Inferring Symptom State of Generalized Anxiety Disorder: A Bayesian Network Approach

Instead of viewing psychiatric disorders as latent causes that lead to observable symptoms, a network view of psychiatric disorders argues that each disorder can be regarded as a complex network of interacting symptoms. Such a network view of psychiatric disorders enables the analysis of the inter-dependencies between individual symptoms. Here, I modeled a set of binary symptoms in Generalized Anxiety Disorder (GAD) as a Bayesian network and performed Belief Propagation on this symptom network to infer the potential states of unobserved symptom variables. In the learned symptom network, the interactions between GAD symptoms were directly supported by empirical investigation of the co-occurrences or causal relations between them. The symptom network enabled one to infer the state of unobserved symptom variables given partial observation. Furthermore, predicting symptom states on the Bayesian network out-performed state-of-the-art machine learning methods that did not explicitly model the interdependencies between symptom variables. Together, this study proposes a novel and reliable approach for measuring the risk of certain GAD symptoms for a patient by inferring the likelihood of developing the symptoms of interest on the Bayesian symptom network. The learned symptom network also predicts novel interdependencies between symptoms that can be verified in future empirical research. The Bayesian network model of GAD provides a potential mechanistic account underlying the co-occurrence of symptoms in GAD.