Our cross-sectional analyses showed clear and consistent associations between physical activity and body composition (both BMI and percent body fat). The odds of being overweight or overfat were substantially increased among girls who had low levels of physical activity (15th percentile) in comparison with girls who were active (85th percentile), and odds ratios ranged from 1.59 to 3.82. The associations tended to be stronger in the sixth grade than in the eighth grade, and, as expected, when data from the two grades were combined the estimates were intermediate.
The results from our cross-sectional analyses are consistent with previous work done by Lohman et al. (21
) using data from the sixth-grade TAAG girls. The objective of that analysis was to determine the nature of the relation (linear, curvilinear) between body composition (the dependent variable) and physical activity (the independent variable). Lohman et al. found a linear association, in which a 1-standard-deviation increment in fat mass index (fat mass/height squared) was associated with an MVPA level that was lower by 22.8 MET-weighted minutes (21
). That study did not include data from eighth-grade girls or longitudinal analyses.
Here, the sizes of the effects observed in the cross-sectional analyses were larger than those observed in the longitudinal analyses. For example, for each minute of difference in MVPA between girls in the cross-sectional analysis (with sixth- and eighth-grade data combined as in ), the effect on percent body fat was over four times larger than that associated with a one-unit change within girls in the longitudinal analysis (0.1249 vs. 0.0260). In addition, the range of physical activity levels between girls was larger than the range of changes in physical activity within girls. Between the 15th and the 85th percentiles, the difference in MVPA was 21.9 minutes (31.8 (85th) minus 13.7 (15th)) per day in sixth- and eighth-grade girls combined, whereas the difference in change in physical activity from sixth to eighth grade within girls was approximately half that size, at 10.7 minutes (6.2 (85th) minus -4.5 (15th)).
Although the coefficients were small, associations were detected between changes in weighted and unweighted MVPA and percent body fat. No statistically significant associations were found in the continuous analysis of BMI or in the incidence analysis in which we examined a number of different relevant outcomes, although results generally pointed in the hypothesized direction. The fact that we did find a small, statistically significant effect in the most statistically powerful and precise analysis (with continuous percent body fat) may indicate that changes in the incidence of overweight and/or high body fat (≥32 percent) might be found if follow-up were longer or if the range of changes in physical activity were larger. Here, with 2 years of follow-up, there were only 44 new cases of overweight85 and only 31 new cases of overweight95.
Similarly to this study, our work in the Pathways Study (9
) showed longitudinal associations between physical activity and percent body fat but not with BMI. In three other studies of accelerometry and body composition in youth (7
), a dichotomy in the results found with BMI as opposed to percent body fat was not apparent. In the Pathways Study (9
) and in the study by Ekelund et al. (11
), associations between physical activity and body composition were seen only in youth who were normal-weight at baseline, indicating an interaction between physical activity and baseline weight status. No such interaction was found in TAAG (data not shown).
To our knowledge, this is the first longitudinal study to examine associations between objectively measured physical activity and body composition among girls in their middle school years. Previous studies that included older girls and young women (4
) used self-report question-naires to examine these associations. Studies using only self-reported measures of physical activity are subject to imprecision and bias. Generally, physical activity assessments obtained by self-report produce correlations with reference methods in the range of 0.2-0.6 (6
The greatest strength of this study was the measurement of physical activity using 6 days of accelerometry in a large,diverse sample of girls. To our knowledge, our sample included the largest number of adolescent girls ever examined longitudinally using accelerometry. Earlier work by TAAG investigators showed that 85 percent of the variance in energy expenditure was explained by accelerometry (25
). Other strengths of our study were that age- and gender-appropriate cutoffs were used to define MVPA (18
) and that missing accelerometry data were imputed (19
One important limitation of this work is that we did not have a measure of pubertal stage. It is well known that puberty is associated with a substantial increase in percent body fat among girls. Associations of pubertal status with physical activity could have resulted in confounding of our analyses. In addition, we estimated percent body fat using anthropometry rather than more precise measures such as dual-energy x-ray absorptiometry. Although our preliminary work indicated high validity of the method used (14
), more precise measures are always preferred.
Finally, a limitation of this work was that energy intake was not assessed. In the current study, the only feasible method of assessing energy intake would have been self-reporting, which is expensive and is subject to measurement error and bias related to obesity status (26
). Although accurate measures of energy intake would have enriched our ability to interpret the results, lack of those measures does not negate the relevance of this study to public health. It was not our purpose to prove that, at the same level of energy intake, greater energy expenditure is associated with declines in obesity measures. Other types of more tightly controlled studies have made this point very clear (28
). Rather, the purpose here was to show the levels of physical activity associated with cross-sectional differences and longitudinal changes in body composition in a free-living population of middle-school girls. Nevertheless, it is possible that associations between physical activity and body composition would have been stronger if energy intake had been controlled.
Decreases in energy expenditure through physical activity have been implicated as an important cause of the growing prevalence of obesity (30
). Accelerometry provides a method of assessing the intensity and duration of physical activity, and thus it can provide important information on the type and amount of activity needed to prevent obesity. This work indicates that large samples, long periods of follow-up, multiple assessments, and careful measurement of confounders may be needed to detect associations between changes in activity and changes in body composition in free-living populations. More studies are needed to understand the amount of physical activity needed to prevent the development of obesity and to provide evidence-based physical activity recommendations to the public.