Semi-Supervised Learning Algorithm for Identifying High-Priority Drug-Drug Interactions Through Adverse Event...

Area of Science:

• Pharmacovigilance
• Biomedical Informatics
• Machine Learning

Background:

• Drug-drug interactions (DDIs) pose a significant risk to patient safety, necessitating effective alert systems.
• Current DDI alerting systems face challenges with alert overload and user fatigue, hindering their clinical utility.
• Prioritizing high-risk DDIs is crucial for optimizing medication safety alerts.

Purpose of the Study:

• To develop and validate a machine learning framework for identifying high-priority DDIs.
• To leverage FDA adverse event reports for feature extraction and DDI classification.
• To reduce alert fatigue by focusing on clinically significant drug-drug interactions.

Main Methods:

• Utilized a machine learning framework incorporating feature extraction from FDA adverse event reports.
• Employed an autoencoder-based semi-supervised learning algorithm for DDI identification.
• Integrated stacked autoencoders and weighted support vector machines for enhanced classification performance.

Main Results:

• Demonstrated the effectiveness of adverse event feature representations in distinguishing high-priority DDIs.
• The proposed algorithm achieved superior performance (F-measure and AUC score) compared to competing methods.
• Successfully identified potential high-priority DDI candidates for medication alerts.

Conclusions:

• The developed framework offers a practical approach for pre-screening high-priority DDI candidates.
• Integrating multiple data sources and domain knowledge enhances DDI alert system optimization.
• This method contributes to improving patient safety by refining DDI alert prioritization.