Inferring Skin-Brain-Skin Connections from Infodemiology Data using Dynamic Bayesian Networks

Abstract

ABSTRACTBackgroundThe relationship between skin diseases and mental illnesses has been extensively studied using cross-sectional epidemiological data. Typically, such data can only measure association (rather than causation) and include only a subset of the diseases we may be interested in.ObjectiveIn this paper, we complement the evidence from such analyses by learning an overarching causal network model over 12 health conditions from a Google search trends public data set.MethodsWe learned the network model using a dynamic Bayesian network, which can represent both cyclic and acyclic causal relationships, is easy to interpret and accounts for the spatio-temporal trends in the data in a probabilistically rigorous way.ResultsThe averageR2for a condition given the values of all conditions in the previous week is 0.67: in particular, 0.42 for acne, 0.85 for asthma, 0.58 for ADHD, 0.87 for burn, 0.76 for erectile dysfunction, 0.88 for scars, 0.57 for alcohol disorders, 0.57 for anxiety, 0.53 for depression, 0.74 for dermatitis, 0.60 for sleep disorders and 0.66 for obesity. Results confirm the large number of cyclic relationships between the selected health conditions and the interplay between skin and mental diseases. For acne, we observed a cyclic relationship with anxiety and attention deficit hyperactivity disorder (ADHD) and an indirect relationship with depression through sleep disorders. For dermatitis, we observed directed links to anxiety, depression and sleep disorders and a cyclic relationship with ADHD. We also observe a link between dermatitis and ADHD and a cyclic relationship between acne and ADHD. Furthermore, the network includes several direct connections between sleep disorders and other health conditions, highlighting the impact of the former on the overall health and well-being of the patient.ConclusionsMapping disease interplay, indirect relationships and the key role of mediators, such as sleep disorders, will allow healthcare professionals to address disease management holistically and more effectively. Even if we consider all skin and mental diseases jointly, each disease subnetwork is unique, allowing for more targeted interventions.