Classification of Itch Patterns and Sentiment Analysis in Patients With Atopic Dermatitis: Clustering Analysis of Patient‐Reported Data

Atopic dermatitis (AD), a prevalent inflammatory skin disease, significantly impairs quality of life. Pruritus manifests variably, complicating its classification and management. This study analyzed real-world, self-reported data from Atopiyo, Japan's largest AD smartphone application (October 2021-October 2024), to explore itch patterns, treatment behaviors, and patient sentiments. Data from 98 users with ≥ 6-months of itch scores and ≥ 11-months of posting history (73.5% female; 82.7% aged 18-50 years) were assessed. The mean itch score (0-5 scale) was 1.80, with a median of 47 posts/user and itch score posts peaked during spring and autumn transitions. The face (56.7 posts per patient) was the most frequently affected area. Cluster analysis grouped users into four categories based on patient-reported itch metrics (minimum, maximum, median, and itch score variability): mild (Cluster 1, n = 38), moderate (Cluster 2, n = 27), high-variability (Cluster 3, n = 16), and severe (Cluster 4, n = 17). Cluster 4 reported numerically lower prescribed medication use (35.3% vs. 57.9%-68.8% in other clusters). Symptom-related keywords most commonly included "itch," followed by "inflammation" and "pain". Treatment-related keywords frequently included "topical corticosteroids" and "prescription moisturizers". Topic modeling identified six major themes, with symptoms and treatment concerns emerging prominently. The themes were: daily symptom management, symptoms and triggers, symptom impact on daily life, advanced treatment options, topical therapies, and symptom control. Topic-"symptom control" was most prominent in the mild itch cluster, while Topic-"symptom impact on daily life" was especially notable in the high-variability itch cluster. In the moderate and severe itch clusters "symptoms and triggers" and "symptom impact on daily life" emerged as the leading topics. Sentiment analysis revealed a predominance of negative-tone posts (46.3%) compared with positive (21.7%) or neutral (32.0%) posts across clusters. Overall, this study used real-world smartphone data to reveal changing patients' needs and itch patterns in AD. Beyond symptom monitoring, these insights reveal potential gaps in treatment adherence and patients' emotional well-being improvement, underscoring the need for personalized, patient-centered care strategies.