Predictive modeling of susceptibility to substance abuse, mortality and drug-drug interactions in opioid patients

Abstract

Opioid addiction causes high degree of morbidity and mortality. Preemptive identification of patients at risk of opioid dependence and developing intelligent clinical decisions to deprescribe opioids to the vulnerable patient population may help in reducing the burden. Identifying patients susceptible to mortality due to opioid-induced side effects and understanding the landscape of drug-drug interaction pairs aggravating opioid usage are significant, yet, unexplored research questions. In this study, we present a collection of predictive models to identify patients at risk of opioid abuse, mortality and drug-drug interactions in the context of opioid usage. Using publicly available dataset from MIMIC-III, we developed predictive models (opioid abuse models a=Logistic Regression; b=Extreme Gradient Boosting and mortality model= Extreme Gradient Boosting) and identified potential drug-drug interaction patterns. To enable the translational value of our work, the predictive model and all associated software code is provided. This repository could be used to build clinical decision aids and thus improve the optimization of prescription rates for vulnerable population.