A machine learning framework to predict the risk of opioid use disorder

Abstract

Opioid overdose epidemic is a national public health crisis in the US. Little is known about how large-scale data analytics can be leveraged to help physicians predict whether a prescription opioid user will develop opioid use disorder. To that end, we proposed a machine learning framework for identifying potential risk factors of opioid use disorder from a large-scale healthcare claims data. These risk factors identified by the proposed framework can be used to predict which patient will be at higher risk of opioid use disorder following an opioid prescription. We utilized clinical diagnosis and prescription histories from Massachusetts commercially insured individuals who were prescribed opioids. We performed several feature selection techniques on a class imbalanced analytic sample to identify patient-level demographic and clinical features that were influential predictors of opioid use disorder. We, then compared the predictive power of four well-known machine learning algorithms: Logistic Regression, Random Forest, Decision Tree, and Gradient Boosting to predict the patients’ risk of opioid use disorder. The study results showed that the Random Forest model achieved superior predictive performance in terms of AUC and recall. Alongside the higher predictive accuracy, the random forest model identified clinical features, some of which were fairly consistent with prior clinical findings. In addition, our proposed framework is capable of extracting some other clinical features, which are predictive of opioid use disorder and indicative as the proxies of patients’ health status. We anticipate that the findings of our study will potentially help reduce in-appropriate and over prescription of opioids.

Keywords

• Opioid use disorder
• Opioid addiction epidemic
• Risk prediction
• Big data analytics
• Machine learning
• Predictive analytics