Abstract
To the Editor, Food-induced anaphylaxis (FIA) is a severe, potentially lifethreatening allergic reaction with highly heterogeneous clinical features. 1 Clinical predictors of FIA severity are still not available. 2 The possibility to predict patients at risk of severe reactions would be key to reduce anaphylaxis morbidity and mortality. In this context, clustering analysis has been used to describe clinical phenotypes of different allergic diseases such as atopic dermatitis, asthma and eosinophilic esophagitis.</p><p>We aimed to describe the clinical heterogeneity of a cohort of children and adolescents with FIA using an unsupervised statistical cluster analysis approach and identify potential clinical predictors of severity. A retrospective study was conducted on 137 patients (60% males) with FIA (27% with more than one FIA) after presenting at the Paediatric Clinic in Pavia, Italy, between April 2001 and October 2021. The study was approved by the local Ethical Committee (protocol number 7969/22). The diagnosis of anaphylaxis was established according to the guidelines of the European Academy of Allergy and Clinical Immunology. A well-trained physician-reviewed medical records to collect demographic data (date of birth, age at diagnosis, gender and ethnicity), current comorbidities (asthma, allergic rhinitis, atopic dermatitis and eosinophilic gastrointestinal disorders), food triggers, cofactors (infections, drugs and physical exercise), clinical manifestations (cutaneous, gastrointestinal, respiratory, cardiovascular and neurological symptoms), severity grading according to Sampson's criteria and treatments of anaphylaxis. The complete-linkage clustering algorithm used 15 clinical (allergic comorbidities, symptoms, adrenaline injection and total serum IgE) and aetiological (food triggers) variables as input (Table 1).</p><p>Gower's general dissimilarity coefficient was used to compute the distance matrix. A hierarchy of partitions was determined (2-6 groups), and the one with the largest Silhouette index was chosen. We used Fisher's exact test (categorical variables) or the Mann-Whitney-Wilcoxon rank sum test (numerical variables) to perform paired comparisons between the clusters. Holm's method was used to adjust the p values for multiple comparisons.