Cardiovascular diseases (CVD) are a major health problem in the industrialized world. More than 40% of all deaths in Denmark (24.156 of 60000 in 1998) are caused by CVD. The highest prevalence of several CVD risk factors including obesity, smoking and low physical activity level is observed in adults of low socio-economic status (SES). 123. In fact, CVD show a relatively clear socio-economic gradient in adults 4567. Naturally, this gives rise to concern not only as regards the health of adults of low socio-economic status but also as regards their children. Parents have been shown to be effective teachers of health habits, and children's lifestyle, health beliefs and behaviour are significantly influenced by their parents' behaviour 8. A number of studies have been designed to investigate the association between socio-economic status and coronary heart disease risk factors in children and adolescents. However, the results have so far been inconsistent – thus the picture is less clear than concerning adults 910. It has been suggested that this inconsistency could be, at least partly, due to differences in the ages of the populations being studied but further investigations are necessary 9.

In order to work out effective preventive strategies against CVD it is important to determine at which stage in life the relatively clear socio-economic gradient in obesity, as well as other CVD risk factors, observed in adults start developing. Furthermore, due to the expected changes in lifestyle/risk status between groups of SES at some point in childhood or adolescence, it is appropriate to examine the stability of risk factors within different groups of SES. It is often advocated that an important aspect of the preventive effort against CVD is to identify subjects at risk of developing CVD at an early point in time, so preventive strategies or clinical treatment can be applied at early stages. The rationale behind this preventive strategy is that measurements of risk factors for CVD early in life predict values later in life. Several studies have described tracking of CVD risk factors in children 11121314. However, to our knowledge, no study has analysed and compared tracking of CVD risk factors among children from families of different socio-economic status and thereby posed the question: Are the fundamental possibilities of predicting risk at an early point in time the same for different groups of socio-economic status?

The aim of this article is twofold: Firstly, to analyse tracking of physical fitness (PF) and tracking of body mass index (BMI) in Danish adolescents from families of different socio-economic status. Secondly, to investigate whether social class differences in the prevalence of overweight and low PF exist or develop within the age range given.

Table 1 presents mean values of BMI and PF together with prevalences of overweight and low PF in 8–10-year-old children within groups of SES. Due to the significant dropout effect observed in EYHS-II, data are presented for the total sample of participants in EYHS-I. Of the 589 children who participated in EYHS-I, 42 families did not provide the necessary information for the classification of SES and 50 subjects failed to meet the criteria for exhaustion in the PF-test. In general, results showed a consistent pattern of more favourable values of the parameters in question in the group of high SES compared to the group of low SES. However, only differences in mean values of PF and in the prevalence of low PF (boys only) were significant. When interpreting the results, it is important to remember that importance should not be attached to the absolute prevalence of low PF, since the definition of low PF is based on relative cut-points. Instead, prevalences of low PF should be used only for relative comparisons between groups of SES.

Tables 2, 3, 4 present results from the longitudinal analyses performed on the 384 subjects who participated in both EYHS-I and EYHS-II. Stability coefficients of physical fitness and BMI are shown in Table 2. When comparing stability coefficients between groups of SES no statistical difference is seen. However the result indicates that the stability of PF and BMI is moderate and moderately high, respectively, independently of socio-economic position. It has been suggested that values of 0.30 to 0.60 show moderate stability while values between 0.60 to 0.90 indicate moderately high stability 25.

In Table 3 predictive values of childhood measurements of PF and BMI are presented. An odds ratio of 4.29, as seen in the group of low SES with respect to physical fitness, indicates that the odds of maintaining low PF throughout the whole measurement period is 4.29 times bigger than the odds of developing low PF at some point later than T₁. The closer odds ratios get to 1 the lower the predictability, because the odds of maintaining risk then will be close to the odds of developing risk and, in such cases, information about risk status at T₁ would offer no information regarding the risk profile at T₂.

Table 4 presents odds ratios for being overweight/having low PF at the age of 14–16 years in the group of low SES compared to the group of high SES. As can be seen from the table, both the risk of being overweight and the risk of having low PF at the age of 14–16 are approximately twice as high in the group of low SES compared to the group of high SES. In order to learn about how these gradients have evolved, Table 4 also presents odds ratios for maintaining and developing risk within the measurement period. The social gradients observed at T₂ are the combined result of the absolute prevalence of risk at T₁, shown in Table 1, and the development and persistence of risk within the time range from T₁ to T₂. Both the odds of maintaining and developing overweight were more than twice as big in the group of low SES compared to the group of high SES whereas the corresponding odds for low PF were 1.36 and 1.94 times bigger in the low SES group, respectively. It should be noted, however, that only odds ratios regarding the development of overweight and low physical fitness came out significant and borderline significant, respectively.

To give an impression of the absolute probabilities from which the relative odds ratios in Tables 3 and 4 are based, Table 5 shows the raw percentages of subjects who remain "at risk" throughout the measurement period and the percentage of subjects who develop into "risk". As can be seen from the table the socioeconomic differences in the absolute probabilities are considerable of size and thus important from a public health perspective.

Interaction terms between gender and the independent variables were added in all tracking and logistic regression models and came out "not significant". Therefore, analyses were performed for the two genders simultaneously, but controlled for maturity state.

The first aim of the present study was to analyse tracking of physical fitness and tracking of body mass index from childhood to adolescence within low and high SES groups. In general, results showed no significant differences in tracking of PF and BMI between the two groups of SES – both the stability coefficients and the predictive values of measurements recorded in childhood did not differ significantly between the two groups. However, the stability and predictability coefficients were highly significant when calculated independent of SES on the total sample and they were moderate and moderately high for PF and BMI, respectively.

The interpretation of the size of the stability coefficients was based on predetermined cut-points distinguishing between low, moderate and high tracking effects. However, it is important to realise that the results of a tracking analysis depends on the length of the measurement period, the age of the subjects being examined, the statistical methods used, and the testing methods 26. Thus, the cut-points used cannot be expected to hold true for all other tracking studies, and it can be difficult to compare results of tracking analyses between studies, as also pointed out by Twisk et al. 22. No earlier study has analysed and compared tracking of CVD risk factors between children from families of different SES, thus it is not possible to compare the current findings to the work of others. Even when looking at the results independently of SES, few studies are directly comparable to our study in all of the above mentioned aspects, but despite differences in methods and ages of the populations being studied, a review of relevant literature revealed that our conclusion regarding tracking of BMI and PF, independent of SES, is in line with previous studies 1112272829.

The implication of the results of the tracking analysis is that SES does not affect tracking and thereby the potential of early identification of subjects who will be unfit or overweight at a later stage in life. Furthermore, intervention programs targeting subjects who are classified as being overweight in childhood could play an essential role in the strategy for lowering the prevalence of overweight in late adolescence. In contrast, childhood measurements of PF are only moderate predictive of PF values in adolescence, resulting in more difficult conditions for identifying adolescents of low fitness at an early point in time.

The second aim of the study was to investigate whether social class differences in the prevalence of overweight and low PF exist or develop within the age range given. Longitudinal analyses showed that the onset of the development towards a social gradient in the prevalence of overweight occurs as early as during the age span of 8–16 years. As regard PF, socioeconomic differences were already present at the age of 8–10 years, indicating that the onset of the development towards a social gradient in the prevalence of low PF occurs earlier than at the age of 8–10 years.

As described in the introduction, earlier studies on the association between SES and coronary heart disease risk factors in children and adolescents have produced inconsistent results 910. It has been suggested that the inconsistencies, at least partly, could be due to differences in the ages of the populations being studied 910. If socioeconomic gradients do not develop until a certain stage in life is reached, naturally the association between SES and CVD risk factors will vary according to age.

Different theories have been put forward as to which stage in life the relatively clear socio-economic gradient in obesity, as well as other CVD risk factors observed in adults, start developing. Van Lenthe et al. reported no differences in biological CVD risk factors (including body fatness and PF) in both 12 and 15-year-olds but, in contrast, they observed that the low SES group had or developed a less healthy life style between 12 and 15 years of age compared to subjects in the high SES group. They hypothesised that SES differences in behavioural risk factors develop during adolescence, and will cause differences in biological risk factors to develop at the end of puberty or during early adulthood 10.

The discrepancy between the study of Van Lenthe et al. and the results of this study cannot be explained by age differences, as indirectly suggested in the study by Sobal and Stunkard 9, since socioeconomic gradients in CVD risk factors were observed in the younger of the two populations. A review of relevant literature revealed additional examples of inconsistencies between studies when taking into account the age of the subjects. For instance our cross-sectional finding that the prevalence of overweight did not differ significantly between groups of SES at the age of 8–10 years contradicts the results of another large study from a developed European country, which reported an inverse social gradient in the prevalence of overweight as early as at the age of 5–7 years 30.

A possible, but probably incomplete, explanation of the inconsistencies could be that methodological differences between studies affect the ability to detect differences in risk factors between groups of SES. For instance, different definitions of SES might not capture the factors responsible for the development of social gradients to the same extent. We tested this theory indirectly by applying a new definition of SES to our data. SES was redefined and now based on the male or female parent of the household with the highest status according to The International Standard Classification of Occupation scheme. In general the results showed the same overall pattern. The significant differences in low PF and insignificant differences in the prevalence of overweight, seen in Table 1, were reproduced. Also, the significant socioeconomic gradient in PF at the second measurement was maintained. However, the odds ratios of developing and being overweight at T₂ across socioeconomic strata were no longer significant, although only small reductions were observed in the size of the odds ratios. This change in the results could very well reflect a stronger association of childhood overweight to the SES of the mother rather than that of the father, in accordance with earlier findings in the literature 1920. However the point is that the new definition of SES would in fact give rise to some changes in the conclusions, if the interpretations were based entirely on the observed p-values.

A further source of inconsistency between studies, when exploring at which age social gradients develop, could be differences in sample power size between studies. At which point differences between groups of SES become significant is highly dependent on sample size. For instance, the insignificant socio-economic differences in the prevalence of overweight, seen in Table 1, could in fact represent a small but true difference and thus have turned out significant given a higher power.

Finally, it can be questioned whether differences between groups of SES emerge at the exact same age in all developed countries or if individual differences exist between developed countries, analogous to the fact that the association between SES and CVD risk factors differs between developed and developing countries 3132.

To which extent these methodological issues can account for previous inconsistent findings regarding the association between SES and CVD risk factors, is without doubt an important research question for future studies. Consistent information about the timing and development of socio-economic gradients in CVD risk factors is important to help set the scene for more detailed research into the possible physiological and psychosocial mechanisms responsible for the socio-economic inequalities.

In conclusion, tracking of PF and BMI from childhood to mid-adolescence is not significantly different between groups of SES, thus indicating that the fundamental possibilities of predicting risk at an early point in time are the same for different socio-economic status groups.

However, social inequalities in the absolute prevalence of overweight and low PF are developing or already present during this particular age period. Results revealed the presence of a social gradient in the prevalence of low PF at the age of 8–10 years, favouring the children of high socioeconomic status and demonstrated that a social gradient in the prevalence of overweight develops no later than during the age period from 8–16 years, partly due to a higher incidence of overweight in the low SES group. These results underline the need for preventive efforts targeting children of low SES in early childhood and furthermore indicate that efforts should not be based solely on the principle of tracking (i.e. aimed exclusively at children who are already classified as being "at risk" at the time given), but initiatives should also be taken to lower the occurrence of new cases of overweight and low PF in early childhood within the group of low SES.

The author(s) declare that they have no competing interests.

NCM has contributed to the analysis and interpretation of data and in preparing the manuscript. CNB has contributed to the acquisition of data and in preparing the manuscript. KF, NW and LBA participated in the design of the study and contributed to the analysis and interpretation of data. PLK performed the statistical analysis and interpretation of data, contributed to the acquisition of data and wrote the paper. PLK revised it with contributions from all authors. All authors read and approved the final manuscript.