Key Points
Question
What is the relationship between severity of atopic dermatitis and frequency of flares and what variables can potentially predict flares and severity?
Findings
In this cohort study of 878 patients with atopic dermatitis, frequency, duration, and severity of flares were predictive of future severity of atopic dermatitis and were associated with lower quality of life.
Meaning
Information on flares should be better integrated in the clinic and in treatment decisions.
Abstract
Importance
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.
Objective
To validate the predictability of flares and disease severity in patients with AD and quantify the importance of predictors.
Design, Setting, and Participants
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.
Main Outcomes and Measures
Severity of AD as well as frequency, duration, and severity of flares were the main variables under consideration.
Results
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.
Conclusions and Relevance
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.
This cohort study uses data from the Danish Skin Cohort to examine the predictability of flares and disease severity in patients with atopic dermatitis.
Introduction
Atopic dermatitis (AD) is a common skin disease characterized by eczematous skin lesions that are often itchy and painful. Disease onset typically occurs in early childhood and affects approximately 20% of children and up to 10% of adults.1,2 Atopic comorbidities, such as asthma and allergic rhinitis, are commonly observed among patients with AD, thereby adding to the burden of disease.3,4
Disease severity may be assessed by symptoms or clinical evaluation.5,6,7 Mild cases can typically be managed with topical therapy, while moderate to severe AD may warrant systemic treatment. Although systemic therapy has proven efficacious in the treatment of AD, many patients remain undertreated with inadequate treatment results and/or satisfaction.8,9,10,11
Despite flaring being frequently reported by patients with AD, it is not routinely considered in severity classifications or the development of treatment guidelines. This may potentially prevent some patients from qualifying for more potent treatments that could regain disease control or avoid disease progression. Especially patients with mild to moderate disease but persistent flaring and an unstable disease course may be at risk.12 This study used statistical methods and predictive machine learning models to validate the predictability and identify predictors of flares and disease severity.
Methods
This study is based on data from the Danish Skin Cohort collected through surveys at 2 different periods: January 14, 2022, through February 6, 2022, and January 3, 2023, to January 31, 2023. The data included patient-specific information, such as sex, current age, and comorbidities. In addition, it included disease-specific information, such as age at onset, number of flares within the past 12 months, and Dermatology Life Quality Index (DLQI). Furthermore, a range of patient-reported severity measures was surveyed, such as Patient-Oriented Scoring of Atopic Dermatitis (PO-SCORAD) and the Patient-Oriented Eczema Measure (POEM). Ethical approval was not obtained because it is not necessary for questionnaire studies conducted in Denmark. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Patients with a dermatologist-verified diagnosis of AD in adulthood were included. A flare was defined, according to the European Task Force Atopic Dermatitis and the European Academy of Dermatology and Venerology Eczema Task Force, as an acute worsening of symptoms requiring escalation of medication.13 Patients were informed of this definition and instructed to consider multiple consecutive days with acute worsening as a single flare. They were further advised to report a maximum number of 20 annual flares to avoid misconceptions and records of inordinately many flares.
Mild AD was defined by having PO-SCORAD less than or equal to 25 and POEM less than or equal to 7, while severe AD was defined as having PO-SCORAD greater than 50 or POEM greater than 17. The remaining patients were considered to have moderate AD.
Statistical Analysis
Patient baseline characteristics were reported as frequencies with percentages for categorical variables and medians with IQRs for numeric variables. Patients were stratified according to number of flares during the past 12 months (0, 1-5, 6-10, and >10).
Based on data collected in 2023, distributions of DLQI, PO-SCORAD, and POEM were visualized as densities stratified by the number of flares reported in 2022.
Sankey diagrams visualized how patients switched between flare strata from 2022 to 2023, and how they switched between groups combining severity and number of flares during the past year (mild AD with 0-5 flares, mild AD with >5 flares, moderate to severe AD with 0-5 flares, and moderate to severe AD with >5 flares).
Quantile regression models assessed the correlation between the number of flares reported in 2022 and various patient-reported outcomes from 2023. This validates the predictability of flares and identifies potential predictors. In addition, a logistic regression was conducted to estimate the odds of progressing from mild AD to moderate or severe AD in 2023 based on the number of flares reported in 2022. Crude and PO-SCORAD–adjusted results are presented.
Gradient-boosted decision trees were used to validate the predictability of the number of flares (regression) and disease severity (classification) and to quantify the importance of predictors contributing to the model. The main classification model was binary and classified patients in mild vs moderate to severe, whereas the second model was multiclass, classifying each severity group as mild, moderate, and severe separately. Shapley additive explanation14,15 values were used in a beeswarm plot to visualize variables in a sequence of importance for the prediction model, with the most important variables at the top. To ensure robust evaluation, nested cross-validation with 10 folds in the outer loop and 5 folds in the inner loop was used, and the performances were reported as the mean of the area under the receiver operating characteristics over all 10 folds. The dataset was first divided into 10 equal partitions in the outer loop, with 1 partition serving as the test data in each iteration. The inner loop split the training partitions into 5, with 1 functioning as validation set and the remaining 4 being used to train the model. The inner loop was essential for fine-tuning model parameters and selecting the optimal model configuration. The performance of the multiclass model was evaluated based on a 1-vs-1 approach, where each class was compared to all other classes individually.
All statistical tests were 2 sided and P values < .05 were considered statistically significant. All statistical analysis and data visualization was performed in Python version 3.9.2 (Python Software Foundation).
Results
This study comprised 878 patients with AD, with 26 (3.0%) reporting 0 flares in 2022, 405 (46.1%) reporting 1 to 5 flares, 169 (19.2%) reporting 6 to 10 flares, and 278 (31.7%) reporting more than 10 flares. A total of 235 patients (26.8%) had mild AD, 575 (65.5%) had moderate AD, and 283 (32.2%) had severe AD. Overall, 279 patients (31.8%) were male, the median (IQR) age in 2022 was 49 (39-59) years, and median (IQR) age at diagnosis was 2 (1-8) years. Almost half of patients (421 [48.8%]) had concomitant asthma, and more than half had allergic rhinitis (580 [66.7%]). Overall, 721 (83.2%) were actively treated for AD and 519 (59.1%) reported a family history of AD (Table 1).
Table 1. Baseline Characteristics Reported in 2022 Stratified on Number of Flares Reported in 2022.
| Characteristic | Overall (N = 878) | No. of yearly flares | |||
|---|---|---|---|---|---|
| 0 (n = 26) | 1-5 (n = 405) | 6-10 (n = 169) | >10 (n = 278) | ||
| Male, No. (%) | 279 (31.8) | 6 (23.1) | 151 (37.3) | 49 (29.0) | 73 (26.3) |
| Female, No. (%) | 599 (68.2) | 20 (76.9) | 254 (62.7) | 120 (71.0) | 205 (73.7) |
| Age, median (IQR), y | 49.0 (39.0-59.0) | 48.0 (41.8-60.2) | 50.0 (40.0-59.0) | 46.0 (38.0-60.0) | 48.5 (38.0-57.0) |
| Age at diagnosis, median (IQR), y | 2.0 (1.0-8.0) | 2.0 (1.0-7.0) | 2.0 (1.0-8.0) | 2.0 (1.0-6.0) | 2.0 (1.0-9.8) |
| Age at diagnosis, No. (%) | |||||
| 0-6 mo | 64 (15.8) | 1 (7.7) | 30 (16.1) | 16 (20.5) | 17 (13.3) |
| 6 mo to 5 y | 165 (40.7) | 8 (61.5) | 69 (37.1) | 31 (39.7) | 57 (44.5) |
| 5-18 y | 125 (30.9) | 4 (30.8) | 67 (36.0) | 21 (26.9) | 33 (25.8) |
| >18 y | 50 (12.3) | 0 | 19 (10.2) | 10 (12.8) | 21 (16.4) |
| Asthma, No. (%) | 421 (48.8) | 10 (38.5) | 184 (46.3) | 87 (51.8) | 140 (51.7) |
| Allergic rhinitis, No. (%) | 580 (66.7) | 15 (57.7) | 262 (64.9) | 113 (67.7) | 190 (69.9) |
| BMI, median (IQR) | 25.5 (22.5-29.0) | 25.7 (23.2-29.5) | 25.6 (22.6-28.7) | 25.0 (22.3-29.8) | 25.4 (22.3-29.2) |
| Lifestyle, No. (%) | |||||
| Sedentary | 161 (18.3) | 5 (19.2) | 74 (18.3) | 30 (17.8) | 52 (18.7) |
| Low | 511 (58.2) | 12 (46.2) | 232 (57.3) | 103 (60.9) | 164 (59.0) |
| Moderate to vigorous | 202 (23.0) | 9 (34.6) | 97 (24.0) | 35 (20.7) | 61 (21.9) |
| No. of flares, median (IQR) | 6.0 (3.0-12.0) | 3.0 (2.0-4.0) | 8.0 (6.0-10.0) | 20.0 (15.0-20.0) | |
| Mean flare duration in last 12 mo, No. (%) | |||||
| 1-7 d | 327 (38.9) | NA | 141 (35.2) | 67 (39.9) | 119 (43.8) |
| 1-2 wk | 202 (24.0) | NA | 82 (20.5) | 52 (31.0) | 68 (25.0) |
| 2-4 wk | 149 (17.7) | NA | 84 (21.0) | 28 (16.7) | 37 (13.6) |
| 1-2 mo | 79 (9.4) | NA | 52 (13.0) | 17 (10.1) | 10 (3.7) |
| >2 mo | 83 (9.9) | NA | 41 (10.2) | 4 (2.4) | 38 (14.0) |
| Severity of last flare, median (IQR) | 5.0 (4.0-7.0) | 5.0 (3.0-7.0) | 5.0 (4.0-7.0) | 6.0 (5.0-8.0) | |
| AD in close relations, No. (%) | 519 (59.1) | 16 (61.5) | 241 (59.5) | 101 (59.8) | 161 (57.9) |
| Currently receiving treatment, No. (%) | 721 (83.2) | 22 (84.6) | 318 (79.5) | 135 (80.4) | 246 (90.1) |
| Satisfaction with current treatment, No. (%) | |||||
| Extremely satisfied | 49 (6.8) | 6 (27.3) | 29 (9.1) | 9 (6.7) | 5 (2.0) |
| Very satisfied | 191 (26.5) | 10 (45.5) | 100 (31.4) | 29 (21.5) | 52 (21.1) |
| Somewhat satisfied | 181 (25.1) | 3 (13.6) | 73 (23.0) | 43 (31.9) | 62 (25.2) |
| Neither satisfied nor dissatisfied | 109 (15.1) | 0 | 34 (10.7) | 21 (15.6) | 54 (22.0) |
| Somewhat dissatisfied | 108 (15.0) | 0 | 39 (12.3) | 21 (15.6) | 48 (19.5) |
| Very/extremely dissatisfied | 83 (9.5) | 3 (11.5) | 43 (10.6) | 12 (7.1) | 25 (9.0) |
Abbreviations: AD, atopic dermatitis; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); NA, not applicable.
The percentage of patients receiving active treatment was higher among patients with more than 10 yearly flares (246 [90.1%]), followed by patients with 0 yearly flares (22 [84.6%]), 6 to 10 yearly flares (135 [80.4%]), and 1 to 5 yearly flares (318 [79.5%]).
Annual Flares, Disease Severity, and Quality of Life
The distributions of quality of life, measured by DLQI, and AD severity, measured by PO-SCORAD and POEM, in 2023 were visualized and stratified by annual flares during 2022. The results indicate that patients who experienced flares in 2022 reported higher PO-SCORAD and POEM 1 year later (Figure 1A). Similar patterns were found when considering only patients classified as having mild AD in 2022 (Figure 1B).
Figure 1. Density Distribution of DLQI, PO-SCORAD, and POEM Reported in 2023 Stratified by the Number of Flares Reported in 2022.
DLQI indicates Dermatology Life Quality Index; POEM, Patient-Oriented Eczema Measure; PO-SCORAD, Patient-Oriented Scoring of Atopic Dermatitis.
The number of flares was compared across both years, following the same flare strata (eFigure 1 in Supplement 1). Among patients with a known number of annual flares in both 2022 and 2023, 28.6% of patients (4 of 14) who reported 0 flares in 2022 also reported 0 flares in 2023. Furthermore, 69.8% of patients (185 of 265) with 1 to 5 flares, 36.0% (45 of 125) with 6 to 10 flares, and 58.0% (127 of 219) with more than 10 flares in 2022 remained in the same groups in 2023, while 71.4% of patients (10 of 14) with 0 flares, 27.5% (73 of 265) with 1 to 5 flares, and 28.8% with 6 to 10 flares (36 of 125) in 2022 switched to a group with more frequent flares in 2023 (eFigure 1 in Supplement 1). When patients were stratified by both severity and annual flares, a similar pattern was observed (eFigure 2 in Supplement 1). A continuous scale for the number of flares and the proportion of patients reporting the same or more flares is displayed in eFigure 3 in Supplement 1.
Regression Models: Associations Between Flares and Severity Measures
Both current, worst in the past 12 months, and worst ever body surface area reported by patients in 2023 increased with the number of flares reported in 2022 (eg, median [IQR] current body surface area of 2.0 [1.0-5.0] among patients with 0 flares, 5.0 [2.0-10.0] among patients with 1-5 flares, 5.0 [3.0-10.5] among patients with 6-10 flares, and 8.0 [3.0-20.0] among patients with >10 flares reported in 2022 in 2023). A significant correlation was found with a quantile regression coefficient (0.26 [95% CI, 0.17-0.36]) when adjusting for sex and age. However, the correlation was no longer significant when also adjusting for AD severity in 2022 measured by PO-SCORAD (Table 2).
Table 2. Outcome Measures Reported in 2023a.
| Overall | No. of yearly flares | Coefficient (95% CI) | Adjusted coefficient (95% CI)b | ||||
|---|---|---|---|---|---|---|---|
| 0 | 1-5 | 6-10 | >10 | ||||
| Body surface area, median (IQR) | |||||||
| Current | 5.0 (2.0 to 15.0) | 2.0 (1.0 to 5.0) | 5.0 (2.0 to 10.0) | 5.0 (3.0 to 10.5) | 8.0 (3.0 to 20.0) | 0.26 (0.17-0.36) | 0.07 (−0.04 to 0.18) |
| Worst in 12 mo | 10.0 (4.0 to 20.0) | 5.0 (2.0 to 8.8) | 8.0 (3.0 to 20.0) | 9.5 (4.0 to 20.0) | 10.0 (4.0 to 30.0) | 0.28 (0.14-0.42) | −0.07 (−0.23 to 0.08) |
| Worst ever | 25.0 (10.0 to 70.0) | 10.0 (5.0 to 65.0) | 25.0 (8.5 to 70.0) | 25.0 (10.0 to 57.5) | 32.5 (10.0 to 75.0) | 0.63 (0.07-1.20) | −0.14 (−0.63 to 0.35) |
| PO-SCORAD | 33.1 (23.1 to 45.3) | 19.4 (14.0 to 23.6) | 29.2 (21.1 to 41.1) | 33.8 (24.3 to 42.2) | 39.0 (27.9 to 51.4) | 0.66 (0.45-0.88) | NA |
| POEM | 10.0 (5.0 to 15.0) | 4.5 (2.0 to 9.8) | 8.0 (5.0 to 13.0) | 10.0 (5.0 to 15.0) | 13.0 (8.0 to 18.0) | 0.29 (0.20-0.38) | 0.11 (0.03 to 0.19) |
| DLQI, median (IQR) | 3.0 (1.0 to 7.0) | 1.0 (0.0 to 2.0) | 2.0 (1.0 to 5.0) | 3.0 (1.0 to 6.0) | 5.0 (2.0 to 8.8) | 0.15 (0.11-0.20) | 0.08 (0.04 to 0.12) |
| Multidimensional fatigue inventory, median (IQR) | |||||||
| General fatigue | 12.0 (8.0 to 16.0) | 9.0 (5.0 to 15.0) | 12.0 (7.0 to 15.0) | 12.0 (9.0 to 16.0) | 12.0 (8.0 to 17.0) | 0.09 (0.01-0.17) | 0.04 (−0.04 to 0.13) |
| Physical fatigue | 10.0 (7.0 to 13.0) | 9.0 (5.0 to 12.0) | 10.0 (6.0 to 13.0) | 10.0 (7.0 to 14.0) | 10.5 (7.0 to 14.0) | 0.06 (0.00-0.13) | 0.04 (−0.04 to 0.11) |
| Reduced activity | 8.0 (6.0 to 12.0) | 8.0 (6.0 to 11.0) | 8.0 (5.0 to 12.0) | 9.0 (6.0 to 12.0) | 9.0 (6.0 to 13.0) | 0.03 (−0.03-0.10) | −0.006 (−0.07 to 0.06) |
| Reduced motivation | 7.0 (5.0 to 10.0) | 9.0 (5.0 to 12.0) | 7.0 (5.0 to 10.0) | 7.0 (5.0 to 10.0) | 8.0 (5.0 to 11.0) | 0.07 (0.02-0.11) | 0.04 (−0.009 to 0.09) |
| Mental fatigue | 8.0 (5.0 to 12.0) | 6.0 (4.0 to 13.0) | 8.0 (4.0 to 12.0) | 9.0 (5.0 to 12.0) | 9.0 (5.0 to 13.0) | 0.06 (0.00-0.13) | 0.03 (−0.04 to 0.09) |
| Trouble sleeping, past 3 d, median (IQR) | 3.0 (0.5 to 5.0) | 1.0 (0.0 to 4.0) | 2.0 (0.0 to 5.0) | 3.0 (0.0 to 5.0) | 3.0 (1.0 to 6.0) | 0.06 (0.02-0.11) | −0.007 (−0.05 to 0.04) |
Abbreviations: DLQI, Dermatology Life Quality Index; NA, not applicable; POEM, Patient-Oriented Eczema Measure; PO-SCORAD, Patient-Oriented Scoring of Atopic Dermatitis.
Patients are stratified based on the reported number of flares in 2022. Coefficients and 95% CIs are based on quantile regressions performed with each variable reported in 2023 as the dependent variable and the number of flares reported in 2022 as an independent variable. The model is adjusted for sex and current age.
Adjusted for PO-SCORAD at baseline (2022).
The severity of AD measured by POEM in 2023 increased significantly from a median (IQR) of 4.5 (2.0-9.8) among patients with no flares registered in 2022 to 13.0 (8.0-18.0) among patients with more than 10 yearly flares reported in 2022 (quantile regression coefficient, 0.29 [95% CI, 0.20-0.38]). There was also a significant correlation when adjusting for the severity in 2022 measured by PO-SCORAD (regression coefficient, 0.11 [95% CI, 0.03-0.19]) (Table 2).
Additionally, the impairment in life quality measured by DLQI in 2023 was higher among patients with more flares in 2022 (median [IQR] DLQI, 1.0 [0.0-2.0] for patients with 0 flares, 2.0 [1.0-5.0] for patients with 1-5 flares, 3.0 [1.0-6.0] for patients with 6-10 flares, and 5.0 [2.0-8.8] for patients with >10 yearly flares in 2022). However, when focusing on clinical significance, the median DLQI scores remained within the group of small effect on patient’s life (score of 2-5).16 This correlation was significant both when adjusting for sex and age (quantile regression coefficient, 0.15 [95% CI, 0.11-0.20]) and when adjusting for PO-SCORAD reported in 2022 (quantile regression coefficient, 0.08 [95% CI, 0.04-0.12]) (Table 2).
Using a logistic regression model to estimate the odds of patients having moderate or severe AD compared with mild AD demonstrated a higher odds ratio (OR) for both moderate (OR, 1.06 [95% CI, 1.03-1.09]) and severe disease (OR, 1.11 [95% CI, 1.08-1.15]) in the unadjusted model. This association was not confirmed when adjusting for PO-SCORAD (eTable in Supplement 1). Furthermore, the odds of increasing one severity category from 2022 to 2023 was lower for patients with more flares in the unadjusted model (OR, 0.90 [95% CI, 0.82-0.99]), which may be explained by the already high number of flares and a potential ceiling effect. However, no differences were found when adjusting for the severity reported in 2022 (eTable in Supplement 1).
Validation of Predictive Models and Identification of Potential Predictors
The primary model using binary classification to distinguish between mild vs moderate to severe AD achieved an area under the receiver operating characteristics of 65.9%, which indicates a moderate ability to predict severity based on the included variables. The severity of the last flare and the number of annual flares were the variables with the highest predictive power, with higher severity and more flares contributing toward more severe disease. Additionally, larger mean flare duration and higher patient weight were predictors of more severe disease (Figure 2).
Figure 2. Binary Prediction Model to Classify Patients With Mild or Moderate to Severe Atopic Dermatitis (AD) in 2023 Using Variables Collected in 2022.
For sex, red indicates females and blue indicates males. For family history of AD, red indicates yes and blue indicates no.
For the secondary multiclass classification model with a class for each severity group (mild, moderate, and severe), an area under the receiver operating characteristics of 63.0% was achieved. Feature importance was visualized separately for the 3 classes based on Shapley additive explanations values (eFigure 4 in Supplement 1). The patterns of feature importance for the mild and severe classes were similar to those obtained from the binary models. However, in addition to flare severity, weight, current age, and family history of AD were the most predictive variables in the distinction of moderate disease.
In the regression model to predict the number of annual flares, the number of annual flares in the previous year, as well as severity measured by both PO-SCORAD and POEM, had the highest predictive power among the included variables, with more flares and higher severity in 2022 contributing toward more annual flares in 2023 (Figure 3).
Figure 3. Regression Model With Number of Flares Reported in 2023 as Target Variable and Predictors Collected in 2022.
For sex, red indicates females and blue indicates males. For family history of atopic dermatitis (AD), red indicates yes and blue indicates no. BSA indicates body surface area; POEM, Patient-Oriented Eczema Measure; PO-SCORAD, Patient-Oriented Scoring of Atopic Dermatitis.
Discussion
In this study a temporal association was demonstrated between the number of flares and disease severity in patients with AD as measured by POEM and between DLQI and the number of flares. This finding suggests that flares may affect the severity of AD and cause a greater reduction in quality of life, regardless of AD severity. Furthermore, PO-SCORAD, POEM, and current age were identified as the most predictive variables for the number of annual flares, while flare severity, annual number of flares, flare duration, and patient weight were identified as key predictors of disease severity 1 year later.
These results concur with 2 previous cross-sectional studies on the Danish Skin Cohort, which demonstrated an association between AD severity and the number of annual flares and flare severity reported at a single point.17,18 However, the current analysis is a longitudinal extension that considers the time aspect of the disease course. Based on past data, it thus contributes information on predictors of future flares and disease severity. Furthermore, the result aligns with a previous study investigating predictors of the patient-perceived burden of AD, which found flare type and frequency to be predictors of DLQI.19
Flares may indicate inadequate disease control, which has been associated with higher burdens of disease, work impairment, and lower disease-specific quality of life.12 A longitudinal study on predictors of depressive symptoms in patients with AD found an inverse correlation between disease control and mental health.20 In light of the presented results, this supports the notion that disease control is an important treatment goal for patients with AD and that frequent flaring, higher severity, and lack of control may result in a higher burden for patients.
Although the primary aim of flare management is to achieve short-term control without compromising long-term disease control, flaring could pose challenges for patients with mild disease who do not progress to more severe disease and fail to achieve relevant and meaningful disease control with the prescribed medicine.13 In many cases, flares may be managed with topical corticosteroids, but in cases where only short-term or no relevant control is achieved, alternative treatment options could be considered. However, determining a specific threshold for the number of flares, which could necessitate a change in treatment to achieve better long-term disease control, remains an area for further research.
Although the models presented in this article serve as an exploratory tool to validate the predictability of flares and disease severity and quantify the importance and predictive power of different variables, rigorous development and extensive testing are necessary prior to implementation in clinical settings. However, this is a proof of concept of the potential of a data-driven approach to inform and support individualized patient care. Advancements in the application of machine learning within dermatology may facilitate predictive modeling of disease courses and treatment outcomes. Attempts have been made to predict the daily severity of AD at the individual level based on variables such as environmental factors.21,22 In this study, predictive models identified key variables related to disease severity, highlighting the significant relevance of flare patterns, including both frequency and severity.
Strengths and Limitations
An important strength of this study is the longitudinal aspect of the data, allowing for the identification of predictors of future flares and disease severity. Additionally, the Danish Skin Cohort comprises a large number of diagnosed patients and contains information on many types of variables, including detailed information on flaring and disease severity. While previous investigations of flaring and inadequate disease control may have been limited by enrollment bias, because patients seek medical help when they experience flares, the Danish Skin Cohort follows the patient irrespective of disease status.
The study is also subject to limitations, including potential residual confounding, recall bias, recency bias, and information bias. Despite patients being informed of the definition of flares, the patient-reported nature of the data gives rise to potential misidentification of flares. Although the questions were asked in the context of AD, patients may not possess adequate education and training to accurately distinguish AD flares from other similarly presenting skin conditions.
Conclusions
In this study, the self-reported number of flares in the previous year, in addition to disease severity, duration and severity of the flares, was found to be predictive of future disease severity and frequency of future flares. Currently, eczema area and severity index assessment remains the criterion standard for disease severity and treatment evaluation in Europe, whereas the US and Canada primarily use the Investigator’s Global Assessment Scale. Although a consensus on how many flares are too many remains to be established, the current findings suggest that flares might serve as an early indicator of disease progression or inadequate disease control, highlighting that flares could be relevant in clinical decision-making to support optimal treatment strategies.
eTable 1. Odds ratios
eFigure 1. Sankey diagram – changes in flare strata
eFigure 2. Sankey diagram – changes in flare and severity groups
eFigure 3. Proportion of patients with at least same number of flares in 2023 as in 2022
eFigure 4. Multiclass classification model in investigate predictors of flares
Data sharing statement
References
- 1.Langan SM, Mulick AR, Rutter CE, et al. Trends in eczema prevalence in children and adolescents: a Global Asthma Network phase i study. Clin Exp Allergy. 2023;53:337-352. doi: 10.1111/cea.14276 [DOI] [Google Scholar]
- 2.Kay J, Gawkrodger DJ, Mortimer MJ, Jaron AG. The prevalence of childhood atopic eczema in a general population. J Am Acad Dermatol. 1994;30(1):35-39. doi: 10.1016/S0190-9622(94)70004-4 [DOI] [PubMed] [Google Scholar]
- 3.Hadi HA, Tarmizi AI, Khalid KA, Gajdács M, Aslam A, Jamshed S. The epidemiology and global burden of atopic dermatitis: a narrative review. Life (Basel). 2021;11(9):936. doi: 10.3390/life11090936 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Spergel JM, Paller AS. Atopic dermatitis and the atopic march. J Allergy Clin Immunol. 2003;112(6)(suppl):S118-S127. doi: 10.1016/j.jaci.2003.09.033 [DOI] [PubMed] [Google Scholar]
- 5.Severity scoring of atopic dermatitis: the SCORAD index. Consensus Report of the European Task Force on Atopic Dermatitis. Dermatology. 1993;186(1):23-31. doi: 10.1159/000247298 [DOI] [PubMed] [Google Scholar]
- 6.Hanifin JM, Thurston M, Omoto M, Cherill R, Tofte SJ, Graeber M; EASI Evaluator Group . The eczema area and severity index (EASI): assessment of reliability in atopic dermatitis. Exp Dermatol. 2001;10(1):11-18. doi: 10.1034/j.1600-0625.2001.100102.x [DOI] [PubMed] [Google Scholar]
- 7.Charman CR, Venn AJ, Williams HC. The patient-oriented eczema measure: development and initial validation of a new tool for measuring atopic eczema severity from the patients’ perspective. Arch Dermatol. 2004;140(12):1513-1519. doi: 10.1001/archderm.140.12.1513 [DOI] [PubMed] [Google Scholar]
- 8.Sidbury R, Tom WL, Bergman JN, et al. Guidelines of care for the management of atopic dermatitis: section 4: prevention of disease flares and use of adjunctive therapies and approaches. J Am Acad Dermatol. 2014;71(6):1218-1233. doi: 10.1016/j.jaad.2014.08.038 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Yew YW, Thyssen JP, Silverberg JI. A systematic review and meta-analysis of the regional and age-related differences in atopic dermatitis clinical characteristics. J Am Acad Dermatol. 2019;80(2):390-401. doi: 10.1016/j.jaad.2018.09.035 [DOI] [PubMed] [Google Scholar]
- 10.Czarnowicki T, He H, Krueger JG, Guttman-Yassky E. Atopic dermatitis endotypes and implications for targeted therapeutics. J Allergy Clin Immunol. 2019;143(1):1-11. doi: 10.1016/j.jaci.2018.10.032 [DOI] [PubMed] [Google Scholar]
- 11.Eichenfield LF, Gower RG, Xu J, et al. Once-daily crisaborole ointment, 2%, as a long-term maintenance treatment in patients aged ≥ 3 months with mild-to-moderate atopic dermatitis: a 52-week clinical study. Am J Clin Dermatol. 2023;24(4):623-635. doi: 10.1007/s40257-023-00780-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Wei W, Anderson P, Gadkari A, et al. Extent and consequences of inadequate disease control among adults with a history of moderate to severe atopic dermatitis. J Dermatol. 2018;45(2):150-157. doi: 10.1111/1346-8138.14116 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Wollenberg A, Christen-Zäch S, Taieb A, et al. ; European Task Force on Atopic Dermatitis/EADV Eczema Task Force . ETFAD/EADV Eczema task force 2020 position paper on diagnosis and treatment of atopic dermatitis in adults and children. J Eur Acad Dermatol Venereol. 2020;34(12):2717-2744. doi: 10.1111/jdv.16892 [DOI] [PubMed] [Google Scholar]
- 14.Lundberg S, Lee S-I. A unified approach to interpreting model predictions. arXiv. Preprint posted online November 25, 2017. doi: 10.48550/arXiv.1705.07874 [DOI]
- 15.Lundberg SM, Erion G, Chen H, et al. From local explanations to global understanding with explainable AI for trees. Nat Mach Intell. 2020;2(1):56-67. doi: 10.1038/s42256-019-0138-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Finlay AY, Khan GK. Dermatology Life Quality Index (DLQI): a simple practical measure for routine clinical use. Clin Exp Dermatol. 1994;19(3):210-216. doi: 10.1111/j.1365-2230.1994.tb01167.x [DOI] [PubMed] [Google Scholar]
- 17.Nymand L, Nielsen M-L, Vittrup I, et al. Atopic dermatitis phenotypes based on cluster analysis of the Danish Skin Cohort. Br J Dermatol. 2024;190(2):207-215. doi: 10.1093/bjd/ljad401. [DOI] [PubMed] [Google Scholar]
- 18.Nielsen M-L, Nymand LK, Domenech Pena A, et al. Characterization of patients with atopic dermatitis based on flare patterns and severity of disease: a Danish population-based study. J Eur Acad Dermatol Venereol. 2025;39(2):301-307. doi: 10.1111/jdv.20160 [DOI] [PubMed] [Google Scholar]
- 19.Paul C, Griffiths CEM, Costanzo A, et al. Factors predicting quality of life impairment in adult patients with atopic dermatitis: results from a patient survey and machine learning analysis. Dermatol Ther (Heidelb). 2023;13(4):981-995. doi: 10.1007/s13555-023-00897-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Chatrath S, Lei D, Yousaf M, Chavda R, Gabriel S, Silverberg JI. Longitudinal course and predictors of depressive symptoms in atopic dermatitis. J Am Acad Dermatol. 2022;87(3):582-591. doi: 10.1016/j.jaad.2022.04.061 [DOI] [PubMed] [Google Scholar]
- 21.Hurault G, Domínguez-Hüttinger E, Langan SM, Williams HC, Tanaka RJ. Personalized prediction of daily eczema severity scores using a mechanistic machine learning model. Clin Exp Allergy. 2020;50(11):1258-1266. doi: 10.1111/cea.13717 [DOI] [PubMed] [Google Scholar]
- 22.Hurault G, Delorieux V, Kim YM, Ahn K, Williams HC, Tanaka RJ. Impact of environmental factors in predicting daily severity scores of atopic dermatitis. Clin Transl Allergy. 2021;11(2):e12019. doi: 10.1002/clt2.12019 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eTable 1. Odds ratios
eFigure 1. Sankey diagram – changes in flare strata
eFigure 2. Sankey diagram – changes in flare and severity groups
eFigure 3. Proportion of patients with at least same number of flares in 2023 as in 2022
eFigure 4. Multiclass classification model in investigate predictors of flares
Data sharing statement
