Research uses machine learning to highlight predictors of atopic dermatitis flare

Recent machine learning findings suggest that duration, frequency, and severity of atopic dermatitis flares predict future severity of atopic dermatitis, and these factors are also linked to poor quality of life.¹

These data resulted from studies conducted to test the predictability of flares and the severity of atopic dermatitis, in addition to quantifying predictive variable values. This analysis was led by investigators such as Mia-Louise Nielsen, MSC, PhD, and others in the dermatology department at the University Hospital of Copenhagen, Bispebelg, Copenhagen University Hospital, Denmark.

Although systemic drugs have been shown to be effective in treating atopic dermatitis, Nielsen et al. emphasized that many affected individuals remain untreated due to inadequate drug outcomes and poor satisfaction.²

“In particular, there are mild to moderate illnesses, but there are potential risks of sustained flaring and unstable courses of illness,” Nielsen and co-authors wrote.^1,3 “This study used statistical methods and predictive machine learning models to validate predictability and identify predictors of flare and disease severity.”

Trial design and survey results

The research team used a comprehensive dataset drawn from a Danish skin cohort. Its contents highlight records of Danish adults with atopic dermatitis. These data also include information about the severity of the disease and patterns related to flare patterns. The team examined these findings to examine the link between the frequency of flares in such patients in 2022 and the 2023 patient self-report severity.

Nielsen and colleagues used the quantile regression model to assess this relationship. They implemented a boosted random forest algorithm with the aim of identifying important predictors of both annual flare frequency and atopic dermatitis severity. Individuals who had been provided with a dermatologist-filled diagnosis in adulthood were deemed eligible for inclusion in investigator analysis.

At 2, survey responses were collected by Nielsen et al.: January 14th to February 6th, 2022, and January 31st, 2023 – January 31st, 2023. Their research was designed to capture general data for patients, such as age, gender, and comorbidities, and to capture specific information related to Atopic dermatitis. The research team concluded a study that evaluated 878 participants. [IQR]39.0-59.0 years).

Of the 878 trial subjects, 26 were reported by the team to have no atopic dermatitis flares in 2022.¹ Additionally, 405 reported 1-5 disease flares, 169 reported 6-10 flares, and 278 reported 10 or more flares in the same year. In investigator quantile regression results, the data provided an important link between the flare count in 2022 and the severity outcomes of several patient-reported reports in 2023.

Even after adjusting for baseline scores for patient-oriented scoring for atopic dermatitis (PO-scorad), flare frequencies remained significantly linked to scores for patient-oriented eczema measurements (poems) and dermatology quality of life index (DLQI) scores.¹

Furthermore, Nielsen and colleagues' machine learning analyses showed that flare properties, i.e. duration, frequency, and severity, were later considered to be among the most powerful predictors of the overall severity of a patient's disease. They also concluded that the severity of atopic dermatitis itself emerged as an important factor predicting the frequency at which subjects burn each year.

Overall, this study demonstrated that not only the duration and severity of the disease, but also the number of self-reported flares in the previous year predicted the severity of future atopic dermatitis and the frequency of future flares.

“While there is no established consensus regarding some too many flares, the current findings suggest that flares may serve as an early indicator of disease progression or inadequate disease control, highlighting that flares may be relevant to clinical decision-making to support optimal treatment strategies,” the investigator concluded.¹

reference

1. Predictors of flare and disease severity in patients with atopic dermatitis using Nielsen M, Nymand LK, Pena AD, and other machine learning. Jama dermatol. Released online on July 16th, 2025. doi: 10.1001/Jamadermatol.2025.2073.

2. Charman CR, Ben AJ, and HC Williams. Patient-oriented eczema measurement: Development and initial verification of new tools for measuring the severity of atopic eczema from a patient's perspective. Arch dermatol. 2004; 140 (12): 1513-1519. doi: 10.1001/archderm.140.12.1513.

3. Wei W, Anderson P, Gadkari A, et al. The extent and outcome of inadequate disease control in adults with a history of moderate to severe atopic dermatitis. J Dermatore. 2018; 45(2): 150-157. doi: 10.1111/1346-8138.14116.