Page 71 - Go4it
P. 71

Sample size
The sample size calculation is based on the primary outcome of the trial i.e. BMIsds: the number of participants needed to detect a difference in BMIsds of 0.29 (10%) between the intervention and control group after 18 months with a standard deviation of 0.5 is 43 subjects per group with an alpha of .05 and a power of .80. A sample of 108 persons (n=54 per group) was required taking into account a dropout rate of 25%.
Statistical analysis
Baseline characteristics were analysed by t-tests for continuous variables and Chi-square tests for categorical variables. Scoring and substitution of missing values was performed according to existing manuals. In the case of 50% or less missing per subscale, substitution by the mean was used (24,25,28). Subscales with higher amounts of missing values were considered missing resulting in a varying number of participants included in the analyses. Group comparisons were performed according the intention-to-treat principle whereby all subjects were analysed in the group to which they were initially assigned. Linear mixed models were applied to assess the effect of the intervention over time. A random intercept and a random slope with time were assumed. Age-, sex- and, ethnicity adjusted analyses were performed with intervention as the categorical variable and time as the continuous variable; an interaction term for intervention and time was also included. B-coefficients (between group difference), 95% confidence intervals and p-values were calculated. This approach has increased statistical power as it accounts for within-person correlations over time and allows different numbers of assessments. All assessments, including baseline, were taken into account. A p-value of <0.05 was considered statistically significant. Effect modification by sex, age and ethnicity was checked by adding an interaction term between group allocation and the potential moderator. For effect modification, a p-value of <0.10 was considered statistically significant. Finally, effect size estimations (Cohen’s d) were calculated in order to decide whether statistical differences were clinically relevant. Effect sizes relate to the difference in mean scores between the dispersion of the scores: [mean baseline – mean follow-up]/pooled standard deviation (29). Following Cohen’s d effect size, d=0.2 was taken to indicate a small effect size, d=0.5 a moderate effect size, and d=0.8 a large effect size (30). All analyses were performed using SPSS software (version 18.0, 2009 SPSS Inc., Chicago, Illinois, USA).
Results
Subjects
Figure 1 shows the consort diagram for the Go4it trial. Of the 219 adolescents who were assessed for eligibility, 122 consented to the trial and were randomly
Quality of life
 71


























































































   69   70   71   72   73