Random effects will account for within-participant correlation; fixed effects will be time and intervention group and the interaction of intervention and time.
A three-way rmANOVA, with factors hemisphere, intervention and time, showed significant interactions between all factors (P = .005, F1,17 = 10.6), a significant interaction of intervention× time (P = .03, F1,17 = 5.6) and of hemisphere× intervention (P = .011, F1,17 = 8.1).
We did secondary analyses by testing interactions of the intervention with sex, grade, and baseline body mass index, categorised into two groups above and below the median.
Appropriate interactions of year, intervention, ethnicity, age and hospital will be investigated with higher order interactions being removed if not important and the analysis being split where non-ignorable interactions exist.
When we found a significant interaction, the participants were divided into two groups according to the median of the baseline score of the outcome variable and the interactions of the intervention were analyzed again.
Third we assessed the interactions of the intervention according to the baseline scores for the outcome variables by adding interactions to the analysis of the main effect using GEE in order to examine in more detail the effects of the intervention program.
There was not a significant (p = 0.96) interaction effect of intervention × time.
The variables of main interest were group (intervention vs control group), time (pre-test vs follow-up), and the group x time interaction (effect of intervention).
More precise estimations of knowledge gain were obtained by the regression analysis controlling for between-group differences in ScoreT0 and allowing for an interaction effect of intervention (group) and ScoreT0.
Results of Model 1 revealed a statistically significant, albeit small, interaction effect of intervention over time (IRR 1.011, 95 % CI = 1.004 1.018, p = .003) on non-adherence indicating that the change in adherence over time differed between the study groups.
