Atopic dermatitis (AD) is a prevalent inflammatory skin disease and a major source of disease burden in children. Biomarker studies in childhood AD span genetic, immune, microbial and metabolic domains, but prior reviews have often focused on single molecular layers, specific sample sites or clinical classification. As a result, the field lacks an integrated, systems-level synthesis that compares and contextualizes biomarkers across domains while clearly distinguishing evidence strength. The rapid growth of literature in this field also poses practical challenges for traditional manual review workflows. To address these gaps, we conducted an AI-augmented, multi-domain review of childhood AD biomarkers. ASReview supported title and abstract screening, while ChatGPT assisted structured data extraction with human validation. Across 526 studies, we identified 141 genome, 95 immunome, 57 microbiome and 75 metabolome childhood AD biomarkers. The most frequently reported biomarkers included Filaggrin, IgE, CCL17, Staphylococcus, Bifidobacterium and vitamin D. Using a structured evidence-grading framework, eight biomarkers were categorized as having strong evidence: IgE, CCL17, CCL27, eosinophil cationic protein, eosinophil, IL-18, IL-31 and Escherichia. By synthesizing evidence across biomarker domains, we developed a systems-level, conceptual AD model in which barrier defects, Th2 inflammation, microbial dysbiosis and metabolic imbalance drive a self-perpetuating cycle of inflammation and barrier dysfunction. We also developed a web app for exploration of the biomarker findings: https: //leejw. shinyapps. io/eczemaᵣeview₅26/. This review provides a broad synthesis of childhood AD biomarkers and frames the evidence within an integrated, multi-domain conceptual model. The findings support the rationale for approaches that consider multiple biological nodes, including barrier repair, immune modulation, microbiome-directed strategies and metabolic factors, while underscoring the need for further validation before clinical implementation. Methodologically, the study illustrates how a hybrid human-AI review workflow can support scalable biomedical evidence synthesis without replacing human oversight.
Lee et al. (Sat,) studied this question.