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

Area of Science:

• Healthcare research
• Computational biology
• Medical informatics

Background:

• Medication adherence is a significant challenge in managing chronic diseases.
• Understanding the complex interplay of factors influencing adherence is vital for effective patient care.

Purpose of the Study:

• To apply a Bayesian network model for analyzing causal dependencies in medication non-adherence.
• To explore interrelationships among predictors of non-adherence in chronic patients.

Main Methods:

• Utilized adherence data from the BEAMER project involving chronic patients.
• Employed Hill Climb Search algorithm for structure learning.
• Applied Maximum Likelihood Estimation for parameter learning in the Bayesian network.

Main Results:

• Identified significant relationships, such as the link between polypharmacy, number of medications, and oncology diagnoses.
• Performed causal and diagnostic inference using Conditional Probability Distributions.
• Results align with existing domain knowledge on medication adherence.

Conclusions:

• Bayesian networks offer a powerful probabilistic graphical approach to understand medication non-adherence.
• The findings support the development of tailored interventions and personalized care strategies for chronic patients.
• Future research could explore advanced learning techniques and approximate inference methods.