A Bayesian network approach for understanding causal dependencies in chronic patients' medication non-adherence

Adherence to medication represents a current challenge in healthcare. Considering the importance not only of its predictors but also the interrelationships across them, this study explores the application of a Bayesian network model to analyze the complexity and causal dependencies on non-adherence. By using adherence data from chronic patients from the BEAMER project, the Hill Climb Search algorithm and Maximum Likelihood Estimation for structure and parameter learning were performed. The results provided relevant relationships, such as the relationship between the number of medications and the presence of oncology diseases in patients with polypharmacy. Moreover, both causal and diagnostic inference were performed via the computed Conditional Probability Distributions. The insights obtained are well aligned with established domain knowledge and this research validates the potential of these probabilistic graphical approaches in understanding causes of non-adherence. Future research lines could explore alternative structure learning techniques to enhance these findings, as well as an assessment using approximated inference.Clinical Relevance- This research provides valuable insights into the complex interrelationships between the factors influencing adherence to medication for enhancing tailored interventions and personalized care in chronic patients.