Machine learning-derived asthma phenotypes in a Swedish adult population-representative sample

<bold>Background:</bold> Asthma is a heterogeneous airway disease characterized by numerous phenotypes. Identification of clinically meaningful phenotypes, using advanced machined based methods and large population-based sample is paramount for optimizing asthma management. <bold>Objectives:</bold> To use a novel machine learning approach to determine asthma phenotypes based on a broad range of characteristics in a representative sample of Swedish adults. <bold>Methods:</bold> Using deep learning to analyze data from the 1,895 individuals included from the clinical sample of West Sweden Asthma Study (a population-based adult cohort), asthma phenotypes were identified based on a wide range of variables, including demographics, risk factors, and clinical profiles. <bold>Results:</bold> Four distinct phenotypes were identified: (1) Burdensome late-onset, non-atopic asthma with smoking (n=458, 24.2%); (2) female-dominated early adult-onset asthma (n=545, 28.7%); (3) adult-onset asthma with high inflammation (n=358, 18.9%); and (4) early-onset, mild, atopic asthma (n=534, 28.2%). The phenotypes differed with respect to asthma onset age, severity, triggers, symptoms, demographics, risk factors, pulmonary function, and markers of inflammation. <bold>Conclusion:</bold> Our findings – partly overlapping with previously described phenotypes – indicate that distinct asthma phenotypes exist in the adult population, with clear differences in a multitude of dimensions. Future research will further elucidate underlying immunological mechanisms, risk factors, comorbidities, and the broader health and societal impacts.