School level data
Data on participating schools are collected through a
questionnaire administered to head teachers or a nominated
representative. Information requested includes details
on school food and physical activity policies and
any relevant initiatives or programmes delivered through
school.
Process evaluation
Implementation fidelity is assessed throughout the intervention
year using a range of methods including direct
observation, logbooks, parent and school staff questionnaires,
research staff experiences and qualitative evaluation.
The methods are described in detail elsewhere [17].
Justification of sample size
Sample size calculation is based on the primary outcome
(BMI z-score). Further calculations were also performed
to estimate power for the secondary outcome of percentage
of children overweight or obese. Planned analysis of
the WAVES study will compare outcomes for control
and intervention schools at follow-up times, adjusting
for baseline measurements. Therefore power calculations
undertaken were based on repeated measures methods
using estimates of correlation between before and after
measurements. A modified version of the design effect
[35] was used to estimate sample size and accommodate
varying cluster sizes (using the estimated: mean cluster
size (n = 25; SD = 23). For the primary outcome of BMI
z-score, a follow-up sample size of 1000 children split
across 50 schools gives the study greater than 90% power
to detect a difference of 0.25 BMI z-score between intervention
and comparator groups (equivalent to approximately
0.5 kg body weight for a 7-year old child) under all
likely estimates of the intraclass correlation coefficient
(ICC = 0 to 0.04, estimated correlation between before and
after measures = 0.9 and estimated dropout rate = 20%). A
change of 0.25 in BMI z-score has been shown to be associated
with clinically detectable benefits in obese adolescents
[36] and longitudinal studies demonstrate a linear
relationship between BMI z-score in children as young as
7 and heart disease events in adulthood [4]. Under more
conservative estimates for the ICC, this sample size would
provide more than 80% power to detect a 0.125 difference
in BMI z-score. A BMI z-score difference as low as 0.125
is the primary outcome of choice for other childhood
obesity prevention trials [37]. Allowing for school dropout
(~8%), 54 schools were therefore invited to take part.
For the secondary outcome of percentage overweight/
obese, this sample size (with an estimated correlation of
before and after measures of 0.7 and an ICC of between
0 and 0.02) provides greater than 80% power to detect a
difference in the change of proportion of children who
are overweight/obese from baseline to follow-up in control
compared to intervention schools of about 7% (exact
value depends on baseline values). All power calculations
were carried out in STATA using the clustersampsi function
[38].