Bayesian network analysis of antidepressant treatment trajectories

Area of Science:

• Psychiatry
• Computational Neuroscience
• Health Informatics

Background:

• Selecting optimal antidepressant medication for individual patients remains a significant clinical challenge.
• Existing data on patient characteristics, treatment choices, and outcomes are often underutilized for pattern discovery.

Purpose of the Study:

• To identify patterns in patient characteristics, antidepressant treatment choices, and clinical outcomes.
• To evaluate the feasibility of combining Bayesian network analysis with natural language processing (NLP) for psychiatric data analysis.

Main Methods:

• Retrospective analysis of adult patients treated with antidepressants between 2014 and 2020 at two Dutch mental healthcare facilities.
• Application of Bayesian network analysis and NLP to extract treatment outcomes (core complaints, social functioning, well-being, patient experience) from clinical notes.
• Construction and comparison of Bayesian networks integrating patient demographics, treatment details, and extracted outcomes.

Main Results:

• Antidepressant continuation rates were 66% and 89% across different trajectories.
• Score-based network analysis identified 28 significant dependencies between treatment choices, patient characteristics, and outcomes.
• Treatment outcomes and prescription duration were closely linked, influenced by co-medications like antipsychotics and benzodiazepines.
• Tricyclic antidepressant use and diagnosis of depressive disorder predicted antidepressant continuation.

Conclusions:

• Combining network analysis with NLP offers a feasible method for discovering patterns in complex psychiatric data.
• Identified patterns in patient characteristics, treatment choices, and outcomes warrant further prospective investigation.
• The findings suggest potential for developing clinical decision support tools to aid antidepressant selection.