Background
Despite previous attempts to classify atopic dermatitis (AD) into subtypes (e.g. extrinsic vs. intrinsic), there is a need to better understand specific phenotypes in adulthood.

Objectives
To identify, using machine learning (ML), adult AD phenotypes.

Methods
We used unsupervised cluster analysis to identify AD phenotypes by analysing different responses to predetermined variables (age of disease onset, severity, itch and skin pain intensity, flare frequency, anatomical location, presence and/or severity of current comorbidities) in adults with AD from the Danish Skin Cohort.

Results
The unsupervised cluster analysis resulted in five clusters where AD severity most clearly differed. We classified them as ‘mild’, ‘mild-to-moderate’, ‘moderate’, ‘severe’ and ‘very severe’. The severity of multiple predetermined patient-reported outcomes was positively associated with AD, including an increased number of flare-ups and increased flare-up duration and disease severity. However, an increased severity of rhinitis and mental health burden was also found for the mild-to-moderate phenotype.

Conclusions
ML confirmed the use of disease severity for the categorization of phenotypes, and our cluster analysis provided novel detailed information about how flare patterns and duration are associated with AD disease severity.