Statistical methods
Age at AD debut (≤/> 1 yr), aeroallergen sensitization (yes/no) and allergic rhinitis (yes/no) were categorized as binary variables. Severity of AD by means of SCORAD was included in the models as a continuous variable including the highest SCORAD level measured in the child (early-onset from 0-1 year and late-onset from 1-6 years).
Associations between AD (binary age at onset or continuous SCORAD) and aeroallergen sensitization and allergic rhinitis were analyzed by logistic regression analyses expressing results as odds ratios (OR) with 95% confidence intervals (CI). Furthermore, we used a logistic regression general estimating equations (GEE) model to compute the overall OR for allergy endpoints using compiled data from both time points and accounting for repeated measures. All results are calculated as crude and adjusted for potential confounders including sex, older siblings, maternal allergic rhinitis, paternal allergic rhinitis, breastfeeding and smoking during 3rdtrimester.
The analysis of whether the association between AD and later sensitization/allergic rhinitis depended on the AD-debut age (yes/no) was analyzed using a logistic regression model where AD status is a three-level variable (early, late, never). The analysis of whether the association between AD and later sensitization/allergic rhinitis depended on the AD-debut age (SCORAD) was based on two models using the quantitative measure of SCORAD as a predictor; one based on early debut, and one based on late debut. As all non-AD children had a SCORAD of 0, these are excluded from the analysis. From these models the slopes are compared using a Wald’s tests associated with H0: b(early) = b(late) which hence reveal whether the association between severity and AR/sIgE is different between early and late AD-debuts.
In the analyses of early-onset AD (debut≤1yr, yes/no) we used all the other children as a control group, whereas in the analyses of late-onset AD (debut 1-6 years, yes/no) we excluded the children with an early-onset AD debut from the control group.
We additionally stratified for FLG mutation. Further, effect modification by FLG was evaluated by second order interaction models by adding cross-products to the models.
All statistical analyses were performed with the statistical software package R version 4.0.2 and RStudio version 1.1.442 (RStudio Inc, Boston, MA, USA). Missing observations were treated as missing data. P-values <0.05 were considered statistically significant.