To the Editor,
The article by Tuten Dal S et al on house dust mite molecular allergen
(MA) sensitization among children appear to be consistent with previous
studies in that: (1) sensitization levels and number of individual
sensitizations increase with age; (2) certain MA mono-sensitizations are
clinically relevant (e.g., Der p 20 and scabies, Der p 23
monosensitization) and (3) allergic sensitization to house dust mites
can cause development of other allergies [1-3].
However, the hierarchical cluster analysis (HCA) employed in the study
by Tuten Dal S et al were unable to address if a particular MA was
associated with increased risk of development of allergic rhinitis, or
if another MA was associated with decreased risk of atopic dermatitis.
Hierarchical cluster analysis, although easy to employ and understand,
relies on several arbitrary decisions (such as distance metric or
linkage criteria, not provided by Tuten Dal S et al) that allocate the
data into clusters and therefore not only prone to misinterpretation but
also often a poor solution. In contrast, Bayesian Network Analysis such
as latent class analysis provide separate probabilistic connections
between individual MA and patient profiles thereby building the Bayesian
directed acyclic graph and also a network [4, 5].
Considerable use of latent class analysis discussed in papers by Yuriev
S et al (Ukraine, root node set to Der p 23 as it remained high in all
age groups) and Hou X et al (China, whole allergen extracts) show the
importance of understanding regional sensitization profiles and age
groups before considering allergen immunotherapy products [2, 6].
Der p 7 was associated with development of allergic rhinitis while
sensitization to Der f 2, Der p 2 and Der p 23 increased risk of
developing atopic dermatitis. In contrast, Der p 10 sensitization
reduced risk of atopic dermatitis [2]. Similarly, Hou X et al showed
that patients in pollen sensitization clusters have increased risks of
dermatitis and allergic conjunctivitis in contrast to those in Class 1
who were highly allergic to house dust mites and low rates of other
allergens [4].
Such analysis is not possible using HCA and large datasets that are
generated using microarray platforms should ideally be analysed using
Bayesian statistics that employ prior/posterior probabilities and
well-equipped to handle missing data.
Sujoy Khan FRCP FRCPath
Consultant Immunologist
Hull University Teaching Hospitals NHS Trust
Castle Hill Hospital, Castle Road
Cottingham, HU16 5JQ
Conflicts of interest: None declared.
Financial Support: None