The adolescents selected for this study participated in the Dutch Obesity Intervention in Teenagers (DOiT). The DOiT-study is a randomized controlled trial in which the effectiveness of a school-based multi-component intervention aimed at preventing excessive weight gain was evaluated (for details see Singh et al 28). Data were collected at baseline (September/October 2003; T0) and after eight (T1), twelve (T2) and 20 months (T3). All aspects of the study were approved by the ethics committee of the VU University Medical Center. To minimize confounding effects of the DOiT-intervention, only the adolescents assigned to the control group were included in the current analyses.

Data of 465 adolescents (53% boys) were available for the present analyses (e.g. at T1, T2 and T3). Due to missing data sample sizes varied between 247 and 177 for boys and between 218 and 173 for girls.

Descriptive subject characteristics (mean (SD)) were calculated at each time point. Differences between boys and girls were examined using the nonparametric Mann-Whitney U test. Since a previous study found that dose-response associations between screen time and overweight were only found among adolescent girls 16, separate analyses were performed for boys and girls. Multi-level linear regression analysis (MLwiN, version 2.22; Centre for Multilevel Modeling) was used to examine whether the amount of screen time predicted changes in measures of body fatness, and whether measures of body fatness predicted changes in the amount of screen time. In these analyses the outcome variable at time point t was related to the value of the predictor at time point t-1, adjusted for the value of the outcome variable at t-1 (autoregressive model). In addition, adjustments were made for ethnicity (based on the country of birth of the parents) and aerobic fitness. The association found can be interpreted as an association of the predictor with changes in the outcome measure, independent of ethnicity and aerobic fitness. Finally, interaction effects for time were checked by including a 'time × predictor' interaction term in the analyses. Using multilevel analyses, regression coefficients can be adjusted for the clustering of observations within one school and/or class. Three levels were defined: 1) student, 2) class, and 3) school. The level of significance was set at P < 0.05.

Body fatness did not predict any changes in TV viewing time and computer time in adolescent boys and girls (Table 2).

TV viewing time significantly predicted changes in BMI and HC in boys (Table 3). TV viewing time did not predict changes in body fatness in girls (Table 3). Computer time significantly predicted changes in skinfold thickness in boys and girls and in BMI in boys (Table 3). Furthermore, time significantly modified the association between computer time and skinfold thickness in girls (b = 0.026), indicating that the strength of this association increases in time. These results indicate that when using the computer for 119 min per day (average time spent using the computer for girls in this study; Table 1), the sum of skinfolds may increase with 1.79 cm in girls in a subsequent time period of 2 years (b = 0.015; Table 3).

This is the first study exploring whether higher levels of body fatness were associated with subsequent changes in screen time in adolescent boys and girls. No evidence was found for a prospective association between indicators of body fatness and screen time, which is in contrast to the findings of Ekelund et al 27 in adults. One explanation for these contrasting results could be the follow-up duration in the study of Ekelund et al 27 was 5.6 years, whereas in our study this was 1 year. This could indicate that body fatness only predicts sedentary behaviour over a relatively long period of time. Another explanation could be a difference in the type of measurement of time spent sedentary. Ekelund et al 27 calculated sedentary time as all minutes below the flex heart rate, which was defined as the mean of the highest resting heart rate and the lowest heart rate while exercising. Using this measure of sedentary time Ekelund et al 27 included all types of sedentary behaviours, whereas in the current study only self reported screen time (i.e. TV viewing and computer use) was included. These different measures of sedentary time could lead to differences in the total amount of screen time, resulting in differences in the associations between body fatness and screen time.

The present study showed that TV viewing time predicted changes in two indicators of body fatness (BMI and HC) in adolescent boys, but not in girls. In addition, we showed that computer time predicted changes in two indicators of body fatness (BMI and skinfold) in adolescent boys and one indicator in girls (skinfold).

Our finding that TV viewing time did not predict changes in body fatness among adolescent girls is in agreement with the results of two high quality studies included in a recent review comprising only prospective studies 73233. In addition, we showed that TV viewing time was related to changes in BMI and HC, but not WC and skinfold thickness in adolescent boys. Skinfold thickness is demonstrated to be a better predictor of body fatness in youth than BMI 34. Therefore, we conclude that our results did not convincingly demonstrate that TV viewing time predicts body fatness in adolescent boys. This is in contrast to the findings of three high quality studies included in the recent review of Chinapaw et al 7123235.

The lack of a convincing positive association between TV viewing time and indicators of body fatness in adolescent boys and girls in the present study could indicate that this association is not linear. However, when data for TV viewing were categorized using cut-offs of 2 and 4 hours/day, such a potential prospective association between TV viewing time and indicators of body fatness could not be confirmed.

The present study demonstrated that computer time predicted changes in skinfold thickness among adolescent boys and girls, and changes in BMI among adolescent boys. The differences in the association of computer time with BMI and skinfold thickness are in line with the finding of Nooyens et al 34, demonstrating that skinfold thickness is a better predictor of body fatness in youth than BMI. Based on the large regression coefficients for skinfold thickness and the consistent findings for boys and girls, our findings indicate that computer time predicts increases in body fatness (as measured by skinfold) in adolescents.

To summarize, our results did not convincingly demonstrate that TV viewing time predicts increases in body fatness, whereas computer time did predict increases in body fatness in adolescents. A positive association between sedentary time and body fatness among adolescent boys and girls could therefore only partly be confirmed.

Strengths include the prospective design and the sample size of the study population. Moreover, the statistics used (autoregressive models) in the current study enabled research into prediction rather than mere association. A study limitation is the reliance on self-reported screen time, which is sensitive to recall bias and socially desirable answers. Although brief self-reported questionnaires have shown to be adequate for group comparisons regarding TV viewing and computer use 36, a misclassification of the amount of screen time could have masked the association between screen time and body fatness in the present study. In addition, adolescents with higher levels of body fatness may have underreported the actual screen time, as also has been observed for food intake. Indeed in the study of Slootmaker et al 37 it was observed that adolescents overrated their physical activity level. Accelerometry is increasingly used as an objective measure of sedentary behaviour; however, it cannot be used to distinguish between TV time and computer time. Another limitation of the present study is the number of statistical tests performed, which implies that the results should be carefully interpreted. Finally, some relevant factors were not included in our statistical analysis (e.g. diet, cigarette smoking, alcohol use), which might have attenuated the associations found.

Adjustments for aerobic fitness were made in order to investigate the prospective association between screen time and body fatness independent of PA. Although a direct and objective measure of PA would have been a better predictor of the time spent in PA, aerobic fitness has shown to be related to time spent in PA in adolescents 3839.