Machine learning for predicting opioid use disorder from healthcare data: A systematic review

Area of Science:

• Health Informatics
• Machine Learning
• Public Health

Background:

• The US faces a severe opioid epidemic, driving the need for advanced analytical tools.
• Machine learning (ML) offers potential for predicting opioid use disorder (OUD) by analyzing healthcare data.
• This review synthesizes research on ML applications for OUD prediction.

Approach:

• Systematic review of peer-reviewed papers (post-2012) using healthcare data for OUD prediction.
• Searched major academic databases (Google Scholar, PubMed, etc.) in September 2022.
• Extracted data on study objectives, datasets, ML models, evaluation metrics, and techniques.

Key Points:

• 16 papers were analyzed, utilizing diverse datasets (public, private, custom) and ML models.
• Most models achieved high predictive performance (ROC AUC > 0.8), with interpretable models showing strong results.
• Significant lack of detail in reporting ML techniques and tools, with limited source code availability.

Conclusions:

• ML holds potential for OUD prediction, but current research lacks transparency and detailed methodology.
• Recommendations are provided to enhance the rigor and reproducibility of ML studies in OUD.
• Increased transparency and standardized reporting are essential for translating ML findings into clinical practice.