Background
Atopic dermatitis (AD) presents with the wide spectrum of clinical phenotypes within and between various populations. Recent study showed low frequency of filaggrin loss‐of‐function (FLG LOF) mutations in Croatian AD patients. At present, there are no data on biomarkers of immune response in Croatian AD patients that might be useful in the selection and monitoring of novel immune therapies.

Objectives
To investigate levels of cytokines of various signature in the stratum corneum (SC) collected from lesional and non‐lesional skin of AD patients and healthy controls and to evaluate their relationship with the severity of disease and skin barrier function.

Methods
SC samples were collected from 100 adult patients with moderate‐to‐severe AD and 50 healthy controls. The levels of 21 cytokines were measured by multiplex immunoassay. We conducted machine learning analysis to assess whether a small number of cytokine measurements can discriminate between healthy controls and AD patients and can predict AD severity (SCORAD).

Results
The SC levels of thirteen cytokines representing innate immunity, Th‐1, Th‐2 and Th‐17/22 immune response showed significant differences between healthy and AD skin. Our analysis demonstrated that as few as three cytokines measured in lesional skin can discriminate healthy controls and AD with an accuracy of 99% and that the predictive models for SCORAD did not achieve a high accuracy. Cytokine levels were highly correlated with the levels of filaggrin degradation products and skin barrier function.

Conclusions
Stratum corneum analysis revealed aberrant levels of cytokines representing innate immunity, Th‐1‐, Th‐2‐ and Th‐17/22‐mediated immune response in Croatian AD patients. Increased Th‐2 cytokines and their strong association with natural moisturizing factor (NMF) can explain low NMF levels despite of low frequency of FLG LOF mutations in Croatian population. Predictive models for SCORAD identified cytokines associated with SCORAD but warrants further investigation.