A Big Data Analytics Framework to Predict the Risk of Opioid Use Disorder

Overdose related to prescription opioids have reached an epidemic level in the US, creating an unprecedented national crisis. This has been exacerbated partly due to the lack of tools for physicians to help predict the risk of whether a patient will develop opioid use disorder. Little is known about how machine learning can be applied to a big-data platform to ensure an informed, sustained and judicious prescribing of opioids, in particular for commercially insured population. This study explores Massachusetts All Payer Claims Data, a de-identified healthcare dataset, and proposes a machine learning framework to examine how naïve users develop opioid use disorder. We perform several feature selections techniques to identify influential demographic and clinical features associated with opioid use disorder from a class imbalanced analytic sample. We then compare the predictive power of four well-known machine learning algorithms: Logistic Regression, Random Forest, Decision Tree, and Gradient Boosting to predict the risk of opioid use disorder. The study results show that the Random Forest model outperforms the other three algorithms while determining the features, some of which are consistent with prior clinical findings. Moreover, alongside the higher predictive accuracy, the proposed framework is capable of extracting some risk factors that will add significant knowledge to what is already known in the extant literature. We anticipate that this study will help healthcare practitioners improve the current prescribing practice of opioids and contribute to curb the increasing rate of opioid addiction and overdose.