Résumé
Background: Phenotypic characteristics of patients with eosinophilic and noneosinophilic asthma are not well characterized in global, real-life severe asthma cohorts.Research question: What is the prevalence of eosinophilic and noneosinophilic phenotypes in the population with severe asthma, and can these phenotypes be differentiated by clinical and biomarker variables?Study design and methods: This was an historical registry study. Adult patients with severe asthma and available blood eosinophil count (BEC) from 11 countries enrolled in the International Severe Asthma Registry (January 1, 2015-September 30, 2019) were categorized according to likelihood of eosinophilic phenotype using a predefined gradient eosinophilic algorithm based on highest BEC, long-term oral corticosteroid use, elevated fractional exhaled nitric oxide, nasal polyps, and adult-onset asthma. Demographic and clinical characteristics were defined at baseline (ie, 1 year before or closest to date of BEC).Results: One thousand seven hundred sixteen patients with prospective data were included; 83.8% were identified as most likely (grade 3), 8.3% were identified as likely (grade 2), and 6.3% identified as least likely (grade 1) to have an eosinophilic phenotype, and 1.6% of patients showed a noneosinophilic phenotype (grade 0). Eosinophilic phenotype patients (ie, grades 2 or 3) showed later asthma onset (29.1 years vs 6.7 years; P < .001) and worse lung function (postbronchodilator % predicted FEV1, 76.1% vs 89.3%; P = .027) than those with a noneosinophilic phenotype. Patients with noneosinophilic phenotypes were more likely to be women (81.5% vs 62.9%; P = .047), to have eczema (20.8% vs 8.5%; P = .003), and to use anti-IgE (32.1% vs 13.4%; P = .004) and leukotriene receptor antagonists (50.0% vs 28.0%; P = .011) add-on therapy.Interpretation: According to this multicomponent, consensus-driven, and evidence-based eosinophil gradient algorithm (using variables readily accessible in real life), the severe asthma eosinophilic phenotype was more prevalent than previously identified and was phenotypically distinct. This pragmatic gradient algorithm uses variables readily accessible in primary and specialist care, addressing inherent issues of phenotype heterogeneity and phenotype instability. Identification of treatable traits across phenotypes should improve therapeutic precision.