figure 3

Percentiles of fasting blood glucose (measured by point-of-care analysis) from normal weight children of the IDEFICS cohort (2007–2010).

Full size image
Figure 4
figure 4

Percentiles of HOMA-IR from normal weight children of the IDEFICS cohort (2007–2010). Insulin was measured by an electrochemiluminescence immunoassay, glucose was measured by point-of-care analysis.

Full size image

For insulin, 5th and 95th percentiles range from 4.2–49.3 and 3.5–41.0 pmol l−1 in 3–<3.5-year-old girls and boys, respectively, to 25.7–100.7 and 19.4–88.2 pmol l−1 in 10.5–<11-year-old girls and boys. For glucose, 5th and 95th percentiles were 3.6–5.2 and 3.7–5.3 mmol l−1 in 3–<3.5-year-old girls and boys up to 4.2–5.7 and 4.3–5.9 mmol l−1 in 10.5–<11-year-old girls and boys. For HOMA-IR, 5th and 95th percentiles ranged from 0.1–1.5 and 0.1–1.3 in 3–<3.5-year-old girls and boys to 0.8–3.4 and 0.6–3.0 in 10.5–<11-year-old girls and boys.

A table showing percentiles for children of all weight groups is provided in Supplementary Table A. Percentile values for glucose show only negligible differences as compared with the values for normal weight children. Inclusion of overweight and obese children has little impact on lower percentiles of insulin and HOMA-IR but has a very pronounced effect on the upper percentiles, especially in the older age groups. The 95th percentile of insulin in 10.5–<11-year-old girls for instance, is 125.7 compared with 100.7 pmol l−1 in normal weight girls of the same age. A sensitivity analysis was conducted to illustrate the influence of BMI on age-specific values of insulin, glucose and HOMA-IR. A strong dependency on weight status was observed for insulin and HOMA-IR but not for glucose. Insulin percentile curves for children of all weight groups versus only normal weight girls and boys are depicted in Figure 5. Results of the sensitivity analyses for glucose and HOMA-IR are provided in Supplementary Figures A and B. Three different reference populations for BMI were used to test the robustness of the model.20,30, 31, 32 Preference was given to the classification by Cole et al.,20 as differences between the three reference systems were not substantial and this classification led to the smallest number of excluded children.

Figure 5
figure 5

Sensitivity analysis showing percentiles of fasting serum insulin according to different definitions of normal weight and for the whole study group including children of all weight groups (classification of normal weight according to Cole et al./International Obesity Taskforce,20 Centers for Disease Control and Prevention30 and World Health Organization31,32).

Full size image

Percentile curves for HbA1c are provided in Supplementary Figure C. Serum concentrations show nearly no variation for younger children (median of 26.8 for girls and 27.9 mmol mol−1 for boys at 3 years of age to 29.0 mmol mol−1 for both sexes at 8 years of age) and a slight increase for the 8–<11-year-old children (median of 34.4 mmol mol−1 for girls and 33.3 mmol mol−1 for boys at 10.9 years of age).

Discussion

This study presents age- and sex-specific reference percentiles of insulin, glucose and HOMA-IR for pre-pubertal children on the basis of the IDEFICS cohort of children from eight European countries. This is the first time, reference percentiles of these biomarkers are based on a multinational children’s cohort. At the same time, with over 7000 subjects, it is by far the biggest cohort ever used to generate such reference data. Nevertheless, previous studies of local populations have shown similar distributions of insulin, glucose and HOMA-IR values in children. As the age group of 9-year-olds was considered most frequently, this age group will be used for comparison in the following.

In our reference population the 5th and 95th percentile values of insulin are 15.3–81.3 pmol l−1 for 8.5–<9.5-year-old girls and 12.5–74.3 pmol l−1 for 8.5–<9.5-year-old boys. Insulin values were also measured in a representative sample of 2244 school children from Quebec who were 9, 13, and 16 years old in 1999 (ref. 11) (insulin measured by the ultrasensitive insulin assay on the Access immunoassay system by Beckman Coulter, Mississauga, ON, Canada), in a representative sample of 1976 healthy French individuals aged 7–20 years in 2006–2008 (ref. 12) (insulin measured with a microparticle enzymoimmuno assay on an AxSYM analyser by Abbott, Abbott Park, IL, USA) and in a random sample of 1137 healthy 9- and 15-year-old school children examined in a cross-sectional study in Sweden in 1998/99 (ref. 17) (insulin assay not reported). In the pre-pubertal group of 9-year-olds, the 5th and 95th percentiles of insulin were 11.8–59.0,11 13.9–83.3 (ref. 12) and 11.1–78.5 pmol l−1 (ref. 17) in girls and 10.4–54.2,11 10.4–69.5 (ref. 12) and 10.4–64.6 pmol l−1 in boys. Overall, insulin concentrations were showing a positive trend with age until puberty and revealed pubertal peaks, which were sharper in females than in males. These results are in good accordance with the percentile values from our study. Studies in mixed non-Caucasian populations on 6132 school children aged 6–18 years in Mexico13 (insulin measured with a microparticle enzymoimmuno assay on an AxSYM analyser by Abbott) and 2153 normal weight children and adolescents in Chile16 (insulin measured by chemiluminescence on the ADVIA Centaur CP Immunoassay System, Bayer HealthCare AG, Leverkusen, Germany) also showed a gradual increase of insulin values between 6 and 13 years of life. Possibly due to the fact that older children were included, insulin concentrations were higher in the Chilean population than in our study cohort: the 5th and 95th percentiles were 29.9–150.0 pmol l−1 for girls and 26.4–117.4 pmol l−1 for boys in Tanner stages I and II.16

For glucose, the 5th and 95th reference percentiles in our population are 4.0–5.6 mmol l−1 for 8.5–<9.5-year-old girls and 4.1–5.8 mmol l−1 for 8.5–<9.5-year-old boys. Remarkably, glucose curves showed a linear trend in this pre-pubertal age group. To verify linearity of this functional relationship, we also applied a cubic model to the data. As both curves looked very similar, we preferred to use the simpler model. In addition, we know from literature12 that glucose curves only start to flatten during puberty, which may lead to a linear functional relationship for the age range that we have considered here. Glucose values were very similar in the three above-mentioned studies from Canada,11 France12 and Sweden.17 For 9-year-olds, the 5th and 95th percentiles were 4.4–5.6,11 3.9–5.2 (ref. 12) and 4.3–5.4 mmol l−1 (ref. 17) in girls and 4.7–5.7,11 4.0–5.2 (ref. 12) and 4.4–5.5 mmol l−1 (ref. 17) in boys. Percentile values of glucose showed a positive trend with age, although the trend was less pronounced as compared with insulin. A distinct pubertal peak of serum glucose concentrations was only observed in females.

For HOMA-IR, the 5th and 95th percentiles in our reference population are 0.4–2.7 for 8.5–<9.5-year-old girls and 0.4–2.5 for boys of this age. Estimates from the 1999 Quebec sample11 (insulin measured by the ultrasensitive insulin assay on the Access immunoassay system by Beckman Coulter) showed slightly lower HOMA-IR values (5th and 95th percentile) of 0.3–2.1 and 0.3–1.9 for 9-year-old girls and boys, respectively. A small Italian cross-section of healthy children and adolescents reported values of 0.6–2.2 for pre-pubertal girls and 0.4–2.2 for the respective boys15 (insulin measured by radioimmunoassay, Radim Kit, Rome, Italy). The above-mentioned studies in Mexico13 (insulin measured with a microparticle enzymoimmuno assay on an AxSYM analyser by Abbott) and Chile16 (insulin measured by chemiluminescence on the ADVIA Centaur CP Immunoassay System, Bayer HealthCare AG) also showed an increase of HOMA-IR from the age of 6 to the age 13 years. Similar to insulin, percentile values were obviously higher for the combined group of Tanner stages I and II, as compared with our study population, that is, HOMA-IR values were 0.9–4.9 for girls and 0.8–3.9 for boys. A HOMA-IR value of 2.0 was suggested as a cut-off point for pre-pubertal children in a population-based small Italian study14 (insulin measured by an immunoenzymometric assay, AIA-Pack IRI, Tosoh, Tokyo, Japan). Cut-off values were also derived from oral glucose tolerance testing in studies with obese children. Two Turkish studies on 148 and 82 obese paediatric patients, proposed cut-off points of HOMA-IR ⩾2.7 for both sexes33 (insulin was measured using the IMMULITE immunoassay, IMMULITE Diagnostic Products Corporation, Los Angeles, CA, USA) or HOMA-IR ⩾2.2 for girls and ⩾2.7 for boys34 (insulin measured by an immunoradiometric assay kit, INS-Irma Biosource, Nivelles, Belgium).

HbA1c as a standard marker for glycaemic control in diabetes has rather little relevance in healthy children. However, as it is frequently measured in population-based studies, we included it in our analysis and we present the results in Supplementary Figure C. To our knowledge, there is only one recent study that aimed to define paediatric reference values for HbA1c.35 This Swedish cohort investigated healthy children from 6 months to 18 years of age and suggested cut-off values of 16.9–27.9 mmol mol−1 (2.5th and 97.5th percentile), which are lower than the 5th and 95th percentiles observed in our study population, that is, 15.8–34.4 and 16.9–34.4 mmol mol−1 for 3-year-old girls and boys, and up to 26.8–38.8 and 15.8–39.9 mmol mol−1 for 10.9-year-old girls and boys.

Overall, the distributions of insulin, glucose and HOMA-IR shown in previous studies are in good accordance with the reference percentiles in the present analysis. For the subgroup of 9-year-old girls and boys, the 5th and 95th percentile values are very similar to cohorts from European countries or Canada. Concentrations were generally higher in countries with mixed populations including people with indigenous ancestry or when adolescents were included (for example, combined analysis for Tanner stages I and II), which corresponds well with the pubertal peak described for all of the considered markers. Differences in insulin and HOMA-IR levels could also be due to the variety of different insulin assays used in the referenced studies.

In our study, only normal weight children were included into the modelling of percentile curves as the insulin level is known to be strongly influenced by BMI.36 Despite the strong influence of BMI on markers of insulin resistance, the relationship between BMI and HOMA-IR has been shown to be much weaker in children than in adults.37 Within the growing group of children with overweight and obesity, paediatricians are faced with the challenge of identifying individuals at the greatest risk of comorbidity. It has been pointed out by other authors that interventions to halt weight gain and promote weight loss in children are of limited success and demand significant resources and continuous follow-up and monitoring.38 This puts the caregiver in the dilemma of where to allocate the limited available resources and who among the obese children will benefit most from an intervention.1 The provision of paediatric reference curves for insulin resistance is thus long overdue.

The main strengths of the present study are the large sample size and the standardised assessment of anthropometrical and laboratory measurements. The statistical modelling was done according to the most advanced methodology, which has also been used by the most recent studies in this field.29,39,40 Some limitations of the study should also be accounted for: even though the study was designed to reach all eligible children in the selected study regions via their settings, the overall participation rate was just above 50%,22 and a non-response bias, for example, towards higher social status may well be present. Nevertheless, the influence of social status on insulin resistance is likely to be mediated by weight status, which is one of the reasons why we restricted our analyses to normal weight children. Another problem might be caused by the fact that pubertal stages were not assessed in the IDEFICS surveys. Even though children of our age range are considered to be pre-pubertal, we cannot exclude that a few of them might already have developed signs of beginning puberty that might better be excluded from the reference population.

In summary, the reference percentiles presented here are to our knowledge the first to be based on a multinational children’s population that is also the largest cohort that was up to now used to model reference percentiles for markers of insulin resistance (insulin measured by electrochemiluminescence immunoassay). These reference values may help to identify children with insulin resistance who have an elevated risk for cardiovascular disease and type 2 diabetes mellitus. These reference values will thus allow for a more focused and earlier behavioural or therapeutic intervention. As insulin resistance has been shown to be reversible in most cases by healthy eating and physical activity,41,42 these reference values may hopefully contribute to reduce the burden of disease in the long run.