Data quality and management
All study data are stored in a password-protected customised
database, hosted by the University of Birmingham.
Paper-based information is held in locked filing cabinets
in the study office. For all data entry, a minimum 10%
sample is checked to monitor error rates. Potential errors
are identified and checked using a range of techniques.
These include clinical and data-driven range checks, and
cross validation between variables where a correlation
would be expected and when the same information is
obtained from different sources.
Planned statistical analysis
Trial analyses will be undertaken after the second followup
measures are completed and there will be no interim
analyses.
The baseline pupil (including sex, ethnicity, deprivation
[based on IMD scores derived from home postcode]
anthropometric measures, dietary intake, physical activity
levels, psychological variables) and school level characteristics
(school size, ethnic mix of pupils and % eligible for
FSM) will be summarised by control and intervention
arms, using numbers and proportions, means and standard
deviations or medians and inter-quartile ranges.
Analyses of outcomes will be by intention to treat. As
randomisation will be at the school (cluster) level, appropriate
statistical methods to account for the clustering
within schools (detailed below) will be used in the analysis.
Analysis of outcomes will be for both 3- and 18-month
follow-up stages.
We will use a mixed model ANCOVA with follow-up
outcome values as the dependent variable and baseline
values and treatment arm as the independent variables,
to investigate effectiveness. These will be fitted using mixed
models in STATA to allow for clustering. We will allow for
clustering at the school level and explore the possibility
of allowing for an additional level of clustering at the
class level.
The primary analysis will be adjusted for baseline values
for all outcomes. Secondary analysis will additionally adjust
for pre-specified baseline school and child level covariates.
These will include school level factors which were
used in the randomisation (school size, % pupils eligible
for free school meals, ethnic mix of pupils) and pupil level
factors (sex, baseline BMI z-score, ethnicity, deprivation from home postcode, baseline total energy intake and
baseline total physical activity). We will not adjust for age
as all children will be of a very similar age. We will adjust
at the school and pupil level for both ethnicity and
deprivation as the school population is expected to differ
from the consented study population.
Outcomes are either binary (e.g. non-overweight vs.
overweight), or continuous (e.g. BMI z-score or energy
expenditure), and therefore either log or linear link functions
will be used, with transformations where appropriate
to accommodate any non-normality. All model
assumptions will be checked. We will report both relative
and absolute treatment effects.
The primary analysis will be a complete case analysis.
However, missing data will be reported and associations
between outcomes explored. Depending on the nature of
these associations and the extent of the missing data,
sensitivity analysis will be undertaken using multipleimputation
techniques.
The primary outcome and primary sub-group comparisons
at both time points will be considered significant
at the 5% level (and so 95% CIs reported); whereas other
secondary outcomes will be deemed significant at the 1%
level (and so 99% CIs reported). This difference in levels
of significance, gives more weight to the primary outcomes.