Participants
The data were derived from the Danish part of the European Youth Heart Study (EYHS); an international study addressing cardiovascular disease risk factors in children and adolescents[15]. The study was approved by the ethic committee of Vejle and Funen.
The subjects all lived and attended school in the community of Odense in 1997 and participated in both EYHS-I, as 8–10-year-old third grade children, and in the follow-up study, EYHS-II, in 2003 as 14–16-year-old ninth grade adolescents.
Sampling
The sampling frame was a complete list of public schools in the Municipality of Odense. Schools were stratified according to location (urban, rural) and the socio-economic character of its uptake area. From each stratum, a proportional, two-stage cluster sample of children was selected.
The primary units were the schools. Schools were selected using probability proportional to school size. Each school on the sampling list was allocated a weighting equivalent to the number of children in the school who were eligible to be selected for the study.
The secondary units were the children in the schools. Equal numbers of children were sampled from each school. Children in the appropriate age band were allocated code numbers and randomly selected using random number tables.
A more detailed description of the sampling procedure has been given elsewhere [15].
Study sample
In all, 771 third grade children from 25 different schools were invited to participate in EYHS-I. A total number of 589 children (310 girls and 279 boys) participated in the study, corresponding to 76.4 % of the invited children.
Six years later, 384 of these children (214 girls and 170 boys) were re-examined in EYHS-II as ninth grade students, corresponding to 49.8 % of the original sample. One of the reasons for non-participation in the follow-up study was reluctance to travel to the city of Odense for the examinations by several of the participants in EYHS-I, who no longer lived in the Community of Odense at the time EYHS-II was initiated. If the coefficient of participation is calculated only taking into account subjects who still attended school in the Municipality of Odense in 2003, the coefficient rises from 49.8 % to 57.0 %.
Representativeness
The sample of 589 participants in EYHS-I has been shown to be a representative subset of the total sample of 771 invited children with respect to physical activity level and body composition [16]. Other parameters have not been investigated.
Possible dropout effects in EYHS-II were examined by comparing baseline values between the group of children who participated in both studies and the group who only participated in EYHS-I. The socio-economic and sex-specific differences in mean BMI/PF between participants and dropouts were tested.
With respect to BMI no significant difference was found between dropouts and participants in either of the two SES groups. However, a significant dropout effect was observed in girls of high SES and in boys of low SES with respect to physical fitness. In these groups, the children who only participated in EYHS-I possessed a lower physical fitness level than the children who also participated in EYHS-II.
Measurements
Physical fitness
In both studies physical fitness was determined by a progressive maximal cycle ergometer test. The protocol has been described in detail elsewhere [15]. The test was validated in both children and adolescents with a correlation coefficient of r = 0.89 and r = 0.90, respectively, to directly measured VO2-max. The cycle-ergometer used was a computerized Monark 839 Ergomedic, and it was programmed to increase the workload every third minute by a predefined quantity. The workload was increased until exhaustion. During the test the heart rate was monitored using a Polar Vantage NV monitor and at the time of exhaustion, the maximal heart rate and the total time were registered.
Criteria for exhaustion were:
-
1.
A heart rate above or equal to 185 beats per minute
-
2.
Failure to keep a pedalling frequency of at least 30 revolutions per minute
-
3.
Subjective valuation by the test personnel
In the absence of objective cut points for low physical fitness, subjects who belonged to the lower sex and age group specific quartile of physical fitness were defined as having a low physical fitness level (Low PF).
Height/weight
Body height was measured to the nearest half centimetre using a stadiometer and body mass was determined by a beam-scale weight to the nearest 100 g. The subjects were only wearing underwear and a T-shirt during the examinations.
Body fatness
Body fatness was assessed by BMI calculations. The classification of overweight was based on BMI cut-points proposed by the International Obesity Task Force (IOTF) [17]. The cut points for overweight used included obese cut point values.
Pubertal stage
Assessment of biological maturation was based on Tanner's pubic hair stages for boys and Tanner's breast development stages for girls [18].
Socio-economic status
The classification of socio-economic status was based solely on information regarding the adult female in the household at the first measurement point, since recent evidence suggests that at least one of the risk factors considered in this study is stronger associated with the SES of the mother rather than that of the father. Two European studies on large cohorts of children have shown that the risk of being overweight in childhood/early adolescence is related to the educational level of the mother rather than that of the father [19, 20]. The mothers were categorised into two different socio-economic groups. The two groups represented blue-collar and white-collar occupation respectively and they were defined by The International Standard Classification of Occupation scheme [21], which holds nine major categories. The categories 1–4 correspond to "blue-collar" occupation while the categories 5–9 correspond to "white-collar" occupation. Mothers working as housewives (a total number of 5 in the present sample) were classified as housekeepers (i.e. classified as belonging to the blue-collar group).
In the following text, subjects in the blue-collar group will be considered as having low SES and subjects in the white-collar group as having high SES.
Statistics
Descriptive statistics
To test the difference in mean BMI/PF between groups of SES, while controlling for biological maturation state, regression analyses were applied. The corresponding differences in proportions of overweight/low PF level were tested by Fisher's exact test.
Tracking
The term "tracking" can be defined as: 1) the overall stability of a given variable through time (from T1 to Tn). 2) the predictability of information regarding risk status at T1 on risk status at Tn [22].
To assess the stability of PF and BMI throughout the measurement period, the following statistical model was used:
where Yi1 is the initial observation for subject i, Yi2 is the second observation for subject i, β1 is the regression coefficient used as the stability coefficient, j is the number of time-independent covariates, Xij is the time-independent covariate j of individual i and εi is the error term for subject i. The model can be fitted by traditional regression techniques with robust standard errors and is derived from a more general formula for assessing tracking, outlined by Twisk [23].
In order to account for scale differences, all continuous variables were transformed into subgroup specific z-scores before entering the model.
Differences between stability coefficients were tested by the following test statistics:
where and are subgroup specific standardized regression coefficients.
To assess the predictive value of being overweight/having low PF at T1 on the "risk status" at T2, logistic regression was used. The odds ratio comparing odds of maintaining overweight/low PF during the measurement period with the odds of developing overweight/low PF at some point later than T1, expresses the predictability in question. Differences in predictability between groups of SES were tested by interaction terms in the logistic regression model.
Logistic regression models were also used to assess the risk of both developing and maintaining overweight/low physical fitness within the measurement period in different groups of SES.
All tracking analyses and logistic regression models were controlled for biological maturation state, and interaction terms between gender and the independent variables were added in all models to investigate whether different models should be fitted for the two sexes separately.
Dropout
In order to compensate for the dropout effects observed, individual measurements of PF were weighted in all tracking and logistic regression models by the inverse probability of a subject turning up at the follow-up examination given the gender, SES and the PF level at baseline [24].