Atopic Dermatitis in the Era of Exposome and Artificial Intelligence

Atopic dermatitis (AD) is a heterogeneous, relapsing inflammatory skin disease in which epidermal barrier dysfunction, immune dysregulation, microbial imbalance, genetics, and the environment interact continuously over time. Although a substantial body of literature has linked AD with air pollution, weather, allergens, stress, infections, sleep disturbance, skin care practices and microbial perturbation, most studies have assessed single exposures in isolation. This reductionist approach only partially reflects the real-world biology of AD, where patients experience dynamic combinations of external and internal exposures whose effects are non-linear, cumulative and modified by age, phenotype, treatment and susceptibility. The exposome paradigm offers a broader conceptual lens by considering the totality of exposures over the life course. However, exposomic research in AD is methodologically challenging because the data are multimodal, high-dimensional, time-dependent, and often incomplete. Artificial intelligence (AI) can help address this complexity by integrating heterogeneous data layers, modelling temporal dependencies, discovering latent patterns, and generating individualized predictions. This narrative review examines current evidence and proposes a theoretical framework that integrates AI and the exposome in AD. We define the principal domains of the AD exposome, describe a multilevel analytical architecture spanning environmental, clinical, behavioural, microbiome, and multi-omics data. We distinguish applications evaluated in AD studies from exploratory analyses and proposed objectives: prediction of annual flare burden remains exploratory, whereas exposome-informed short-term flare alerts are proposed applications. Although selected AI applications have been evaluated in AD, the added predictive value of exposure data and the clinical utility of a fully integrated exposome–AI system remain to be established. We also outline key methodological and translational priorities, including exposure assessment, temporality, treatment-confounding, interpretability, fairness, data governance, and external validation. The proposed framework offers a structured basis for future research, with clinical translation dependent on prospective validation and evidence of benefit to patient care.