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

Abstract

Objective: The objective of this study was to leverage machine learning techniques to analyze administrative claims and socioeconomic data, with the aim of identifying and interpreting the risk factors associated with high-dose opioid prescribing.

Design: We applied six machine learning algorithms to a dataset integrating Medicaid claims from Missouri (2017-2021) and 2018 United States Census Bureau data. High-dose prescribing was defined as dosages ≥120 morphine milligram equivalent/day. SHapely Additive exPlanations methods were utilized to enhance model interpretability, ensuring transparent insights into the predictors of high-dose prescription risks.

Results: Our findings reveal that sociodemographic factors like age, race, and sex, along with socioeconomic variables such as percentages of veterans, disability, and primary care physicians (PCPs) per capita, have associations with high-dose prescription risks. Notably, higher percentage of veterans and PCPs per capita within counties correspond with increased high-dose prescriptions, while older age groups and patient sex also predict a greater risk.

Conclusion: This analysis underscores the significant influence of sociodemographic variables on high-dose opioid prescriptions. The interplay of these factors highlights the need for multifaceted public health strategies to address the underlying complexities of the opioid crisis. The integration of machine learning methods with traditional epidemiological techniques represents a promising approach for gaining a comprehensive understanding of intricate patterns not captured in traditional statistical analysis, thereby enabling effective mitigation of the opioid crisis.