Supervised multivariate analysis
Principal component analysis (PCA) was performed with the c-IgE values
for the preschool children. PC1 accounted for 20.3% of the variance and
PC2 for 11.7%. Overall, PCA did not allow differentiation between NSRW
and SRW. Random forest analysis was then performed. Similarly, it did
not allow discrimination between NSRW and SRW, with an estimated
out-of-bag error rate of 43.1% and a ROC AUC of 0.56 (Sup Fig 3).
Among school-age children, PCA, with PC1 explaining 24.1% of the
variance and PC2 10.4%, did not allow differentiation between NSA and
SA. Similarly, random forest analysis did not allow discrimination
between NSA and SA, with an estimated out of bag error rate of 33.9%
and a ROC AUC of 0.53 (Sup Fig 4).