Statistical analysis
To calculate participation rates, the denominator was the number of
adolescents and the numerator the number of core questionnaires returned
with at least one data symptom. For prevalence estimations, positive
answers to a specific symptom in the centre was divided by the number of
completed questionnaires. There was no data imputation.
All variables included in the GAN environmental questionnaire were used
as independent variables in logistic regression analyses (uni and
multivariate) in which the three eczema markers (current symptoms,
severe symptoms, and current eczema) were the dependent variables. Those
analyses were performed within each centre. As the number of cases of
severe symptoms was very low, logistic regression analyses were not
performed. Factors which showed significant (p<0.05) values of
adjusted odds ratios in the multivariate logistic regression analyses in
at least one centre were subsequently meta-analysed (random effects)
including the results of the six centres. A forest plot was built,
including the pooled effect together with the 95%conficence interval
and prediction interval. Measures of heterogeneity, such as Q and
I2 were also calculated.
To test whether the presence of current wheeze modified the associations
between the environmental factors and the eczema markers on the whole
sample of adolescents, a multilevel mixed effects logistic regression
model was performed, including the same variables as in the within
centre logistic regressions; and using the individual as the first level
and the centre at the second. As previously and due to the low number of
cases, no analysis was made for symptoms of severe eczema.
Most statistical calculations were made using Stata SEĀ® v18 software
package (Stata corp., College Station, TX, USA) except meta-analyses
that were carried out using Comprehensive Meta-Analysis (CMA) V4.0
software package (Biostat, Englewood, NJ, USA).