Bayesian Networks with Interpretable Summary Indexes for Modeling Clinicians’ Decision-Making in Treatment Recommendation for Mental Disorders

Bayesian Networks provide a principled framework for developing tailored treatment recommendations by estimating conditional probabilities for diagnosis and treatment response from individual patient profiles. As new information becomes available, these probabilities update dynamically, enabling personalized clinical decision-making. However, applying Bayesian Networks to psychological data is challenging because parameter learning in Bayesian Networks can be sensitive to high-dimensional, correlated behavioral indicators across domains. We therefore introduce a Bayesian network–based modeling framework that incorporates Convex Generalized Structured Component Analysis (Convex GSCA) to derive interpretable, low-dimensional summary indexes from multivariate psychological and behavioral data. These summary indexes serve as nodes in a Bayesian network, yielding a parsimonious, theory-consistent model that reproduces clinicians’ treatment recommendations. We illustrate the approach with an empirical application to cognitive behavioral therapy for insomnia (CBT-I), demonstrating how interpretable summary indexes enhance the construction of a Bayesian network for probabilistic, patient-level treatment guidance. We provide commented code and example data to support step-by-step adoption in applied settings.