Predictors of Flares and Disease Severity in Patients With Atopic Dermatitis Using Machine Learning

The disease course of atopic dermatitis (AD) is characterized by fluctuations and frequent flares, contributing to the disease burden and impairment in life quality. However, flares are not necessarily considered in severity classifications and clinical treatment decisions.

To validate the predictability of flares and disease severity in patients with AD and quantify the importance of predictors.

Using the Danish Skin Cohort, a large population of patients with AD from Denmark with data on disease severity and flare patterns, quantile regression models were conducted to investigate the association between the number of flares reported in 2022 and patient-reported severity measures reported in 2023. Additionally, boosted random forests were used to explore predictors of both annual flares and disease severity. Analyses were conducted from January to December of 2024.

Severity of AD as well as frequency, duration, and severity of flares were the main variables under consideration.

This study included 878 patients with AD (median [IQR] age, 49.0 [39.0-59.0] years), with 26 reporting 0 yearly flares, 405 reporting 1 to 5 yearly flares, 169 reporting 6 to 10 yearly flares, and 278 patients reporting more than 10 yearly flares in 2022. From the quantile regression, the number of annual flares reported in 2022 was significantly associated with most patient-reported severity measures reported in 2023. When adjusting for the Patient-Oriented Scoring of Atopic Dermatitis score at baseline, the number of annual flares reported in 2022 was significantly associated with the Patient-Oriented Eczema Measure and Dermatology Life Quality Index. Using predictive machine learning models, flare severity, duration, and number were among the most important predictors of AD severity, while disease severity was among the strongest predictors of the number of annual flares.

This cohort study found that a higher number of flares was associated with lower quality of life and was identified as a predictor of more severe AD in the following year. These results highlight the relevance of flares in the assessment of severity or disease prognosis and suggest the need for a threshold for an acceptable number of flares in treatment decisions to achieve better disease control and improved quality of life for patients.