Predicting treatment dropout after antidepressant initiation.

Area of Science:

• Pharmacology and Psychiatry
• Health Informatics
• Clinical Decision Support

Background:

• Antidepressants show similar efficacy but varied tolerability in trials.
• Personalizing antidepressant treatment can enhance patient adherence and outcomes.
• Predicting real-world treatment tolerability is crucial for effective depression management.

Purpose of the Study:

• To assess if electronic health record (EHR) data improves prediction of unplanned antidepressant discontinuation.
• To identify patient subgroups at higher risk of treatment dropout.
• To enhance personalized antidepressant selection based on predicted tolerability.

Main Methods:

• Retrospective longitudinal cohort study of 51,683 patients with depression initiating antidepressant treatment.
• Analysis of clinical data including diagnostic codes and medication prescriptions from EHRs.
• Application of machine learning models to predict treatment discontinuation.

Main Results:

• Models incorporating EHR data (diagnostic, procedure codes, prescriptions) improved prediction accuracy (mean AUC 0.69).
• Discontinuation rates varied by antidepressant, with paroxetine (27.71%) higher than venlafaxine (20.78%).
• Machine learning models effectively identified patients at high risk for treatment dropout.

Conclusions:

• Machine learning on coded EHR data can predict antidepressant treatment discontinuation.
• This approach aids clinicians in personalizing antidepressant therapy by considering tolerability.
• Improved prediction facilitates better patient adherence and potentially better depression management outcomes.