Short-term prediction of opioid prescribing patterns for orthopaedic surgical procedures: a machine learning framework

Abstract

ABSTRACT Overprescribing of opioids after surgical procedures can increase the risk of addiction in patients, and under prescribing can lead to poor quality of care. In this study, we propose a machine learning-based predictive framework to identify the varying effects of factors that are related to the opioid prescription amount after orthopaedic surgery. To predict the prescription classes, we train multiple classifiers combined with random and SMOTE over-sampling and weight-balancing techniques to cope with the imbalance state of the dataset. Our results show that the gradient boosting machines (XGB) with SMOTE achieve the highest classification accuracy. Our proposed analytical framework can be employed to assist and therefore, enable the surgeons to determine the timely changing effects of these variables.