1 INTRODUCTION

Inflammation plays a significant role in the pathogenesis of diabetes.1 In the adult population, increased concentrations of proinflammatory and reduced anti-inflammatory markers were significantly associated with the incidence of type 2 diabetes.2-4 However, these associations need to be confirmed for causality, as Mendelian randomisation studies yielded inconsistent results for some inflammatory markers.5-8 Previously, there have been prospective studies that have investigated associations between inflammatory markers and glycaemia or insulin resistance, measured at one-time point.9, 10 However, few studies have addressed longitudinal associations between inflammatory markers and glycaemic traits11-13 and even fewer studies have investigated on how inflammatory markers act in combination.14 Particularly, longitudinal studies investigating the association between low-grade systemic inflammation and markers of glycaemic deterioration/insulin resistance in children are missing.7, 8

Moreover, since higher HbA1c and HOMA-IR are important indicators of vascular complications in prediabetic conditions15 and have also been closely related to higher risk of cardiovascular disease and all-cause mortality in nondiabetic people,16 better biological markers are required to identify the subjects at high risk in very early phases, such as prediabetes which may open new directions for early prevention. As inflammatory markers may be used to refine diabetes risk prediction and thus better target individuals for lifestyle interventions, we aimed to investigate longitudinal associations between pro- and anti-inflammatory markers (individually and combined) and markers of glycaemia (fasting glucose [FG], HbA1c), and insulin resistance (HOMA-IR) in European children.

2 RESEARCH DESIGN AND METHODS 2.1 Study population

The study population was enroled in the pan-European, multi-centre, prospective IDEFICS/I.Family cohort of 16,229 children aged between 2 and 9.9 years at T0, from eight European countries (Belgium, Cyprus, Estonia, Germany, Hungary, Italy, Spain, and Sweden). The children were first examined in 2007 and 2008 with follow-up examinations conducted after two (T1) and six (T3, I.Family study) years; the design of this cohort study has been described in detail elsewhere.17, 18 In the IDEFICS/I.Family study, risk factors of lifestyle-related outcomes were investigated in young children and anthropometric and clinical examinations were conducted at each survey wave. Blood samples were considered fasting if the last meal or drink (other than water) was consumed >8 h before drawing blood. Before children entered the study, parents provided written informed consent. Additionally, children aged 12 years and older gave simplified written consent. Younger children gave verbal assent for examinations and sample collection. Ethics approval was obtained from the institutional review boards of all eight study centres.

2.2 Markers of glycaemia/insulin resistance

At T0, FG was assessed either with capillary blood from finger prick or with venous blood from venipuncture using a point-of-care analyser (Cholestech LDX, Cholestech Corp.) which reports plasma equivalent glucose concentrations/venous plasma glucose concentrations. In T3, an enzymatic UV test (Cobas c701, Roche Diagnostics GmbH) was used for FG analysis from NaF plasma. At T0, serum insulin concentrations were measured by luminescence immunoassay in a central laboratory. We used an AUTO-GA Immulite 2000, Siemens, Eschborn, Germany. At T3, serum insulin was analysed (at the University of Bremen, Centre for Biomolecular Interactions Bremen) by multiplex analysis with electrochemiluminescence technology from Meso Scale discovery (MSD) using a MULTI-SPOT® Assay System; Human Leptin, Insulin Assay Kit. The HbA1c was analysed in K2-EDTA venous blood by high-performance liquid chromatography (AUTOGA variant, Biorad) in a central laboratory at both T0 and T3. HOMA-IR was calculated as fasting insulin (μIU ml−1) × FG (mg/dl)/405.

2.3 Inflammatory markers

Serum samples stored at −80°C were used to detect levels of C-reactive protein (CRP), interleukin-1 receptor antagonist (IL-1Ra), IL-6, 8, 15, interferon gamma inducible protein (IP-10), TNF-α, adiponectin and leptin were measured at T0 and T3, by ELISA using electrochemiluminescent multiplex assay (using either single or MULTI-SPOT® Assay Systems, Meso Scale Discovery). The choice of inflammatory markers were based on their role in endothelial function via either direct or indirect mechanisms such as reducing nitric oxide production and stimulating inflammation-oxidative stress pathways. IL-6, IL-8, TNF-α, IP-10, IL-15 and IL-1Ra were run together on a 6-plex assay, insulin and leptin run together on a 2-plex assay, whereas adiponectin, and CRP on single-plex assays each. The combination of markers for the assays were decided based on the feasibility of combinations with the help of MSD customer support.

2.4 Covariables

Based on the validated and reproducibility tested FFQ data, a Healthy Diet Adherence Score (HDAS) was developed for all the study regions,19, 20 as a proxy-indicator of children's adherence to healthy dietary guidelines including a high consumption of fruits and vegetables, wholemeal, fish consumption of 2–3 times per week and a reduced intake of refined sugars and fat. The HDAS was used for the present analyses as a continuous variable and ranged from 0 to 50. A higher score represented a higher adherence to healthy dietary guidelines. The pubertal status was self-reported by children in T3, and was defined as pre-pubertal or pubertal based on voice change in boys and age at menarche in girls. This definition of pubertal status has given similar results when compared to Tanner stage in this cohort previously.21 We used the number of occasions reported for alcohol intake/cigarette smoking in lifetime to create binary indicator variables for alcohol intake and smoking of ever smokers/drinkers versus nonsmokers/non-drinkers. The alcohol and smoking questionnaire was completed at T3 by study participants 12 years of age or older at the time of examination. Sports club membership (yes/no) as an indicator of physical activity22 and daily TV, DVD, video, computer or games-console use in hours which were summed to obtain the total screen time for the whole week as a proxy for sedentary behaviour were reported by parents in T0. In T3, these proxy measures were reported by parents if the child was younger than 12 years, or self-reported if the child was 12 years or older. Parents self-reported their history of diabetes which was categorised as positive (at least one parent with diabetes), negative (both parents without diabetes), or unknown (if diabetes status of mother and father were unknown). Parents reported medication use and medical history for their children by means of an interview based on the health and lifestyle questionnaire. Mothers were asked to retrospectively report starting and ending months of exclusive breast feeding and breast feeding combinations which were used to derive the total breast feeding duration.23 Information on mother's height and weight assessed at cohort entry was used for calculating maternal BMI. A binary indicator for children delivered at term versus children born preterm (≤37th gestational week) and continuous variable for birthweight were derived from parental questionnaire data. As part of the standardised anthropometric examination protocol, waist circumference (WC; cm) was measured in an upright position with relaxed abdomen and feet together, midway between the lowest rib margin and the iliac crest to the nearest 0.1 cm (inelastic tape: Seca 200; Seca). Height (cm) of the children was measured to the nearest 0.1 cm with a calibrated stadiometer (Seca 225 stadiometer), body weight (kg) was measured in fasting state in light clothing on a calibrated scale accurate to 0.1 kg (Tanita BC 420 SMA, Tanita Europe GmbH). BMI was calculated as weight (kg) divided by height (m) squared.

2.5 Analysis dataset

The present analysis used only T0 and T3 measurements as inflammatory markers were not measured at T1. Our analysis dataset included participants with measurements of at least one inflammatory marker from T0 or T3 (n = 7992). Children diagnosed with type 1 or type 2 diabetes at cohort entry (n = 9) or taking anti-diabetic drugs (ATC codes: A10), anti-inflammatory drugs (M01), or corticosteroids (H02) within the last 14 days of cohort entry or follow-up examination were excluded from the analysis (n = 560). Children with acute infection defined as CRP level ≥10 mg/l at T0 or T3 were also excluded (n = 886). Finally, for non-fasting blood samples the values of FG and HOMA-IR were set to missing, thus leading to a final study sample of 6537 children (Figure S1).

2.6 Statistical analysis

Data were expressed as mean ± SD or median with an interquartile range as appropriate. According to previously described methods,24-27 age- and sex-specific z-scores were derived for waist-to-height ratio, WC, HbA1c, HOMA-IR, triglycerides, SBP, and FG in children and adolescents using the data collected in the IDEFICS/I.Family cohort. Since the laboratory methods to measure FG changed between T0 and T3, age- and sex-specific reference percentiles were estimated for T0 and T3, separately, and were used to calculate the respective z-scores for the analysis. We used stata module STNDZXAGE for calculating z-scores of inflammatory markers by standardising its raw values (irrespective of their distribution with respect to the detection limits) over age, sex, and survey.28 Since the children were newly recruited in all surveys (i.e. T0, T1 and T3; Figure S1), we henceforth use the word ‘baseline’ for cohort entry and follow-up time for representing difference between age at follow-up and age at cohort entry. The follow-up time was used as a continuous variable, as it was different for different study participants.

To model the association between inflammatory markers and markers of glycaemia/insulin resistance, a two-level growth model was used, where one level accounts for differences between individuals and the other level for changes over time within individuals.29 Markers of inflammation (continuous variable) were the exposure variables and markers of glycaemia/insulin resistance (continuous variable) were the response variables. The between-subject effect estimate referred to the association between an inflammatory marker and HbA1c, HOMA-IR or FG, whereas the fixed-effect interaction between follow-up time and inflammatory markers represented the rate of change in the association between inflammation and glycaemic deterioration per 2-year increase in follow-up time.

The description of the crude model is as follows: let be -th measurement of the -th child (e.g., z-scores of HOMA-IR, HbA1c, FG), is an inflammatory marker (e.g., z-scores of CRP, IL-1Ra, IL-6, 8, 15, IP-10, TNF-α, adiponectin and leptin), is the follow-up time since cohort entry and is the error term for individual i at follow-up time j, then the crude model without adjustment was specified as follows: where is the overall mean intercept, is the overall mean slope and and express how much the intercept and slope, respectively, of individual deviates from the average intercept and slope with respect to the individuals's follow-up time. Crude models included age and sex in addition to one exposure variable and follow-up time. The adjusted models furthermore included the minimum sufficient adjustment set (MSAS) for estimating the association between each inflammatory marker and markers of glycaemia/insulin resistance. The MSAS was identified using directed acyclic graph (DAG)30 built using DAGitty version 3.0 (Figure S2) and included age, sex, study region (proxy for ethnicity), waist-to-height ratio, lifetime smoking and alcohol status, family history of diabetes, membership in a sports club (proxy for physical activity), and screen time per week (proxy for sedentary behaviour), HDAS, pubertal status, birthweight and other inflammatory markers. The DAG was built from literature research and expert knowledge (Figure S2). The main assumptions of growth models were checked and confirmed. Post-hoc analyses were performed to evaluate the marginal effect of each inflammatory marker on FG, HbA1c, HOMA-IR at baseline and different follow-up times using effect estimates from the adjusted model. To investigate the differential effect of sex, we performed a sex-stratified analysis. The combined effect of all inflammatory markers was calculated by subtracting the sum of z-scores of the anti-inflammatory markers (IL-1Ra, IL-15, adiponectin) from the sum of z-scores of the proinflammatory markers (CRP, leptin, TNF-α, IP-10, IL-8 and IL-6). This sum was then dichotomised in high (=1; topmost quartile) and low (=0; lower three quartiles) using the 75th percentile as the cut-off, separately for each sex and survey (T0 and T3). Several sensitivity analysis were performed: (i) study participants with >90th percentile of HbA1c/HOMA-IR/FG at baseline were excluded to further eliminate bias due to undetected prevalent diabetes; (ii) to evaluate the robustness of our DAG MSAS, additional covariates were included in the model; (iii) the sample was limited to children with no overall obesity (BMI category upto 0 by Cole & Lobstein), no abdominal obesity (waist circumference z-scores z-scores 

All covariates were treated as time-varying to account for changes in lifestyle and anthropometric factors over time. The results were reported as regression coefficients and their 95% confidence intervals. Bonferroni correction was used to account for multiple testing, that is the statistical significance level was set to α = 0.05/10 = 0.005 (nine independent inflammatory markers and one dichotomised sum score were tested for FG, HbA1c, HOMA-IR). All statistical tests were two-sided. Statistical analyses were performed using Stata 16 and R 4.0.3.

3 RESULTS 3.1 Characteristics of the study population

Table 1 shows the characteristics of the study participants included in the analysis at T0 and T3. The mean age of participants at T0 was 6.17 years (SD = 1.75) and 48% were girls (Table 1). The mean (±SD) duration of follow-up time was 5.3 years (±1.1). The percentage of overweight or obese children were 15.9 and 25.2 at T0 and T3, respectively. The prevalence of diabetes in parents was low at both T0 and T3. There was an increase in the proportion of children with a sports club membership at T3 as well as an increase in screen time over this period. Moderate correlations were observed among most of the inflammatory markers (Figure S3).

TABLE 1. Characteristics of the analysis group at T0 and T3 IDEFICS/I.Family cohorta Parameters T0 (n = 5794) T3 (n = 4393) Year of examination 2007−08 2013–14 Sex, female: n (%) 2781 (48.00) 2129 (48.46) Age: years 6.17 (±1.75) 11.64 (±1.96) Study region Italy: n (%) 641 (11.06) 634 (14.43) Estonia: n (%) 915 (15.79) 731 (16.64) Cyprus: n (%) 824 (14.22) 1061 (24.15) Belgium: n (%) 777 (13.41) 235 (5.35) Sweden: n (%) 838 (14.46) 436 (9.92) Germany: n (%) 1000 (17.26) 541 (12.32) Hungary: n (%) 703 (12.13) 650 (14.80) Spain: n (%) 96 (1.66) 105 (2.39) BMI category by Cole & Lobstein, 2012 Thinness grade 1–3: n (%) 636 (10.99) 369 (8.41) Normal weight: n (%) 4236 (73.11) 2918 (66.42) Overweight/obese: n (%) 922 (15.91) 1106 (25.18) BMI z-score 0.19 (±2.71) 0.89 (±2.84) WC z-score, n (T0) = 5705, n (T3) = 4308 0.13 (±1.35) 0.43 (±1.48) Wasit-to-height ratio z-score, n (T0) = 5705, n (T3) = 4308 0.07 (±1.32) 0.39 (±1.45) Antibiotic intake: n (%) 249 (4.30) 70 (1.59) HDAS, n (T0) = 5397, n (T3) = 4031 21 (15–27) 22.0 (16–28) SBP z-score, n (T0) = 5368, n (T3) = 4197 0.05 (±1.01) 0.03 (±1.02) DBP z-score, n (T0) = 5369, n (T3) = 4197 −0.01 (±1.01) −0.02 (1.00) TRG z-score, n (T0) = 4883, n (T3) = 3986 0.09 (±0.85) 0.10 (±1.04) HDL z-score, n (T0) = 4887, n (T3) = 3976 0.02 (±0.99) −0.08 (±1.02) Pubertal status: n (%) Not observed 1701 (38.72) Ever smoking: n (%) Not observed 203 (4.62) Ever consumed alcohol: n (%) Not observed 665 (15.14) Parental history of diabetes: n (%) 87 (1.50) 117 (2.66) Membership in sports club: n (%) 2617 (45.17) 2700 (61.46) Screen time per week (hours), n (T0) = 5444, n (T3) = 4001 11.97 (±0.10) 17.19 (±0.17) CRP (ng/ml), n (T0) = 3377, n (T3) = 3890 1006.17 (341.28–2720.94) 323.77 (109.85–1048.80) Adiponectin (μg/ml), n (T0) = 2277, n (T3) = 3135 26.09 (19.76–34.84) 2.05 (14.41–29.56) Leptin (pg/ml), n (T0) = 3154, n (T3) = 3969 1555.21 (923.75–2987.10) 5152.26 (2193–13,322.02) IL-1Ra (pg/ml), n (T0) = 3195, n (T3) = 3561 313.92 (205.87–450.17) 272.94 (202.74–395.72) IL-6 (pg/ml), n (T0) = 3124, n (T3) = 3605 0.27 (0.17−0.45) 0.41 (0.28−0.62) IL-8 (pg/ml), n (T0) = 3194, n (T3) = 3633 3.25 (2.35–4.61) 6.18 (4.46–8.98) IL-15 (pg/ml), n (T0) = 3170, n (T3) = 3633 1.79 (1.27–2.49) 2.26 (1.72–2.92) IP-10 (pg/ml), n (T0) = 3195, n (T3) = 3636 175.28 (131.27–253.39) 213.09 (158.63–296.52) TNF-α (pg/ml), n (T0) = 3195, n (T3) = 3633 2.18 (1.68–2.86) 2.57 (2.01–3.42) HbA1cz-scores, n (T0) = 4225, n (T3) = 3889 −0.26 (±0.94) 0.05 (±0.99) HOMA-IR z-scores, n (T0) = 3515, n (T3) = 3172 0.04 (±1.07) 0.18 (±1.18) FG z-scores, n (T0) = 4744, n (T3) = 3220 −0.01 (±1.03) −0.01 (±0.98) HbA1c (%), n (T0) = 4331, n (T3) = 3926 4.7 (4.40–5.00) 5.0 (4.80–5.20) HbA1c 5.7%–6.4%: n (%) 21 (0.36) 40 (0.91) HOMA-IR, n (T0) = 3596, n (T3) = 3303 0.76 (0.45–1.17) 1.22 (0.77–1.86) HOMA-IR > 2.5: n (%) 104 (1.79) 427 (9.72) FG (mg/dl), n (T0) = 4881, n (T3) = 3354 84 (78.00–90.00) 93 (89.00–98.00) FG 100 mg/dL to 125 mg/dL: n (%) 198 (3.42) 584 (13.29) Note: n stated in case of missingness. Abbreviations: CRP, C-reactive protein; DBP, diastolic blood pressure; FG, fasting glucose; HDAS, Healthy Diet Adherence Score; HDL, high density lipoprotein; IL-1Ra, interleukin-1 receptor antagonist; IP-10, interferon gamma inducible protein; NA, not applicable; SBP, systolic blood pressure; TRG, triglyceride; WC, waist circumference. a Characteristics of the study participants are presented as number (percentages) for categorical variables and median (25th and 75th percentiles) or mean (±SD) for continuous variables. 3.2 Two-level growth models

Results of the two-level growth models for the association between inflammatory markers and HbA1c and HOMA-IR are depicted in Table 2. Figure 1 illustrates effect estimates at baseline and different follow-up time-points from adjusted model based on DAG. The interaction between inflammatory markers and follow-up time represents the rate of change in the association between z-scores of an inflammatory marker and z-scores of HbA1c, HOMA-IR or FG per 2-year increase in follow/up time. The increase in leptin levels was observed to be strongly associated with increased levels of HOMA-IR (Figure 1; Table 2). We also observed weak association between leptin and FG (β = 0.04, 95% CI = 0.004 to 0.09; Table S1). Further, a significant interaction was observed between leptin and follow-up time showing that the association between leptin and HOMA-IR increases by 0.05 units per 2-year increase in follow-up time (Table 2). In post-hoc analysis, children with 6 years of follow-up showed that a one-unit increase in z-score of leptin level showed an 0.38 (95% CI = 0.32 to 0.44) unit increase in HOMA-IR z-scores (Figure 1). Association between higher levels of CRP with increasing concentration of HOMA-IR were observed only in crude model (Table 2).

TABLE 2. Association between markers of inflammation and HbA1c/HOMA-IR HOMA-IR Crudea Adjustedb β (95% CI) β (95% CI) CRP 0.06 (0.03 to 0.09) −0.01 (−0.05 to 0.03) CRP × follow-up time 0.04 (0.02 to 0.06) 0.01 (−0.02 to 0.03) Adiponectin 0.00 (−0.02 to 0.02) 0.00 (−0.03 to 0.02) Adiponectin × follow-up time −0.01 (−0.03 to 0.01) 0.00 (−0.02 to 0.02) Leptin 0.39 (0.36 to 0.41) 0.29 (0.25 to 0.33) Leptin × follow-up time 0.04 (0.02 to 0.06) 0.05 (0.02 to 0.07) IL-1Ra 0.01 (−0.01 to 0.04) −0.05 (−0.09 to −0.01) IL-1Ra × follow-up time 0.03 (0.02 to 0.05) −0.02 (−0.04 to 0.00) IL-6 −0.01 (−0.03 to 0.01) 0.01 (−0.01 to 0.03) IL-6 × follow-up time 0.00 (−0.02 to 0.03) 0.03 (0.00 to 0.06) IL-8 −0.03 (−0.06 to −0.01) −0.01 (−0.07 to 0.04) IL-8 × follow-up time 0.02 (0.00 to 0.04) 0.00 (−0.05 to 0.05) IL-15 −0.06 (−0.09 to −0.03) −0.11 (−0.15 to −0.07) IL-15 × follow-up time −0.06 (−0.08 to −0.04) −0.07 (−0.10 to −0.05) IP-10 0.00 (−0.03 to 0.03) 0.00 (−0.03 to 0.04) IP-10 × follow-up time 0.01 (−0.01 to 0.04) −0.01 (−0.04 to 0.01) TNF-α −0.05 (−0.08 to −0.03) 0.01 (−0.05 to 0.06) TNF-α x follow-up time 0.00 (−0.02 to 0.02) 0.02 (−0.01 to 0.05) HbA1c CRP 0.01 (−0.01 to 0.03) −0.02 (−0.05 to 0.02) CRP × follow-up time 0.00 (−0.02 to 0.02) −0.02 (−0.04 to 0.01) Adiponectin 0.02 (0.00 to 0.04) 0.03 (0.00 to 0.05) Adiponectin × follow-up time 0.02 (0.00 to 0.03) 0.03 (0.01 to 0.05) Leptin 0.07 (0.05 to 0.09) 0.03 (0.00 to 0.07) Leptin × follow-up time 0.01 (0.00 to 0.03) 0.01 (−0.02 to 0.03) IL-1Ra 0.01 (−0.01 to 0.03) 0.01 (−0.04 to 0.03) IL-1Ra × follow-up time −0.01 (−0.03 to 0.00) 0.00 (−0.02 to 0.02) IL-6 0.01 (−0.02 to 0.01) 0.00 (−0.02 to 0.02) IL-6 × follow-up time 0.00 (−0.02 to 0.02) 0.00 (−0.03 to 0.03) IL-8 0.01 (−0.01 to 0.02) −0.03 (−0.08 to 0.01) IL-8 × follow-up time −0.02 (−0.03 to 0.00) 0.00 (−0.04 to 0.05) IL-15 −0.03 (−0.05 to −0.01) −0.02 (−0.06 to 0.01) IL-15 × follow-up time −0.01 (−0.03 to 0.00) 0.00 (−0.02 to 0.02) IP-10 −0.02 (−0.04 to 0.00) −0.02 (−0.05 to 0.02) IP-10 × follow-up time 0.00 (−0.01 to 0.02) 0.02 (0.00 to 0.04) TNF-α 0.02 (0.00 to 0.03) 0.03 (0.00 to 0.06) TNF-α × follow-up time −0.02 (−0.03 to 0.00) −0.01 (−0.04 to 0.02) Note: The ß coefficient represents the ß unit change in HbA1c/HOMA-IR z-scores per unit increase in z-scores of inflammatory markers, whereas the interaction ß coefficient represents the rate of change in the association between inflammation (z-scores) and HbA1c/HOMA-IR (z-scores) per 2-year increase in follow-up time. Associations at p < 0.05 are shown in bold. Significant associations after Bonferroni correction are shown in bold and italics. Abbreviations: CRP, C-reactive protein; IL-1Ra, interleukin-1 receptor antagonist; IP-10, interferon gamma inducible protein. a Adjusted for age and sex with follow-up time as a random slope. b Additionally adjusted for study region, waist-to-height ratio, lifetime smoking and alcohol status, pubertal status, birthweight, healthy diet adherence score, family history of diabetes, membership in sport club, screen time/week and other inflammatory markers (minimal sufficient adjustment set).