Statistical Analysis
The primary outcome was total difficulties score for SDQ in each domain:
hyperactivity, emotional symptoms, conduct problems, and peer problems.
Each score can be interpreted as “normal,” “borderline,” and
“clinical,” on a scale from the lowest to highest score. Matsuishi
proposed the total difficulties score classification as “normal,”
“borderline,” and “clinical,” corresponding to 0-12, 13-15, and
16-40 points, respectively. Matsuishi sets the ranges for the domains of
emotional problems, conduct problems, and peer problems as 0-3 for
“normal,” 4 for “borderline,” and 5-10 for “clinical”. The ranges
for the hyperactivity domain were defined as 0-5 for “normal,” 6 for
“borderline,” and 7-10 for “clinical”. In cases where the outcomes
were divided into binary results corresponding to non-occurrence or
occurrence, it is generally recommended that “borderline” results are
included in occurrence when high sensitivity is desired(normal vs.
borderline and clinical) , and excluded when high specificity is
desired(normal and borderline vs. clinical). 20 We
have used the recommended borderline and clinical cutoff scores as
criteria for dichotomizing primary outcome variables.
Exposure variables used in this study were history of allergic symptoms
such as wheezing, eczema, and nose symptoms since birth. The potential
confounding variables were age and sex (demographic characteristics),
PSI-SF scores (parents’ psychological aspect), doctor’s diagnosis, and
family history of allergic diseases such as atopic dermatitis, food
allergy, asthma, rhinoconjunctivitis, and hay fever, as well as other
items of the ISAAC.
We performed logistic regression analyses to estimate crude and adjusted
odds ratios (OR), accounting for propensity scores yielded by all
confounding variables. The items used to determine propensity scores are
shown in Supplementary Table 1. When integrating all confounders into
one variable as a propensity score and determining the propensity score
variables for each exposure, the variables relevant to the simultaneous
factors inducing collinearity, such as a history of diseases or period
prevalence, were excluded. The analyses applied a complete case
analysis. As the online survey prevented responders from submitting
their forms with missing values, the dataset involved no missing values.
All statistical analyses were performed using SPSS version 22.0.