Identifying high-dose opioid prescription risks using machine learning: A focus on sociodemographic characteristics

Area of Science:

• Public Health
• Data Science
• Epidemiology

Background:

• High-dose opioid prescribing poses significant public health challenges.
• Understanding the complex risk factors is crucial for effective mitigation strategies.

Purpose of the Study:

• To identify and interpret risk factors for high-dose opioid prescribing.
• To leverage machine learning (ML) for analyzing administrative claims and socioeconomic data.

Main Methods:

• Applied six ML algorithms to integrated Medicaid claims (Missouri, 2017-2021) and US Census Bureau data (2018).
• Defined high-dose prescribing as ≥120 morphine milligram equivalent/day.
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Sociodemographic factors (age, race, sex) are associated with high-dose prescribing risks.
• Socioeconomic variables, including percentages of veterans and primary care physicians (PCPs) per capita, correlate with increased high-dose prescriptions.
• Older age groups and patient sex were identified as predictors of higher risk.

Conclusions:

• Sociodemographic variables significantly influence high-dose opioid prescribing patterns.
• Multifaceted public health strategies are needed to address the opioid crisis, considering identified risk factors.
• Integrating ML with traditional epidemiology offers a comprehensive approach to understanding and mitigating the opioid crisis.