Using machine learning to study the effect of medication adherence in Opioid Use Disorder

Area of Science:

• Data Science
• Public Health
• Pharmacology

Background:

• Opioid Use Disorder (OUD) and opioid overdose (OD) present significant societal and healthcare challenges.
• Medication for Opioid Use Disorder (MOUD) is an effective treatment for OUD.
• Machine learning can investigate risk factors associated with OUD and OD.

Purpose of the Study:

• To investigate the association between MOUD adherence and other risk factors in patients with OUD and potential OD.
• To evaluate the predictive performance of machine learning models for identifying patients at high risk of opioid overdose.
• To assess the impact of medication adherence on overdose risk.

Main Methods:

• Utilized longitudinal Medicaid claims data for 26,685 patients diagnosed with OUD.
• Included variables such as age, sex, socioeconomic data, comorbidities, MOUD type, and Proportion of Days Covered (PDC) for MOUD adherence.
• Applied and compared four machine learning classifiers, focusing on PDC's predictive importance and risk stratification.

Main Results:

• XGBoost and Logistic Regression (LR) models showed high predictive performance (AUC 0.77 for XGBoost).
• LR model achieved best risk stratification: top 10% risk score patients accounted for 35.37% of OD events.
• Higher PDC (adherence) was associated with lower OD risk; 72.3% of high-risk patients were non-adherent (PDC <0.8).

Conclusions:

• Machine learning models can effectively identify patients at high risk of opioid overdose.
• MOUD adherence (PDC) is a significant factor in reducing OD risk.
• Targeted outreach for MOUD can be directed towards high-risk, non-adherent individuals.