Statistical Methods
Normally-distributed continuous data (age, age at diagnosis, lung function measures) were expressed as mean and standard deviation, and not-normally distributed continuous data (eosinophil counts and total IgE values) as median and interquartile ranges (IQR). Group comparisons carried out using student´s t-test, Mann–Whitney U-test, ANOVA or Kruskal Wallis test as appropriate for the continuous, and the chi-square test or Fisher test for categorical variables.
Latent class analysis (LCA) was performed to identify patterns of clinical clusters using 17 variables (allergic conjunctivitis, eczema, asthma, family history of asthma, family history of allergic rhinitis, skin sensitization to 8 common allergens, tonsillectomy, adenoidectomy). Starting with a latent model including 2 classes, we compared models with increasing numbers of classes using the Bayesian information criterion (BIC). The expectation maximization algorithm was used to estimate relevant parameters, with 100 000 iterations and 500 replications. The optimal number of classes was selected based on the lowest BIC, and interpretability.
Association of LCA-driven classes with clinical outcomes was examined using regression models adjusted for potential confounders, including gender and maternal history of allergic rhinitis. The odds ratio (OR) and 95% confidence interval (CI) were reported. Analyses were performed using SPSS Statistics v21.0 (IBM, Chicago, IL, USA), Stata 16 (StataCorp, College Station, TX), MPlus 8, and R (http://www.r−project.org/).