Predicting Antidepressant Deprescription with Machine Learning Using Administrative Data

Area of Science:

• Health Informatics
• Machine Learning in Medicine
• Pharmacovigilance

Background:

• High rates of failed antidepressant discontinuation attempts pose a clinical challenge.
• Identifying suitable patients for antidepressant deprescription is difficult for healthcare providers.

Purpose of the Study:

• To develop and evaluate supervised machine learning models for predicting successful antidepressant deprescription.
• To leverage longitudinal dispensing data from the Pharmaceutical Benefits Scheme (PBS) dataset.

Main Methods:

• Developed two annotation pathways (retrospective and prospective) to label deprescription success/failure from administrative data.
• Trained supervised machine learning models, including XGBoost and Random Forest, on labelled PBS data.
• Utilized Australian primary care dispensing data for model development and validation.

Main Results:

• The Random Forest model achieved 0.90 accuracy for prospective annotation.
• The XGBoost model achieved 0.81 accuracy for retrospective annotation.
• Demonstrated the feasibility of using administrative healthcare data for predictive modeling.

Conclusions:

• Administrative healthcare data holds potential for creating clinical decision support tools for medication management.
• Further validation of models against actual clinical outcomes is necessary.
• Machine learning approaches can aid in optimizing antidepressant deprescription strategies.