Identification of rhinitis phenotypes using an unsupervised approach in the Constances cohort

Unsupervised approach has been scarcely used for rhinitis in adults in population-based studies. We aimed to identify rhinitis phenotypes using this approach in the French population-based cohort Constances. Among participants answering the 2014 questionnaire, we included those with current rhinitis i.e. reporting sneezing, runny or blocked nose out a cold or the flu, in the last 12 months. Based on 25 variables referring to rhinitis characteristics (persistence, severity, triggers, seasonality, treatments) and co-morbidities, a dimension reduction with Factor Analysis of Mixed Data followed by the K-means algorithm were used to identify clusters. Among 5516 participants with current rhinitis (50 years old, 57% women, 20% ever-asthmatics), three clusters were identified: cluster 1 (C1) n=2586, 47%, C2 n=2379, 43% and C3 n=551, 10%. C1 was characterized by rhinitis with few identified allergic triggers (dust or dust mites, animals and pollens reported by less than 10% of participants and 53% of participants did not know what triggered their symptoms). In C2 and C3, more than 95% of participants reported ever nasal allergies and 100% of participants from C3 had severe rhinitis. Gradual increase was observed from C1 to C3 for asthma co-morbidity (C1=7%, C2=31%, C3=38%), combined medication (oral antihistamines and intranasal corticosteroids) (C1=7%, C2=37%, C3=56%) and eosinophils count (109/L) (C1=181, C2=208, C3=231), all p-trend