Inferring skin-brain-skin connections from infodemiology data using dynamic Bayesian networks

Area of Science:

• Computational epidemiology and network science.
• Investigating complex health condition interdependencies.

Background:

• Existing research on skin diseases and mental illnesses relies on correlational data, limiting causal inference.
• Cross-sectional studies often capture only associations and a limited scope of conditions.

Purpose of the Study:

• To construct a causal network model for twelve health conditions, including skin and mental diseases.
• To explore spatio-temporal trends and causal relationships using public Google Search Trends data.
• To identify mediating factors, such as sleep disorders, in disease interplay.

Main Methods:

• Utilized a dynamic Bayesian network to model causal relationships, accommodating cyclic and acyclic connections.
• Employed Google Search Trends Symptoms dataset for learning the network structure.
• Analyzed spatio-temporal patterns probabilistically.

Main Results:

• Confirmed significant cyclic relationships between various health conditions, particularly between skin and mental diseases.
• Identified specific links: acne with anxiety and attention deficit hyperactivity disorder (ADHD); dermatitis with anxiety, depression, and ADHD.
• Highlighted the crucial role of sleep disorders as a mediator, with direct connections to multiple conditions.

Conclusions:

• The learned causal network provides a comprehensive view of disease interconnections, moving beyond simple associations.
• Understanding these complex relationships, including indirect pathways and mediators like sleep disorders, enables holistic disease management.
• Unique disease subnetworks suggest potential for targeted interventions in clinical practice.