Digital health framework for the predictive surveillance and diagnosis of atopic dermatitis

Atopic dermatitis (AD) is an inflammatory skin disease with immunological and environmental triggers that reduces the quality of life and increases the burden on health services. It is thus important to establish effective surveillance and diagnosis methods for the development of preventive and therapeutic interventions. In line with this, the present study established a digital health framework combining urban big data analytics, machine learning modeling, and environmental bioinformatics for the predictive surveillance and diagnosis of the nationwide AD prevalence in Korea. In this process, urban big data from environmental (e.g., immune response inducers), crowdsourced (web search keywords related to AD symptoms), and municipal microbiome sources (AD-associated bacteria detectable in wastewater) were combined and employed as input variables. Data preprocessing (i.e., feature selection, scaling, and normalization), model testing and selection, and hyperparameter tuning were then used to improve the prediction accuracy for AD prevalence. By applying explainable artificial intelligence methods, highly explanatory predictors, such as specific skin disease keywords associated with AD patients and environmental and inflammatory factors, were identified. Environmental genomics revealed that Streptococcus strains were dominant in human-derived wastewater, with operational taxonomic units that were strongly associated with inflammation-inducing bacteria originating from AD patients. Bioinformatic analysis subsequently revealed the pathogenotype and resistotype of these inflammation-related bacteria. Overall, our digital health framework holds great promise as an alternative to conventional complex and costly surveillance systems for the proactive guidance of the decision-making of health professionals regarding the surveillance, diagnosis, and therapeutic treatment of environmental diseases.