Table 1.
Clinical parameters at inclusion in the entire cohort and according to cluster analysis.
Table 2.
Biological parameters at inclusion in the entire cohort and according to cluster analysis.
Fig 1.
Dendogram for the entire population (n = 214), obtained with a hierarchical bottom-up clustering.
Three clusters were apparent. This agglomerative approach begins with each subject as a separate cluster and merges them into successively larger clusters. By Ward’s linkage, samples were merged into larger clusters to minimize the within-cluster sum of squares. Cluster 1: “AD with low sensitization”, cluster 2: “AD with multiple sensitizations”, cluster 3: “AD with familial history of asthma”.
Table 3.
Parameters at the end of the follow-up (6 years).
Fig 2.
Classification tree for the entire cohort based on two variables.
Each subject was assigned to one of the 3 clusters using the tree; 97% of the subjects were assigned to the appropriate cluster. Tree performance values are given in the table. Cluster 1: “AD with low sensitization”, cluster 2: “AD with multiple sensitizations”, cluster 3: “AD with familial history of asthma”.
Fig 3.
Importance measure (permutation-based mean decrease accuracy) provided by the random forest analysis.
The values are not interpretable but the ranking is of interest since a high value of the importance measure is associated with a high predictive power. Sensitization is defined by specific IgE to one or more allergens ≥0.35 kUI/L. Multiple sensitizations were defined as at least two positive specific IgEs to allergens. Serum total IgE level expressed in kU/L, blood eosinophilia expressed in eosinophils/mm3