Socio-economic inequalities in CVD mortality contribute substantially to the inequality in all-cause mortality in most European countries( Reference Mackenbach, Stirbu and Roskam 1 ). Evidence suggests that CVD originates in childhood( Reference Berenson, Srinivasan and Bao 2 , Reference Juhola, Magnussen and Viikari 3 ) and that a low socio-economic position early in life increases later CVD risk independent of one’s adult socio-economic position( Reference Pollitt, Rose and Kaufman 4 ). However, we lack knowledge on modifiable factors that may help reduce socio-economic differences in cardiometabolic risk from childhood. It is, therefore, relevant to investigate whether socio-economic differences in cardiometabolic risk markers are present in childhood and whether lifestyle factors play a mediating role in these differences.
Some of the relatively few studies that have examined cardiometabolic inequalities in childhood have shown higher blood pressure( Reference Howe, Galobardes and Sattar 5 ), insulin resistance( Reference Lawlor, Harro and Wedderkopp 6 , Reference van den Berg, van Eijsden and Vrijkotte 7 ) and a more adverse lipid profile( Reference Goodman, McEwen and Huang 8 ) among the lowest socio-economic groups. Obesity is a strong driver of cardiometabolic derangements, also in childhood( Reference Weiss, Dziura and Burgert 9 ), and many studies of children from high-income countries have shown that low socio-economic position is associated with overweight or obesity( Reference Shrewsbury and Wardle 10 ). Observational studies in children and adolescents have shown that inequalities in blood pressure and insulin resistance were mediated by fat mass and BMI, respectively( Reference Howe, Galobardes and Sattar 5 , Reference Goodman, Must and Daniels 11 ). BMI is commonly used as a proxy measure of body fatness, but the fat mass index (FMI)( Reference Wells and Cole 12 ) may be better suited for investigating such associations, as it is not influenced by the amount of lean mass. Diets low in added sugar and rich in fruit, vegetables, dietary fibre and n-3 long-chain PUFA (LCPUFA) from fish have been linked to better cardiometabolic profiles in children( Reference Ambrosini, Huang and Mori 13 , Reference Brauchla, Juan and Story 14 ) and lower CVD risk in adults( Reference Hartley, Igbinedion and Holmes 15 – Reference Kotwal, Jun and Sullivan 18 ). Children of lower socio-economic backgrounds have been shown to eat less healthy diets( Reference Brug, van Stralen and te Velde 19 , Reference Rasmussen, Krølner and Klepp 20 ), which could contribute to inequalities in cardiometabolic risk markers between children of higher and lower socio-economic positions. Physical activity and sedentary behaviour may be other potential contributors as higher moderate-to-vigorous physical activity (MVPA) has been associated with better cardiometabolic profiles in schoolchildren( Reference Ekelund, Luan and Sherar 21 ), whereas sedentary behaviour has been linked to increased CVD risk in adults( Reference Ford and Caspersen 22 ). Dietary components, physical activity and sedentary time may affect cardiometabolic risk markers directly or via body fatness. However, to our knowledge, no studies have examined whether socio-economic differences in cardiometabolic risk profiles in children from high-income countries are mediated by dietary components and objectively measured physical activity or sedentary time. Also, it has not been investigated whether dietary components and physical activity play a mediating role independent of body fatness.
The aims of the present study were (i) to explore whether socio-economic differences were present in various cardiometabolic risk markers among 8- to 11-year-old Danish children; and (ii) if differences were observed, to explore whether the following were potential independent mediators: FMI; dietary components such as fruit, vegetables, fibre and added sugar; whole-blood n-3 LCPUFA (as a biomarker of fish intake); and MVPA and sedentary time.
Study design and population
The present cross-sectional study used baseline data collected during 2011 within the OPUS (Optimal well-being, development and health for Danish children through a healthy New Nordic Diet) School Meal Study. The primary aims of the OPUS study were to examine the effect of Nordic school meals on metabolic health and concentration performance in third- and fourth-grade pupils. A detailed description of the study design and methods has been published previously( Reference Damsgaard, Dalskov and Petersen 23 ). In total, 1021 pupils from forty-six school classes at nine schools were invited, and among these 834 (82 %) children participated in the OPUS School Meal Study. The present study included the 715 children from whom baseline data on household education, at least one cardiometabolic risk marker and all potential mediators and relevant covariates were available. The cardiometabolic risk markers included were blood cholesterols, TAG, glucose, homeostasis model assessment of insulin resistance (HOMA-IR), blood pressure and heart rate. Potential mediators examined were FMI; intakes of fruit, vegetables, dietary fibre and added sugar; whole-blood n-3 LCPUFA as a biomarker of fish intake; and MVPA and sedentary time. Additionally, waist circumference and BMI-for-age Z-scores were included as mediators to substantiate the results.
Interviews on socio-economic position, ethnicity and puberty
At baseline, each family was interviewed about socio-economic factors and demographics. We used the highest level of education obtained in the household as the indicator of socio-economic position, since education most consistently predicts CVD risk factors in children and adults compared with income and occupation, both in Denmark and internationally( Reference Goodman, McEwen and Huang 8 , Reference Groth, Fagt and Stockmarr 24 – Reference Winkleby, Jatulis and Frank 26 ). The level of education was divided into categories( 27 ) based on the International Standard Classification of Education (ISCED): lower secondary (≤10 years), upper secondary (12 years), vocational (12 years), short higher (14 years), medium higher (15–16 years) and long higher education (≥17 years). Lower secondary education included compulsory schooling; upper secondary education included programmes qualifying for higher educations, e.g. high school; and the vocational category included practical educations, e.g. sales assistant, carpenter, etc. The short higher education category included e.g. market economist and laboratory technician; medium higher education corresponded to bachelor degree; and the long higher education category included mainly university master and PhD degrees. We pooled data from children in the upper secondary (n 20) and vocational education categories (n 231) since they did not differ on any cardiometabolic risk markers; the group is referred to as ‘vocational education’. Children were defined as immigrants/descendants if all grandparents and at least one parent were born outside Denmark. Pubertal status was self-evaluated by the child using Tanner stages, which are based on breast development in girls and pubic hair in boys( Reference Morris and Udry 28 ). Since only forty-six children had reached Tanner stages 3–5 we used only two pubertal categories: ‘not entered puberty’ (Tanner stage 1) and ‘entered puberty’ (Tanner stages 2–5).
Recording of diet and physical activity
With parental assistance, children recorded their dietary intake for seven consecutive days using web-based dietary assessment software developed for the age group( Reference Biltoft‐Jensen, Trolle and Christensen 29 ). Dietary variables included in the present study were intake of fruit (excluding fruit juice, but including nuts), vegetables (excluding potatoes), dietary fibre and fish measured in g/10 MJ as well as intake of added sugar measured in percentage of total daily energy intake (E%). Only children with at least four days of dietary recording were included in the analyses. The ratio between energy intake (EI) and estimated BMR( Reference Henry 30 ) was used to categorize children as under-reporters (EI:BMR ≤1·05), acceptable reporters (EI:BMR >1·05 and <2·29) and over-reporters (EI:BMR ≥2·29) of energy intake( Reference Black 31 ). Physical activity was measured using an accelerometer (GT3X or GT3X+; ActiGraph, Pensacola, FL, USA) worn tightly at the right hip and affixed with an elastic belt, as previously described( Reference Hjorth, Chaput and Michaelsen 32 ). Recordings of ≥10 h monitor wear time for at least three weekdays and at least one weekend day were included in the analyses. We used 0–100 counts/min and ≥2296 counts/min as the cut-offs for time spent sedentary and in MVPA, respectively( Reference Trost, Loprinzi and Moore 33 ).
Clinical examinations were conducted in a double-decker truck equipped as a mobile laboratory and parked at the schools. Except for body composition, all measurements were performed in the morning after an overnight fast. A total of twenty-one children did not fast; eight of these had consumed only chewing gum or single bites of food. During the anthropometric measurements, the children were wearing light clothing and had emptied their bladders. Height was measured to the nearest 0·1 cm (CMS Weighing Equipment, London, UK) while the examiner made sure that the children’s heads were in the Frankfurt horizontal plane and calculated as the mean of three repeated measurements. Body weight was measured to the nearest 0·1 kg on a digital scale (Tanita 800S, Arlington Heights, IL, USA). BMI was calculated as: [body weight (kg)]/[height (m)]2, and sex- and age-adjusted Z-scores for BMI were calculated using the WHO Growth Reference 2007 for children aged 5–19 years( Reference de Onis, Onyango and Borghi 34 ). The prevalence of underweight, overweight and obesity was based on age- and sex-specific cut-offs defined by centiles passing through BMI of 18·5, 25·0 and 30·0 kg/m2, respectively, at 18 years( Reference Cole, Flegal and Nicholls 35 , Reference Cole, Bellizzi and Flegal 36 ). Waist circumference was measured at the level of the umbilicus and the mean of three repeated measurements was used. Total fat mass was assessed by whole-body dual-energy X-ray absorptiometry scan (Lunar Prodigy Pro™; GE Medical Systems, Madison, WI, USA) completed on the same day as the other measurements. Most children had a standardized breakfast prior to the scanning. FMI was calculated as: [fat mass (kg)]/[height (m)]2. Blood pressure and heart rate were measured three times in the supine position after 10 min of rest using an automated device (ProBP 3400 Sure BP; Welch Allyn Inc., Skaneateles Falls, NY, USA). A second device was used for children with arm circumferences >22 cm (UA-787 Plus; A&D Medical, Tokyo, Japan). The mean of the last two measurements was used.
A 35 ml fasting venous blood sample was drawn from the forearm. Whole-blood glucose concentrations were analysed immediately after sampling, and remaining blood was stored at −80ºC for later analysis. Plasma total cholesterol, HDL cholesterol and TAG, as well as serum insulin, were analysed, and LDL cholesterol concentrations were calculated using Friedewald’s equation, as previously described( Reference Damsgaard, Dalskov and Laursen 37 ). HOMA-IR was calculated as: [glucose (mmol/l)×insulin (mIU/l)]/22·5. Whole-blood n-3 LCPUFA was measured as EPA + DHA by high-throughput GC and is presented as relative weight of the total fatty acids (w/w%)( Reference Damsgaard, Eidner and Stark 38 ).
Differences between the long higher education group (as reference) and the other education groups were evaluated by linear or logistic mixed-effects models including school and class as random effects to account for the clustering of pupils. Our exploratory analysis consisted of four steps inspired by the causal steps approach( Reference Baron and Kenny 39 ). First, we assessed whether educational differences were present in the various cardiometabolic risk markers. Only cardiometabolic risk markers showing educational differences were included in the subsequent steps of the analysis. Second, we assessed whether educational differences were present in the potential mediators. Only potential mediators that showed educational differences were included in the subsequent step. Third, we tested potential associations between the cardiometabolic markers identified in the first step and potential mediators identified in the second step. Potential mediators that were associated with the cardiometabolic markers were included in the fourth step: for each cardiometabolic risk marker, we expanded the models from step 1 by including the relevant mediators identified in step 3. Mediators were included in the models individually and together to evaluate whether they were independent and hence to account for multicollinearity in the interpretations. For example, if a dietary component was still significant in the model after adjusting for body fatness, it was considered to be a mediator independent of body fatness. For each cardiometabolic risk marker, the absolute mediated effects were calculated as β from model without mediator(s)−β from model including mediator(s) and the proportions mediated as [(β from model without mediator(s)−β from model including mediator(s))/β from model without mediator(s)]×100 %. The absolute mediated effects were evaluated using Sobel’s test.
All models were adjusted for age and sex. Analyses of cardiometabolic risk markers were furthermore adjusted for height, puberty and a sex × puberty interaction. Additionally, analyses of HOMA-IR, glucose and blood lipids were adjusted for fasting status, and analyses of blood pressure and heart rate were adjusted for blood pressure device and ambient temperature. Analyses of FMI were additionally adjusted for puberty; analyses of self-reported dietary variables were adjusted for under-, acceptable and over-reporting of energy intake; and analyses of MVPA and sedentary time were additionally adjusted for a sex × age interaction. Associations between whole-blood n-3 LCPUFA and fish intake were evaluated using Pearson’s correlation coefficient. Model assumptions were checked by visual inspection of residual and normal probability plots. When necessary, variables were transformed and results subsequently back transformed( Reference Laursen, Dalskov and Damsgaard 40 ). Statistical analyses were performed using the statistical software package Stata® version 12·1 and R version 3·0·2( 41 ). Statistical significance was established at P<0·05, using two-sided tests.
Characteristics of the study population
The age, height and sex distribution of the children did not differ between the long higher and the other education groups (Table 1). Overweight and obesity proportions were higher in all groups where parents had less than a long higher education level, and it was highest in children from the group whose parents had a lower secondary education level. Children in this group were also more likely to be immigrants or descendants and to have entered puberty, compared with the long higher education group. However, the pubertal difference was no longer significant after adjustment for BMI-for-age Z-score and sex. Under- and over-reporters of energy intake comprised 8·1 % (n 58) and 1·7 % (n 12) of the study population, respectively, but the proportions of under-reporters differed only between the lower secondary and long higher education groups (31·4 % v. 6·0 %, P<0·001). Among the 834 children originally enrolled in the OPUS School Meal Study, the 715 children included in the present study did not differ from the 119 not included with regard to BMI, age or gender, but they were less likely to have parents with a lower secondary education (P<0·01; data not shown).
*P<0·05, **P<0·01 and ***P<0·001 compared with long higher education.
† Analysed by logistic or linear mixed-effects models adjusted for school and school class.
‡ Self-evaluated Tanner stages( Reference Morris and Udry 28 ).
|| Children were defined as immigrants/descendants if all grandparents and at least one parent were born outside Denmark.
Educational differences in cardiometabolic risk markers
The lower secondary education group had higher plasma TAG and HOMA-IR compared with the long higher education group, and the vocational education group had higher total and LDL cholesterol (Table 2). The short higher education group differed from the long higher education group only with respect to plasma TAG. Blood pressure, heart rate, plasma HDL cholesterol and glucose did not differ between the long higher and other education groups.
SBP, systolic blood pressure; DBP, diastolic blood pressure; TC, total cholesterol; LDL-C, LDL cholesterol; HDL-C, HDL cholesterol; HOMA-IR, homeostasis model assessment of insulin resistance.
*P<0·05 and **P<0·01 compared with long higher education.
† All models were adjusted for school, school class, sex, age, height, puberty and a sex×puberty interaction. TAG and HOMA-IR were log-transformed in the analyses, but the results are presented on the original scales.
‡ Additionally adjusted for ambient temperature and blood pressure device.
§ Additionally adjusted for fasting status.
|| Data on HOMA-IR were available for 675 children only: lower secondary (n 31), vocational (n 236), short higher (n 68), medium higher (n 198) and long higher (n 142).
Educational differences in potential mediators
Children of parents with shorter educations had higher FMI, waist circumference, BMI-for-age Z-scores, lower intakes of vegetables and dietary fibre, lower whole-blood n-3 LCPUFA and higher intakes of added sugar compared with the long higher education group (Table 3). In contrast, these groups spent less time sedentary than the long higher education group. Generally, the educational differences were largest between the lower secondary and the long higher education groups. Fruit intake and time spent in MVPA did not differ between the long higher and other education groups, nor did mean accelerometer wear time (P>0·2; data not shown).
n-3 LCPUFA, n-3 long-chain PUFA (w/w% of the total fatty acids measured in whole blood); FMI, fat mass index; WC, waist circumference; E%, percentage of energy intake; MVPA, moderate-to-vigorous physical activity.
*P<0·05, **P<0·01 and ***P<0·001 compared with long higher education.
† All models were adjusted for school, school class, sex and age. Intake of dietary fibre was log-transformed, and intakes of fruit, vegetables and added sugar were transformed using the square root in the analyses, but the results are presented on the original scale.
‡ Analyses of FMI, WC and BMI were additionally adjusted for puberty, and analyses of WC for height.
§ Additionally adjusted for under-, acceptable and over-reporting of energy intake.
|| Additionally adjusted for an age×sex interaction.
Additional analyses showed that 76 % of the children consumed fish and 15 % consumed fish oil during the dietary recording week. The percentage of children eating fish was lower among the children of parents with shorter educations (ranging from 67 % in the lower secondary to 76 % in the medium higher education group) compared with 87 % in the long higher education group (P<0·05). Mean frequency of fish-oil consumption was lower in the vocational, short higher and medium higher education groups (0·3 d/week) compared with 1 d/week in the long higher education group (P<0·001). The n-3 LCPUFA content in whole blood correlated positively with the reported intake of fish (r=0·41, P<0·0001) and fatty fish (r=0·39, P<0·0001).
Associations between potential mediators and cardiometabolic risk markers
Among the putative mediators, FMI was positively associated with total and LDL cholesterol (P<0·01) and with plasma TAG and HOMA-IR (P<0·001). Whole-blood n-3 LCPUFA was negatively associated with plasma TAG (P=0·002) and HOMA-IR (P=0·001). Dietary fibre intake was also negatively associated with HOMA-IR (P=0·001). Intakes of added sugar and vegetables and sedentary time were not associated with any of the cardiometabolic risk markers that showed educational differences. Accordingly, FMI, whole-blood n-3 LCPUFA and dietary fibre intake were evaluated further as potential mediators.
Estimated mediated effects
Table 4 shows how much of the educational differences in the cardiometabolic markers were accounted for by the mediators in absolute values and percentages. FMI accounted for part of the differences in the cholesterol concentrations, plasma TAG and HOMA-IR between the long higher education group and other groups with shorter educations. Whole-blood n-3 LCPUFA also accounted for part of the differences in plasma TAG and HOMA-IR, whereas dietary fibre intake accounted for part of the difference in HOMA-IR only.
TC, total cholesterol; FMI, fat mass index; LDL-C, LDL cholesterol; n-3 LCPUFA, n-3 long-chain PUFA (w/w% of the total fatty acids measured in whole blood); HOMA-IR, homeostasis model assessment of insulin resistance.
*P<0·05, **P<0·01, ***P<0·001.
† All models were adjusted for school, school class, sex, age, height, puberty, a sex×puberty interaction and fasting status. TAG and HOMA-IR were log-transformed in the analyses, but the results are presented on the original scales.
‡ Data on HOMA-IR were available for 675 children only: lower secondary (n 31), vocational (n 236), short higher (n 68), medium higher (n 198) and long higher (n 142).
§ Additionally adjusted for under-, acceptable and over-reporting of energy intake.
The estimated mediated effects of the educational differences in plasma TAG and HOMA-IR became larger when the relevant mediators were entered together in the models. For example, FMI and whole-blood n-3 LCPUFA combined accounted for 0·05 mmol/l (42 %) out of the 0·12 mmol/l difference in TAG, whereas FMI, dietary fibre intake and whole-blood n-3 LCPUFA accounted for 0·32 points (89 %) out of the 0·36 point difference in HOMA-IR between the long higher and lower secondary education groups. All mediators remained associated with the cardiometabolic risk markers when included in the models presented in Table 4. Including waist circumference or BMI-for-age Z-score as mediators, instead of FMI, confirmed the results (data not shown).
The present study showed that shorter parental education was associated with higher insulin resistance and a more adverse lipid profile in Danish 8- to 11-year-olds. As expected, body fatness was an important explanatory factor in these associations. However, as a novel finding, whole-blood n-3 LCPUFA and dietary fibre intake were also seen to mediate part of the educational differences in some of the cardiometabolic risk markers. This indicates that educational differences in intakes of fish and dietary fibre affect cardiometabolic differences as early as in childhood.
Interpretation and comparison with previous studies
To our knowledge, no previous studies have investigated whether socio-economic differences in fish intake, whole-blood n-3 LCPUFA or dietary fibre intake serve as links between socio-economic position and cardiometabolic risk markers. n-3 LCPUFA supplementation consistently lowers plasma TAG in adults( Reference Mozaffarian and Wu 42 ) and has been shown to improve insulin resistance in obese children( Reference López-Alarcón, Martínez-Coronado and Velarde-Castro 43 ). Also, previous studies have found an association between cheek-cell n-3 LCPUFA and the degree of social deprivation in children( Reference Kirby, Woodward and Jackson 44 ) and erythrocyte n-3 LCPUFA and level of education in adults( Reference Cohen, Garg and Ali 45 ). Altogether, this indirect evidence supports our finding that whole-blood n-3 LCPUFA has an explanatory role in the association between socio-economic position and cardiometabolic risk markers. Oily fish is the main dietary source of n-3 LCPUFA in Danes, but fish-oil consumption may also contribute. In our study, an educational gradient was present in fish consumption, whereas this was not the case for fish-oil intake. Also, 76 % of the children consumed fish, whereas only 15 % consumed fish oil. Therefore, fish intake was most likely the main contributor to the educational differences in whole-blood n-3 LCPUFA.
Higher intake of dietary fibre has been associated with a reduced risk of overweight, obesity and impaired glucose metabolism in children( Reference Brauchla, Juan and Story 14 ) and lower risk of type 2 diabetes in adults( Reference Schulze, Schulz and Heidemann 46 ). Low intake of dietary fibre may increase insulin resistance by contributing to overweight. However, to our surprise, we found that dietary fibre and whole-blood n-3 LCPUFA explained part of the difference in HOMA-IR between the children of parents with the shortest and longest educations, independent of body fatness. Similarly, whole-blood n-3 LCPUFA explained part of the educational differences in TAG independent of body fatness. This indicates that whole-blood n-3 LCPUFA and dietary fibre have an explanatory role in relation to cardiometabolic inequalities in children in addition to overweight. Still, our results showed that educational differences in overweight are important as a mediating factor. Moreover, body fatness was the only presumed mediator of the association between shorter parental education and higher total and LDL cholesterol. In support of this finding, increased BMI has been shown to increase childhood concentrations of total and LDL cholesterol( Reference Dai, Fulton and Harrist 47 ).
Differences in HOMA-IR and TAG were observed only between the reference group and the lower secondary and short higher education groups, respectively, the latter two being relatively small groups. Although the small size of the education groups increases the uncertainty of the estimates, the educational differences were comparable to those observed in previous studies( Reference Lawlor, Harro and Wedderkopp 6 , Reference Thomas, Nightingale and Donin 48 ). In contrast, no educational differences were seen in blood pressure. A possible explanation could be a larger variation in this measure compared with the risk markers measured in blood.
Generalizability and implications for public health
The distribution of children in the education groups matched the distribution in Danish families with children fairly well, except that the lower secondary education group was somewhat under-represented, comprising 5 % in our study v. 10 % among all Danish families with children( 49 ). Therefore, the estimates for this group may be more prone to bias than is the case for the other education groups. However, in the case of bias, the estimated differences between this group and the long higher education group would probably be underestimated as the children enrolled from the lower secondary education group were likely to be among the most advantaged children from that education group in the background population. The proportion of immigrant/descendants matched the 10 % reported in the total Danish population( 50 ), the prevalence of overweight and obesity corresponded to that found in a recent cohort of Danish schoolchildren( Reference Pearson, Hansen and Sørensen 51 ) and the diet of the children was similar to that reported among children in the representative Danish National Dietary Survey( Reference Pedersen, Fagt and Groth 52 ). This indicates that the study population was, to a large degree, representative of Danish 8- to 11-year-olds and that the results may be generalizable.
The immediate clinical implications of 0·36 point higher HOMA-IR and 0·12 mmol/l higher plasma TAG in children of parents with the shortest compared with the longest education are uncertain. However, blood lipids show tracking from childhood to adulthood( Reference Juhola, Magnussen and Viikari 3 ) and metabolic dysregulations during childhood may increase the risk of metabolic syndrome and diabetes in adulthood( Reference Morrison, Friedman and Wang 53 ). Thus, since atherosclerosis is a gradual, lifelong process, it is probable that a reduction in socio-economic differences in cardiometabolic risk markers in childhood will beneficially affect later inequalities in CVD risk. As indicated by our results, not only educational differences in overweight but also in consumption of fish and dietary fibre per se were possible explanations for differences in cardiometabolic risk markers and may therefore be important targets in public health strategies to prevent inequalities in children’s cardiometabolic health.
Strengths and limitations
A major strength of the present study is the use of body fatness by dual-energy X-ray absorptiometry scan, a biomarker of fish intake and objectively measured physical activity. Whole-blood n-3 LCPUFA, instead of reported fish intake, was a relevant mediator to include in order to overcome potential misreporting and recall bias. Twenty-four per cent of the children reported not having eaten fish during the 7 d recording period, but they may have eaten fish during previous weeks. The response of whole-blood fatty acids to n-3 LCPUFA supplementation has been shown to be intermediate between that of plasma and erythrocyte fatty acids, which reflect intakes over the last days to weeks and over the last months, respectively( Reference Metherel, Armstrong and Patterson 54 ). Whole-blood n-3 LCPUFA correlated fairly well with fish intakes and, compared with erythrocytes, whole blood required less processing in the laboratory truck. Physical activity was measured by accelerometers that have a much higher validity than self-reported physical activity( Reference Westerterp 55 ). As the categorization of minutes spent in MVPA has not been validated for triaxial data, we used the vertical axis only with a position capturing movements by the centre of mass. The advantage of this position is a relatively close relationship with energy expenditure, whereas the disadvantage is a limited ability to capture bicycling, which often belongs to the MVPA category. Only 82 % of the children reported time spent bicycling, therefore we could not include this parameter in the analyses.
We used a relatively detailed grouping of education, as we wanted to explore whether the educational differences were general trends in the population or related to particular, vulnerable groups. As a limitation, this approach precluded detailed investigation of the interplay between ethnicity, sex and parental education. Under-reporting of energy intake was higher in the lower secondary education group but was adjusted for in the analyses and has therefore probably not affected the results substantially. We also acknowledge that the many statistical tests performed give an increased risk of false positive findings; however, due to the exploratory nature of the study, adjustment for multiplicity was not applied. Finally, the use of cross-sectional data does not allow conclusions on causality. However, the risk of reverse causality is low, as it seems unlikely that children’s cardiometabolic profiles influence their dietary habits or their parents’ education levels, and the potential mediators were chosen based on previous evidence showing that they affect the cardiometabolic risk markers. Still, residual confounding, by, for example, other dietary components or psycho-social influences, cannot be precluded.
Among Danish 8- to 11-year-olds, children of parents with shorter compared with long educations had higher insulin resistance and less favourable lipid profiles. Body fatness, whole-blood n-3 LCPUFA and dietary fibre intake, but not MVPA or sedentary time, were independent mediators of these differences. The mediators explained a greater proportion of the educational differences when combined than when assessed individually. Therefore, not only prevention of overweight but also increasing the intakes of fish and dietary fibre among children of parents with shorter educations could be potential targets for public health interventions to reduce socio-economic inequalities in cardiometabolic risk from childhood. However, as these cross-sectional results do not justify conclusions on causality, confirmation in longitudinal studies is warranted.
Acknowledgements: The authors are grateful to the participating children, their families, the school administrators, teachers and other staff, and thank the entire study team for their contribution to this work. Financial support: The OPUS School Meal Study was supported by the Nordea Foundation (grant number 02-2010-0389). The Foundation did not have any role in the design, data collection, analysis, interpretation of results or the decision to publish. Conflict of interest: None. Authorship: H.H. designed and performed the statistical analysis and wrote the manuscript. M.V.G. designed the analyses and helped interpret the data. C.R. supervised the statistical analysis and helped write the statistics section and interpret the data. A.B.-J. and R.A. collected and processed the dietary data. S.-M.D. collected the data and processed data related to body fatness. M.F.H. collected and processed the physical activity data. A.S. designed the research and supervised the collection of the physical activity data. A.A. and K.F.M. designed the research and supervised the data collection. C.T.D. designed the research and analysis, interpreted the data and helped write the manuscript. All authors critically reviewed the manuscript and approved the final version. Ethics of human subject participation: The study was conducted according to the principles of the Declaration of Helsinki and approved by the Committees on Biomedical Research Ethics for the Capital Region of Denmark (H-1-2010-124), and written informed consent was obtained from all custody holders of the children.